I commented and asked her if she had a source for the outcome numbers. She answered me, “Yes, for the data come to UNBOUND. We will share it.” Given previous HubSpot conferences, this leads me to believe (pure speculation, I have no insider knowledge here) that HubSpot will announce a Context Layer baked into the platform that Breeze agents will run on, and those numbers come from early alpha client success stories.

So the question is: If the highest-quality context produces two to four times the go-to-market outcomes, what does the highest-quality context actually look like — and how do you build it?

That is what this article is about. By the end of the article, you will have a precise and usable definition of the term Context Layer, know what high-quality looks like, and, maybe most importantly, how you can create yours within your organization. Because (spoiler alert): you can’t buy it.

Subscribe now

1. What A Context Layer Actually Is (Definition)

Yamini’s definition of high-quality context is through a CRM/tooling lens. She uses three criteria to define it: depth (completeness of CRM data), specificity (task-specific serving of context), and freshness (data reflecting the latest situation).

But I would argue it still misses critical elements that prevent a company from scaling with AI even if they had all three. And that’s the point most people miss. Entirely.

Point in case: Ask most business leaders how to scale AI in a business, and you will hear the same three requirements everywhere: a unified data foundation, so AI has high-quality information to draw from; operational alignment, so humans and machines work together without friction; and governance, so teams can use AI with confidence. All three are genuinely necessary. But even if you have all three fully in place, your AI could still produce generic slop that needs rework and doesn’t result in compounding business impact.

What the standard answer leaves out is a fourth layer: context. Yes, that includes the knowledge about your customer, your business, and your team. But it also must include the judgment that sits on top of that knowledge. The discernment. The taste. The thousands of micro-decisions your company (or most likely your founder and a few people) makes every day that make it truly yours.

So here is my definition of a context layer:

Let me take this apart.

Uniquely recognizable. This is the whole point of a context layer. You are trying to give your AI an accurate representation of how you would write something, answer a difficult question, and behave in a tricky situation. Context gives the AI everything it needs to know about you to create, act, and behave like you, and only you.

Deliberate. This is important. Tacit and implicit knowledge is made explicit, and definitions, standards, guardrails, process maps, and judgment rules have been deliberately decided. A context layer is a dump of documents that have accumulated in your Drive over eight years; it is what you decided, on purpose, to hold as true.

What your company knows. Who you serve and what actually drives them. What you sell, what it costs, and what it is worth. What you claim and what evidence you got to back those claims up. How you sound in a proposal, in a cold email, and in a support reply.

What your company stands for. Your position, your beliefs, and your stance. What you believe about your market, what you are not willing to do (even if it’s standard practice), what you are for, why you exist (beyond the transaction), and the value proposition that follows from all of it.

And what your company holds by. Your commitments (expressed most often in the form of refusals): Which customers you turn down. Which claims never ship without proof attached. What never leaves the building without a human reading it first. Where you stay silent while competitors shout. These almost never appear in any document, because nobody writes down what they refuse. They are also the most valuable content in the entire layer.

Extracted. Most of this is not written anywhere. Most mid-market companies don’t even have agreed-upon value propositions and messaging. Context lives in the heads and habits of your best people, and it surfaces in how they work and how they decide. This first needs to be carefully extracted.

Codified. Codifying serves two purposes. The first is resolving contradictions: after you extracted all the raw material, you are left with contradictions and half-truths. Someone with the authority has to decide which of the three currently used ICP definitions is the right one (if any). The second job is explicitly capturing the context—the value proposition, the messaging hierarchy, the refusals that nobody ever wrote down. Codifying is part judgment, part authorship.

Structured. Structuring now turns the extracted and codified knowledge into canonical, retrievable, machine-readable assets, organized for using rather than only for finding. So, don’t think of a context layer as a filing cabinet AI needs to make sense of first or a database that only contains data records. It includes all types of knowledge made explicit and accessible.

Now that we know what a Context Layer is, let’s talk about how you can test whether your context is high-quality or not.

2. The Four Tests Of A Real Context Layer

Most companies I talk to have some context: a brand guide, a few positioning documents, an ICP definition somewhere, a shared drive full of proposals. So the question isn’t whether you have context. It’s whether what you have is high-quality enough for AI to run on. These four tests will tell you.

Accessible. Can the knowledge be retrieved and used by an AI without first untangling, reinterpreting, or reformatting it? Technically, your filing cabinet contains the answers too. But it makes everyone assemble those answers from parts, every single time. And everyone assembles them slightly differently, which is exactly the problem you were trying to solve. Also, documents written for humans are different than context created for AI. (One great example is a voice guide. The voice guide for humans includes aspirational adjectives an AI cannot accurately interpret. It needs rules, good and bad examples, and guardrails.)

Canonical. Is there a single agreed-upon version that governs all others, and does everyone know which one it is? Copies and overlap are normal and often even necessary. What matters is that when two documents disagree, the answer is already decided and it is clear which wins. If this isn’t the case. AI will just pick one when it encounters the contradiction. Without asking you but with complete confidence in whichever version it happened to land on.

Adopted. Does the context live where your team already works, in a form that fits how they actually do their jobs? In other words: is everyone using it? If this fails, people use whatever is closest to hand (e.g., documents they have to fill in the blanks or often just prompts they write up in the moment), and the version in circulation becomes the real one regardless of what you decided. You can imagine how messy this gets very quickly.

Maintainable. Can the people who own it update it without breaking everything downstream that depends on it? Building your context is one thing. Maintaining it is another. Companies who figure out how to maintain and iteratively improve their context over time will have a massive competitive advantage. Without being able to maintain it, the knowledge goes stale, and you find out when a customer points out the error or discrepancies. Ouch.

It is important to understand that these four tests are not a scorecard you average out the answers. Instead, think of them as toll gates you have to pass through. A foundation that fails one of them isn’t eighty percent of a foundation. It’s a well-organized filing system, which is a genuinely useful thing to have but not a thing your AI can scale on.

What passing all four produces is simple to state: one truth, one home, one owner.

Now that you know what high-quality context looks like and how to test for it, let’s look at why the fast ways of getting there don’t work.

Subscribe now

3. The Three Shortcuts That Do Not Work

Once leaders accept that context is the constraint, the instinct is to close the gap as fast as possible, and there are three shortcuts I see business leaders try to take. None of them survives the four tests, and it’s worth understanding exactly why, because each one is genuinely appealing.

The first shortcut is data. Connect every system, clean up the CRM, and let AI figure out the rest. The problem is that records tell AI what happened. They don’t tell it what you hold by. Your CRM knows a deal closed. It doesn’t know why you would refuse the same deal today or what your team worked around to get it over the line. Data quality is real and necessary work, which is a critical element for your agents to execute, but it is not the full story because it still lacks the full context.

The second shortcut is documents. Export everything into a single folder, point the AI at it, and call it a knowledge base. But your documents are artifacts of past decisions. They were written by different people, in different times, for different audiences, and they disagree with each other in ways nobody has ever deliberately reconciled. They also skew heavily toward wins, because nobody writes up the deals they walked away from or the customer they should never have taken on.

The third shortcut is the tool. Yes, I know. It is very tempting to just buy a platform that promises to become your company brain. I understand the appeal completely — buying is fast, building is not, and the demo always looks like it solves the problem. Also, once you really think about it, creating a context layer can feel overwhelming and exhausting (there is a reason you avoided this work before). But think of those tools as containers. They are necessary, often excellent, and still empty when you buy them. You will still need to fill them.

Underneath all three sits the same missing step. Decisions. Nobody decided which of the three ICP definitions governs. Nobody decided which claims are safe to ship without proof attached. Nobody decided how the company sounds when it says no. Nobody decided what never goes out without a human reading it first.

Your organization has the knowledge. You just never explicitly decided what it holds by. In my upcoming book, From Scattered to Scaled AI, I call this the Decision Gap: the distance between having knowledge and having decided knowledge. And when you point AI at an undecided pile, it does what AI is built to do with ambiguity: It picks the most plausible version and states it as fact.

This is also, by the way, why you are still reading every AI draft before it goes out. Not because AI isn’t mature enough. But because you are giving it ambiguity. In other words: If you didn’t do the deciding before you give it to AI, you will have to keep making decisions later; now there are exponentially more decisions to make.

From Scattered to Scaled will launch October 1st, 2026. Become part of the movement and join my book lunch team.

Join my book launch team

4. Context Is Nothing New — Now You Just Can’t Get Away Without It Anymore

None of this is new. Playbooks, brand guidelines, positioning documents, documented sales processes — capturing what a company knows and holds by has been around for decades. But companies often skipped most of it and got away with it. Because the founder knew how to pivot the pitch mid-conversation in response to an objection. Because Sandra knew how to write aspirational thought leadership articles that were always on-brand. Humans learn by absorbing what they are exposed to. A new hire ramps slowly, asks questions in the hallway, watches how your senior rep handles the pricing objection, and six months later carries the context in her head like everyone else.

AI does not absorb. It makes assumptions. It fills gaps. And sometimes it hallucinates. Give it half the picture, and it fills the other half with the most plausible guess, delivered confidently, at a pace no human team can match. A salesperson who gets ten meetings wrong can be coached and retrained. An AI workflow running on a wrong foundation produces slightly off, slightly generic output every time it runs. An AI agent operating with inadequate context can cause hard-to-reverse, sometimes irreparable, damage.

And the stakes are about to rise. Gartner expects that 15% of day-to-day business decisions will be made autonomously by AI agents by 2028. Decisions. Not drafts. If you have never told an agent what you hold by, on what basis is it supposed to decide anything on your behalf?

One caveat worth stating: build context as a company asset, not inside a single platform. Whatever you codify should be portable enough that your CRM’s agents, your team’s daily AI tools, and whatever ships in eighteen months can all read from it. A context layer that lives only inside one vendor’s product isn’t an asset but merely a configuration.

5. How A Context Layer Actually Gets Built

So how do you actually build one? You don’t install it, and you can’t buy it. You go through a process, and the sequence matters more than most people expect. I developed a process I call the TrustLeader Method which consists of 5 steps: Extract, Codify, Structure, Implement, and Amplify.

Extract. Pull knowledge out of heads, habits, and recordings — the explicit, the implicit, and the tacit. The richest and most valuable material here is spoken and unstructured. Internal pipeline reviews, sales calls, the way your best people answer a hard question when nobody is polishing the answer for publication.

Codify. This is where you close the Decision Gap. Two jobs happen here. The first is resolving contradictions and committing to positions: definitions, standards, guardrails, process maps, judgment rules. The second is authoring what was never written down in the first place — the value proposition, the messaging hierarchy, the refusals. It is CEO-level work, and no platform will ever do it for you.

Structure. Give every truth one home. Turn the codified decisions into canonical, retrievable assets organized for using rather than only for finding, held in portable forms your systems can actually reach.

Implement. Embed the foundation into the workflows where the work actually happens, and train your team to use it and to QA against it. This is the Adopted test in practice. If it doesn’t reach the daily work, it might as well not exist.

Amplify. Scale what works. Speed becomes safe at this point, because every output is drawing from the same decided foundation, and every win compounds instead of evaporating.

Notice which two moves carry the weight: Extract and Codify are the ones everyone tries to skip, and they are the two no product will ever ship for you, because they are made entirely of your judgment.

This is the work I do at TrustLeader. The Foundation Five engagement exists to build exactly this layer: extracted from your people, decided by you, and structured so that every workflow and every agent you deploy from here on runs on a foundation that is authentically yours and genuinely hard to copy.

Remember: The platforms will continue to ship better containers, and you should use them. What goes inside is the part nobody can sell you.

Does your business have the foundations to scale AI? Find out where you stand and uncover your AI strengths and gaps with The AI Foundation Scorecard. Take the quiz now https://scorecard.trustleader.co/ai-foundation-scorecard.

Leave a comment

Nobody would do this to a human being.

And yet this is exactly what most companies do to their AI every single day.

Every AI tool in your business knows exactly what you explicitly give it about your business (plus what’s on the internet), and nothing else. It gets no shadowing your best people, no hallway conversations, no six months of quietly absorbing how things are done around here.

But AI is like an overconfident, eager-to-please intern. When it doesn't know something specific about your business, it doesn’t stop to ask you. It fills the gap with generic training data and the internet's statistical average, including what your competitors say.

That is exactly where AI slop comes from. Generative AI literally generates output by statistically predicting the next likely word. So, if we just give it a prompt like “You are an experienced blog writer. I want you to write me a blog post of 1200 words on the topic XYZ,” it will go and do exactly that. But everyone can do this. And almost everyone does this. So, the value creation equals ZERO.

So, what would you give your new salesperson? After 25 years of extracting, codifying, and structuring organizational knowledge to write GTM content, I know what it takes to give AI the context it needs to not only sound like you but also know your background, your values, and your experience. This enables your AI not only to create better content that requires less rework, but also to act correctly on your behalf.

This always boils down to five buckets. I call them the Foundation Five. That’s what this article is about.

Subscribe now

What the Foundation Five Is

The Foundation Five are the five pillars of codified knowledge and judgment that every GTM workflow stands on: your customer, your offer, your proof, your character, and your guardrails. They are built once, referenced by everything, and owned at the leadership level.

Three properties make them different from every other document in your company.

1. They are built once and improved forever. Not once per tool. Not once per team. Once. Every workflow, automation, and AI agent you will ever build references the same five pillars. When a pillar improves, everything built on it improves with it.

2. They are referenced, never redefined. A workflow narrows what it uses: this segment of the ICP, this pain point, these claims, this voice register. It never re-decides the foundations. Without this rule, every workflow team quietly answers the foundational questions in isolation. Marketing AI describes the company one way, sales AI another, customer service a third. Twenty AI workflows end up sounding like twenty different companies. I call this Standards Fragmentation, and it is the default state of AI adoption right now.

3. They encode judgment, not just information. Here is the thing most companies miss: a knowledge base can be assembled, but the Foundation Five must be decided. Each pillar holds two kinds of content: what you know, and what you hold by. Which claims you stake the brand on. Who you walk away from. The opinions you are willing to lose prospects over. Those are leadership decisions, and no intern with a folder of Google Docs can make them for you.

   

Share

The Five Pillars

Now that we know what the Foundation Five are, let’s have a closer look at what is in those context buckets:

1. The Customer. Deep context around exactly who you serve (and who you don’t), what fears, worries, and concerns they have, what payoffs they wish to see, what transformation for them looks like, and the events that put them in market.

• This includes the language they would use themselves, the questions and objections they usually raise, and the deeper fear underneath that they will not say in a first meeting.

• The judgment at the core: Who do you walk away from, and why? What problems are you able to solve better than anyone else, and why?

• Without this pillar, AI doesn’t know your exact audience, so the output is generic (it addresses everyone, and therefore no one in particular).

2. The Offer. All the context around what you sell, how you uniquely deliver it, and where you send people next. Your offerings are defined sharply enough that they cannot be confused with a competitor’s same-named offering, your method, your pricing logic, or the specific next steps you ask people to take.

• Most companies cannot clearly describe their value proposition or how they differ from their competitors. Before anyone trusts you, they need proof of three things: that you have the capability to do what you say (competence), that you will do it exactly as promised, every time (reliability), and that you will treat them ethically throughout (integrity). This pillar establishes the context around it.

• This includes what you refuse to sell or scope (even when a competitor does and a prospect asks) and a living register of the exact next step for each offer, so no article or proposal ever ends with nowhere to go.

• The judgment at the core: What will you never promise, no matter how much the deal wants to hear it?

• Without this pillar, AI cannot articulate what makes your offer different. Proposals read as accurate and articles educate your buyers and end without a next step.

3. The Proof. Deep context around why anyone should believe you: every claim you make, sorted by whether it’s safe to ship, requires sign-off, or never leaves the building, each one paired with the evidence that backs it, and where you stand against the alternatives your buyer is actually weighing.

• This includes proof of competence (you have the capability to do what you say), proof of reliability (you’ll deliver exactly as promised, every time), and proof of integrity (you’ll treat them ethically throughout) — the three things a buyer must believe before they trust anyone at all.

• The judgment at the core: Which claims do you stake the brand on, and which do you soften?

• Without this pillar, AI invents statistics or hedges into mush, because nothing was ever staked in the first place.

4. The Character. Deep context around what makes you unmistakably you: your values as they show up in hard decisions, the positions you’ve decided to take out loud, and how all of it sounds on the page — your vocabulary, your sentence patterns, your proprietary terms used exactly as written.

• This includes the practices in your industry you refuse to play along with, and a voice document, which is useful, but only one artifact inside this pillar, not the pillar itself.

• The judgment at the core: What are the opinions you’re willing to lose prospects over?

• Without this pillar, output is polished, correct, and instantly recognizable as machine-made. All manners, no character.

5. The Guardrails. Deep context around where the lines are: what never ships regardless of context, where your expertise genuinely ends, what “safe-to-ship” actually means, and the named situations where AI stops and a human decides.

• This includes the words, claims, and topics prohibited outright, decided once at the top rather than argued over per output.

• The judgment at the core: Where does the line sit when speed and safety pull in opposite directions?

• Without this pillar, the public screw-up isn’t a risk. It’s a schedule.

In addition to the Foundation Five, I always recommend you create a compilation of company facts (I call it the “Fact Book”). This can be a Sharepoint site or anything that can be maintained and updated without having to roll out a version document (think: Masterprompt PDF document) in your organization.

Share

How Are The Foundation Five Used

To pressure-test the Foundation Five, I mapped it against the top 15 GTM workflows for B2B companies with $5M to $25M in revenue: the blog writers, proposal generators, lead scorers, and chatbots that most of you are building or considering right now. For each one, I asked a simple question: how deep does it actually need to draw from each pillar? The pattern that emerged is the best illustration I have found yet for why the Foundation Five works the way it does.

Why Your Existing Documents Do Not Count

You may be thinking: we have most of this. A voice guide, brand book, an ICP slide, a messaging doc from the last agency engagement.

That’s great and that put’s you ahead of many B2B companies out there that are winging it without those elements properly documented. But those documents were written for humans. And that’s a problem for AI. For example: A tone-of-voice guide tells a human writer to sound confident, approachable, and authoritative, and the human fills the enormous gaps between those adjectives with judgment. AI cannot. It needs rules, not aspirations: vocabulary in and vocabulary out, sentence patterns, before-and-after examples, claims paired with evidence. This is a higher documentation standard than your business has ever needed, because for the first time, your reader has no intuition.

The good news: the Foundation Five is finite work. Five pillars, one home for every decision, built once. The alternative is to answer the same five questions in every workflow you ever build, forever, slightly differently each time. The Foundation Five is not the heavy option. It is the only option that ever ends.

How to Get Started

Start with an honest audit, one pillar at a time. For each of the five, ask two questions.

1. Where does this live today, in a document or in someone’s head?

2. Has this actually been decided, or just assumed?

Most leadership teams discover that their customer definition lives in three conflicting versions, their claims have never been sorted, and their guardrails exist only as instincts in the founder’s gut.

That discovery is not a setback. It is your baseline. The extraction and the decisions are leadership work, and they are exactly the work that turns AI from a generic tool into something unmistakably and defensibly yours.

If you want to know which of your five pillars is strongest and which one is quietly undermining everything you build, take the AI Foundation Scorecard. It takes a few minutes, and it will show you precisely where to start.

Take the AI Foundation Scorecard →

Then come back in three months and take it again. The score should move because foundations always show.

Leave a comment

So when the biggest CRM company in the market starts building exactly that, it is worth taking notice (and celebrating).

Here is what happened: HubSpot has now added two tabs into its Breeze menu: Context and Knowledge Vaults. Both exist to do one thing: give HubSpot’s AI more information about your business so its outputs are less generic and more accurate. That is the right idea. It is the idea I have been arguing for. And the fact that HubSpot is building it into the CRM tells you the platforms themselves now agree that context is what makes AI trustworthy at scale.

Subscribe now

What HubSpot Actually Built

HubSpot has had a marketing capability called “Brand” for a while now. This is where you will find your brand kit (logos, colors, etc.), your brand voice (don’t get me started on the HubSpot Brand Voice… that deserves its own article), and Brand Knowledge. Brand Knowledge has its own set of fields: company profile, ideal customer profile, products and services, additional product info, industry classification, customer sentiment, competitive landscape, content themes, tech stack, social responsibility, brand personality, mission and vision. This has been around for at least six months if I remember correctly, but I haven’t seen many companies fill this out yet. Some of it gets actually filled automatically by HubSpot. In addition, HubSpot now even offers to update the brand knowledge sections automatically if you ask it to crawl your website (more on that below).

However, this is different (for now) from the newly added Breeze AI Context and the Breeze AI Knowledge Vaults, even though they are almost one-for-one and are asking the same things. So, I assume (and I don’t have any insider knowledge here) that these will be consolidated soon.

Let’s look at the recently added Breeze AI Context and Knowledge Vaults in more detail. The first is Breeze AI Context. HubSpot describes it as the “information, facts, and knowledge about your organization that empowers AI across the platform.”

The new layer sits under Breeze > Context and is organized into three areas.

1. Business covers your identity and the shape of your company. There is a brand kit (voice, tone, and brand guidelines, though notice this one actually pulls from Marketing > Brand, more on that in a moment). There is identity and classification: name, domain, industry, business type. There is location and scale: headquarters, size, annual revenue, year founded. There is a business profile meant to capture your mission, positioning, and key differentiators. There is a market and ecosystem section for your main competitors and stakeholders, a technology stack overview, and a products and services section.

2. Customer is where you add ideal customer profiles, the characteristics of the companies and buyers that best fit what you offer, and personas, the detailed profiles of your typical customers with their roles, goals, and challenges.

3. Team and Processes is essentially about the individual user. Your name, email, job title, phone number, and an email personality, where you set the tone, style, formality, greeting and sign-off patterns, whether you use bullet points, your common phrases, and so on. I would assume this will be built out in the future.

The second part is the Knowledge Vault. HubSpot draws a clear line between the two: Context is the default, foundational information applied automatically across AI features (at TrustLeader, we have a similar concept called the Foundation Fives), while Knowledge Vaults are additional, use-case-driven context you add manually and assign to specific assistants or agents.

To create a Knowledge Vault, you click “Create Vault”, give it a name, add a description, and add files, HubSpot CRM objects, and lists. Only owners and super admins can manage and use vaults.

Now that you know what was added, let me tell you what I think about it.

Why This Is a Real Step Forward

Let’s start with what the move itself signals. HubSpot is telling you that AI needs a context layer and that it is critical. When a software company this large invests real engineering in that idea, it is strong evidence the idea matters. You do not build this unless you have concluded it is essential.

And you can see the recognition underneath it. HubSpot’s agents and AI are genuinely capable, but technology is not the constraint here. The AI is only ever as good as the data and context it runs on. A clean, well-maintained HubSpot database is necessary, and it is still not enough, because your records tell the AI what happened, not who you are, what you stand for, or what good looks like. HubSpot building a place for that second thing is an admission that the technology alone was never going to close the gap. That admission is the step forward.

There is a business logic to it too, and it is worth naming. The better your context, the better your AI and agents perform, and the better they perform, the more reason you have to keep your work inside HubSpot. Context is what makes the platform stickier. So I would expect serious investment in this part of the product over the next six months. I have no insider knowledge, but the incentives point in one direction.

In a previous article, I talked about the AI Context Gap, which is the space between what your AI knows about your business and what it actually needs to know to represent you accurately. Everything HubSpot has built here is an attempt to narrow that gap. That is the right target. The only question left is whether filling out form fields within a single platform actually closes it, and that is where the good news runs out.

Too Little, Too Late

The trouble begins the moment you actually try to fill it in.

Look at how HubSpot asks you to express your value proposition. You have to create phases consisting of a few words that are created like blog tags. And I want you to sit with how strange that is. How do you write a value proposition as a bundle of tiny phrases? You cannot. A value proposition is a piece of reasoning. It connects a specific buyer to a specific problem to a specific reason you, and not your competitor, are the one to solve it. Reduce it to tags, and you have kept the words and thrown away the logic.

But the value proposition is just an example. For some fields, you can enter a small paragraph. For others, you can choose from a list of preset choices. That is it.

I am currently writing the manuscript for my second book, From Scattered to Scaled AI. There, I draw a line between two kinds of documents. A human-readable document uses aspirational language: be bold, be trustworthy, be authoritative. It is too vague for AI to act on, because AI cannot interpret “authoritative” the way a person can. (BTW: that is exactly how you are supposed to define your brand voice in HubSpot…) An AI-accessible standard uses concrete, bounded, verifiable rules: documented examples, prohibited phrases, approved claims paired with the evidence that backs them, sentence patterns you actually use.

HubSpot’s tagged phrases are neither. They are a third thing, and it is the worst of the three. They are too thin for a human to reason from and too fragmented for an AI to reason from. There is no connective tissue, no before and after, no example of the claim done right versus done wrong. It is the shape of context without its substance.

Who Makes The Decisions & Who Fills In The Fields?

But there is an even deeper problem with this, and this problem does not get resolved when (not if) HubSpot improves the fields. As an organization trying to build your AI Context layer, you are facing empty fields. Who fills in these fields? Most of the time, it is an eager marketing manager or the person responsible for branding. They sit there and feel like they are just filling out some forms, but in reality, this work requires crystal clarity and thousands of microdecisions if done right. For most organizations, the clarity doesn’t exist. The vast majority of small and mid-size B2B companies cannot cleanly state what they sell, who they serve, and what problem they actually solve. Not because they are not good at what they do, but because that knowledge lives in the founder’s head and in the instincts of a few key people, and it has never been made explicit. It has never had to be.

That is what I call the Decision Gap: the space between having knowledge and having a position. The raw material is there, scattered and often contradictory, but nobody has sat down and decided what the company holds by. HubSpot gives you a place to record the decision. It gives you no process for creating it (which is fine, since they are a SaaS company and that really isn’t part of the job description). But it is worth noting that making these decisions is the hard part. Filling out the forms then becomes easy.

In my work with clients here at TrustLeader, we extract, codify, and structure knowledge: We extract the knowledge that lives in the founder's head and a few key people's instincts and get it out into the open where it can be seen. Then we codify it, which means resolving contradictions and deciding what the company actually holds by, so you have real standards rather than competing opinions. And we structure it so that an AI can retrieve it cleanly and you can keep it up to date.

That is the process that turns a scattered, half-articulated company into something an AI can represent consistently. None of it happens by handing someone a form. You are handed the boxes and left to guess at the answers, and if you guess the way most companies guess, you will fill them with the same vague, agency-written, sounds-smart-means-little language that created the problem in the first place.

Share From Scattered To Scaled AI

The Shortcut That Skips the Decisions

HubSpot, like a few other tools I have seen this with, offers to automatically fill out the information by having their AI crawl your website. This is an attempt to encourage more people to complete the brand knowledge section.

I get the intent completely. Filling this out is tedious, and anything that lowers the barrier gets more people to actually do it. This is the eating-your-vegetables problem of AI foundations. Everyone knows they should define who they serve and what they stand for. Almost nobody wants to sit down and do it, because there is always something that feels more urgent and more rewarding, the pizza in front of the salad. So a button that does it for you is genuinely appealing, and I do not blame HubSpot for offering it.

Here is the catch, and it has nothing to do with HubSpot. A scan can only reflect back the decisions you have already made. Ironically, for most companies, the website is one of the least-decided things they own. It was often written by an agency that was never fully clear about what the company does; it optimizes for sounding smart rather than being clear, and it rarely names the actual buyer, the actual pain, or the actual benefit in plain language. Scanning it does not create the clarity. It captures the vagueness that is already there and hands it back to you looking official.

That is the real point, and it holds no matter how good the scanning gets. The reason your context comes out vague is not the tool. It is that the underlying decisions have not yet been made, and no amount of automated fill-in will make them for you.

It Only Captures What Is Already Explicit

Organizational knowledge lives in layers. There is explicit knowledge, the things already written down. There is implicit knowledge, the things that live in people’s heads but could be articulated if someone asked the right questions. And there is tacit knowledge, the deepest layer, the instinct a founder has about whether a prospect is a fit before the first call is over, which they cannot fully explain even when you ask.

HubSpot’s context fields can only hold what is already explicit. They have no mechanism for surfacing the implicit and tacit layers, and those are exactly the layers where your real competitive advantage lives. The generic stuff is easy to capture and easy to copy. The proprietary stuff, the judgment and the point of view, is the hard part, and it is the part a set of form fields cannot reach.

The Two Limits That Matter Most

Everything above is a depth problem. In addition to those, I want to point out two structure problems:

There is no governance and no maintenance. Once you fill these fields in, who owns them? What triggers an update when your pricing changes, your positioning shifts, or you retire a feature? How does that change reach every place the old version was used? This isn’t a HubSpot problem per se, but it's one of the biggest challenges we will need to solve over the next 12-18 months. But it is worth noting because the tool has no concept of ownership or propagation for this. So the context you enter today goes stale. And stale context is not a neutral problem. It is worse than no context, because it is confidently wrong. It will have your AI promoting a price you no longer charge or a capability you no longer offer, and nobody will notice until a customer does.

And it is a silo. This is the one that matters most. However good HubSpot’s context layer gets, it lives inside HubSpot. And only within HubSpot. Your AI assistants (Claude, ChatGPT, Gemini, etc.) don’t see it. Your automations running through Make or n8n cannot see it. Every other tool that generates something in your name is still working from nothing, or from its own separate, quietly different copy of who you are. A context layer trapped within a single platform is not a foundation. It is one more island, and the whole problem you are trying to solve is that your knowledge is scattered across islands.

And here is the part that should give you pause, because it shows the fragmentation is not just between HubSpot and your other tools. It is happening inside HubSpot right now with HubSpot’s older Brand Knowledge and the new Breeze AI Context. That is the same company described in two places in one platform, which is exactly the fragmentation a foundation is supposed to prevent. If it can happen within a single tool, you can imagine how quickly it happens across five tools. A real foundation has a single canonical source of truth that every tool can access. HubSpot’s context can be a consumer as well as the supplier of that foundation. It cannot be the foundation itself.

Share

What Good Actually Looks Like

I am not going to hand you an implementation plan here, because the plan is not the point and it is different for every company. But the shape of the answer is simple to describe.

Your context foundation has to be properly extracted, so it captures the implicit and tacit knowledge, not just what was already lying around. It has to be codified, so the contradictions are resolved, and you have actual standards, guardrails, and definitions rather than fragments. It has to be structured so that an AI can retrieve it cleanly and you can keep it up to date. It has to be owned, so it does not go stale. And it has to live somewhere every tool can reach, so it serves all of your AI, not one platform’s.

HubSpot’s AI Context can be one of the places that foundation flows into. That is a good use of it. What it cannot be is the place the foundation is built, because building it is decision work, extraction work, and governance work, and no field on a form does that for you.

The Right Direction, Not the Destination

I want to end where I started, because I think both halves of this are true at once.

HubSpot building this is genuinely good news. It means the market is catching up to the idea that context, not tooling, is the differentiator. The platforms are validating the thesis. That is worth celebrating.

And a set of fields is not a foundation. Filling them in is not the work. The work is the decisions underneath them, the ones that force you to finally say who you serve and who you do not, what you always say and what you would never say, and no tool, however well built, makes those decisions for you.

If you want to know how far along your own foundation actually is, before it becomes a brand incident or a lost deal or a question from your board you cannot cleanly answer, I built a short diagnostic for exactly that. It takes a few minutes and benchmarks you across the five pillars: Extract, Codify, Structure, Implement, Amplify.

Take the Safe to Scale AI Readiness Scorecard →

Because the goal was never to fill in the fields faster. It was to become the kind of company whose AI is safe to scale from the outset.

Leave a comment

That gap — between what executives believe AI can do and what they are actually using it for — is what the researchers called the growth blindspot. The rest of the article goes on to calculate specific scenarios to prove the point. It’s a really interesting read, and I highly recommend you check it out.

But the reason it made me do a happy dance is that they clearly laid out something I have been noticing among mid-market B2B CEOs as well. Only, I call it the AI efficiency trap. And the mechanism that makes it a trap is more insidious than it looks. That is what this article is about.

If Your Primary Goal For AI Is Efficiency & Productivity, You Are Pursuing Linear Growth

Over 80% of executives cite increased efficiency and productivity as their primary goal in implementing AI. In other words, they are looking to make an existing business process faster or cheaper.

For example, early adopters report that content production time has dropped by 30–50% thanks to AI, and companies using AI for content operations report cutting content creation time by 60% while maintaining quality standards. The mechanism is straightforward: AI drafts the first version, and humans edit and approve it. The process is faster. The output is the same type of content, going to the same channels, serving the same strategy. Salesforce found that marketers save approximately five hours per week using GenAI tools for content-related tasks.

I can give you a dozen more examples of how Generative AI improves productivity and increases efficiency. Generative AI has been shown to increase knowledge workers’ productivity by 40%. Those are real gains, and they are nothing to cough at!

But when you use AI to improve efficiency, you are applying AI to an existing process to make it faster or cheaper. Marketing still produces the same type of content, just more of it. Sales still runs the same follow-up emails, just faster.

The process architecture does not change. The inputs do not change. The logic, the structure, the decisions embedded in the workflow — none of that changes. AI becomes a faster engine bolted onto the same chassis.

And here is where the ceiling appears: efficiency gains are, by definition, linear. A task that took 10 hours now takes 3, saving you 7 hours.

That is the gain.

But that’s also your ceiling.

You cannot reduce the same task below zero. The math has a hard floor.

The HBR’s data puts this in stark contrast. Even under generous assumptions (half of a company’s cost base amenable to AI, with a 10% average reduction), the impact on overall expenses is roughly 5%. The resulting boost to firm value is around 10%. And the article is blunt about this: costs can only be cut to zero, whereas your revenue has no ceiling. And the multiple investors’ place on growth expectations dwarfs the earnings impact of cost reduction. A sustained two-percentage-point lift in organic growth rate can increase firm value by 50%. A four-point lift can more than double it.

Efficiency cannot get you there. But here is the thing: the problem is not just that efficiency has a ceiling. The problem is what optimizing for efficiency does to your organization’s ability to reach scale.

Subscribe now

The Process You Optimize Becomes The Process You Are Committed To

Steve Jobs once said that the most common mistake engineers make is optimizing a process that doesn’t exist. Now, I am not saying that every process you bolted AI onto shouldn’t exist. But we should at least ask ourselves whether the process should be redesigned now that AI automation and agents can scale beyond the limitations of their human counterparts.

When you use AI to optimize a process, you are investing in it. You train people on it. You build automation around it. You create systems, habits, and reporting structures that presuppose it. The process becomes load-bearing. The better you execute it, the more committed your organization becomes to its continued existence.

And that commitment is exactly what makes the shift to scale so difficult. Because scaling with AI is not about making your current process faster. It is about asking whether your current process should exist at all in its current form. That is a fundamentally different question. And it requires a fundamentally different posture. To scale, you have to be willing to redesign — or eliminate — the process you just spent eighteen months optimizing. ve.

Boston Consulting’s research showed that currently, only 5% of companies achieve transformative AI value. They did not layer AI on top of how they already worked. They stepped back and asked: Now that we have these capabilities, what would this process look like if we designed it from scratch? ” Sometimes the answer was a redesigned workflow. Sometimes it was a scrapped business model. Sometimes it was a function that no longer needed to exist.

The companies in the bottom 60% (BCG calls them “laggards”) report minimal revenue gains despite sometimes heavy investment. They built faster versions of the same things because they were trapped in the efficiency trap.

What Scaling Actually Requires

Scaled AI looks different because it comes from different conditions. Scaling means you are not only growing exponentially but also improving as you grow. Getting more and better output with less input.

It requires a documented knowledge foundation — your extracted expertise, your codified standards, your defined guardrails — that AI can actually draw from. The real institutional knowledge that makes your company’s output recognizably yours.

It requires redesigned workflows, not just accelerated ones. Processes that were built with AI as a native component, not retrofitted with AI as an afterthought. This means questioning decision points, eliminating steps that exist only because humans needed them, and designing feedback loops that make each cycle better than the last.

It requires cultural and organizational change. The people, the skills, the way decisions get made. Efficiency optimization can happen within the existing org structure. Scaling cannot. The handoffs change. The ownership changes. The definition of what good work looks like changes.

None of these conditions emerges from optimizing what already exists. You cannot prompt-engineer your way into a redesigned workflow. You cannot automate your way into a knowledge foundation you have not built. You cannot efficiency-optimize your way into an organizational structure that supports compounding returns.

The path to scale runs through a different door entirely.

Why The Trap Closes

Now, here is what genuinely makes this a trap rather than simply a suboptimal choice.

Efficiency optimization produces visible results: Productivity metrics go up, content volume goes up, and turnaround times go down. The CEO can see the numbers moving. The team reports positive adoption. Everything seems to be working.

And because everything seems to be working, the pressure to ask harder questions decreases. There is no visible crisis. The quarterly review looks fine. The efficiency investments are paying off, at least in terms of the current reporting structure.

The majority of mid-market CEOs I am talking to describe themselves as “doing relatively well with AI,” ranging from “doing relatively well with AI” to “AI-forward.” But they don’t realize they are trapped in the AI efficiency trap.

Meanwhile, the companies on the other side — the ones building foundations, redesigning workflows, investing in the conditions that produce compounding returns — are not yet showing up in the data you are watching. Their advantage is not visible yet, but it is accumulating nevertheless.

BCG calls this the vicious cycle. Efficiency optimization yields marginal returns, insufficient reinvestment in the right capabilities, and further marginalization. Every quarter you spend optimizing for efficiency alone is a quarter when the foundation is not being built.

Every quarter, the foundation is not being built, and the gap between you and the companies that are building it widens. And the gap compounds. That is the nature of exponential returns versus linear ones. The distance does not stay fixed. It accelerates.

The Paradox Stated Plainly

The companies that feel most confident about their AI progress are often the ones most deeply caught in this trap. They are doing AI. They have adoption numbers. They have productivity data. They have a story to tell the board. What they do not have is a foundation. They do not have redesigned workflows. They do not have the organizational conditions that allow compounding. They are fast, but they are not scaling.

And the harder truth: the confidence produced by the efficiency results makes it less likely they will ask the question that could break them out of the trap. Because things appear to be working, the urgency to do something different does not feel urgent. Until the gap becomes undeniable — which, by then, means it is also very large.

You cannot optimize your way into transformation. You cannot accelerate an existing process into a fundamentally different one. The efficiency gains are real. They are also a ceiling, and increasingly, a commitment to staying under it.

The companies building the AI advantage that lasts are not investing in efficiency first. They are investing in the foundation that makes AI output authentically theirs — and then scaling from that. The ones investing in efficiency first are not on a path to that foundation. They are on a path away from it.

That is the trap.

Not sure which side of this divide you are on? Take the TrustLeader AI Foundation Scorecard — 20 questions, 5-8 minutes, an honest read on where you actually stand.

This isn’t an enterprise-only problem. The mid-market mirrors it almost exactly. And according to S&P Global Market Intelligence's Voice of the Enterprise: AI & Machine Learning, Use Cases 2025, 42% of companies have now abandoned most of their AI initiatives — up from 17% just a year earlier.

The easy interpretation is that AI is overhyped and doesn’t deliver. But that’s not what the research found. The divide between the 5% that work and the 95% that don’t has almost nothing to do with model quality, regulation, or budget. It comes down to approach — already-brittle workflows, weak contextual learning, and a fundamental misalignment in how the work actually gets done.

Here’s what nearly every one of those failed pilots had in common. They took an existing process and put a tool on top of it.

Subscribe now

The Core Mistake: Adding AI Instead of Rebuilding For It

Slapping a tool onto an existing process makes that process faster. Sometimes it makes it more efficient. But what it does not do is make it scalable or agentic.

When you automate a process as-is, you are automating it in a way that a human used to do the work, including all the steps that only existed because a human was doing them. The handoffs were really just someone walking a file to the next desk. The review stage existed because Sandra always caught the errors. The approval loop was there to protect against a mistake that a redesigned process wouldn’t make in the first place. You’ve made a flawed process run faster. You haven’t built anything that scales.

This is the part most companies skip, and it’s the uncomfortable one: real automation isn’t additive. It’s a teardown. (Cueing in Steve Jobs’ famous quote about the most common mistakes engineers make is to optimize processes that shouldn’t exist.)

If you want scalable automation and agentic systems that actually hold up and deliver trustworthy outcomes at scale, you have to be willing to take the process completely apart. Question whether each step should even exist. Rebuild it around what the work actually requires — not around how it’s always been done. That’s slower, harder, and far less satisfying than buying a tool and pointing it at the problem. It’s also the only thing that works.

And the teardown forces a decision you can’t avoid. For every step that survives, you have to answer one question: who or what owns it?

There are only three answers, and the difference between them is autonomy:

• A human owns the step.

• A workflow runs the step — with rules, structure, and human oversight at the edges.

• An agent owns the step outright, taking action without waiting for anyone’s approval.

That distinction is not academic, because as you move from human to workflow to agent, you gain speed and you give up the safety net. A bad human judgment is caught by the next human. An incorrect workflow output is caught at a checkpoint. But a wrong agentic action has already happened — the email is sent, the customer is charged, the post is live, the request is routed — before anyone knows there was a problem.

So the real question for every surviving step isn’t “can AI do this?” It’s “how much autonomy can this step safely carry?” And that question can only be answered one step at a time. Which is exactly why you have to take the process apart to answer it at all.

Ready to find out where your foundation stands? Take the AI Foundation Scorecard at scorecard.trustleader.co/ai-foundation-scorecard.

The Five-Lane Test

A few years ago, I read Paul Roetzer’s book Marketing Artificial Intelligence: AI, Marketing, and the Future of Business, and absolutely loved his simple, yet highly effective Human-2-Machine scale. Later, in his Smarter X course, he introduced the 4 questions that help organizations identify suitable use cases for generative AI. The questions for the latter were simple:

• Is it repetitive?

• Is it data-driven?

• Is it generative?

• Is it predictive?

These two simple but genius frameworks inspired me to think about how we could solve this problem: deciding which step should be owned by whom.

For now, I am calling it The Five-Lane Test. Once you’ve broken a process into its component steps, you run five questions against each one:

1. Is the input structured and predictable?

Agents run on consistent, well-formed inputs that leave very little margin for error. They don’t compensate for ambiguity the way a person does — they execute on whatever they’re handed. If a step keeps requiring its inputs to be interpreted, normalized, or cleaned up before anything can happen, that points to a human or a workflow, not an agent.

2. Is good output verifiable against a clear standard?

Verifiability isn’t just a quality bar but also a safety mechanism. An agent needs a definition of “done” against which it can be measured. If you can’t say in advance what a correct output looks like, you can’t catch a wrong one before it does damage. Steps where “good” is subjective, relational, or judgment-dependent aren’t ready for autonomous execution.

3. Can this run without contextual judgment?

Contextual judgment is the ability to weigh what isn’t in the brief — the relationship history, the timing, what someone clearly needs but didn’t say. Agents act on what’s in front of them. They don’t read the room or reliably infer intent. Any step that routinely requires a person to make a call, the inputs alone don’t determine that it needs a human in the loop.

4. Can this exist without a human relationship in the loop?

Some steps are relationship moments dressed up as tasks. A follow-up to a warm prospect, a check-in after a hard conversation — these carry relational weight an agent can’t assess. AI can draft, prepare, and suggest. But the step belongs to a human when the relationship is part of the output, not just the channel it travels through.

5. Are the consequences of an error recoverable?

Agentic errors compound fast. The question isn’t whether errors are possible — they always are. It’s what happens when one occurs. If the failure is public-facing, financially material, or damages a relationship that’s hard to repair, the step doesn’t go to an agent — no matter how cleanly it scores on the first four.

The answers assign the step to a lane.

• Five yeses, and the step is Agentic — hand it off and let it run.

• Five nos, and the step is Human — it belongs to a person, and trying to automate it is how you end up in the 95%.

• Anything in between — and it’s Workflow.

Here’s the part most people resist: the majority of your steps will land in Workflow. That is not the test failing. That is the test working. Workflow is where most real work belongs — structured enough to benefit from automation, consequential enough to need rules and oversight, but not safe enough to hand off entirely.

The Five-Lane Test doesn’t push you toward agents everywhere. It protects you from deploying them where they don’t belong, which is precisely the mistake driving that failure rate.

What You Get Is a Redesigned Process, Not a Tool List

The output of this exercise is not a vendor shortlist. It’s a redesigned process map: every surviving or new step is assigned to its lane, with the handoffs between lanes defined in advance. Where a step passes from human to workflow, or workflow to agent, that boundary is now explicit — and those handoffs are where most AI implementations quietly fall apart. They only become visible once you’ve taken the process apart.

This is the actual difference between the 5% and the 95%. The 95% added tools to old processes. The 5% rebuilt the process around what each step genuinely needs, then chose tools to fit. MIT’s clinical phrase for the failure is “flawed enterprise integration.” The plain-language version: they automated work they never redesigned. The redesign comes first. The tools come after. You cannot buy your way past the teardown.

The Window Is Still Open

So look again at that 95%. It isn’t proof that AI doesn’t work. It’s proof that most companies tried to skip the hardest, least glamorous part of the job — taking the process apart and rebuilding it from the ground up.

The Five-Lane Test is how you do that rebuild deliberately, instead of by guesswork. It won’t choose your tools for you. What it will do is make sure that when you finally do choose them, every step is already sitting in the lane where it belongs.

That’s the work the 5% did. It’s still available to everyone else.

Ready to find out where your foundation stands? Take the AI Foundation Scorecard at scorecard.trustleader.co/ai-foundation-scorecard.

Most likely, they will sell. (Or quit on the spot because they think you lost your marbles…) They will use their experience, their general training, and whatever they vaguely picked up about your product. They will have ten conversations that day, and most of those conversations will sound confident, polished, and professional. Some might even close.

But they will struggle to clearly articulate your value proposition, explain product fit, differentiate your products from other solutions, and so on. Now imagine, every time they don’t know something, they quickly fill in the blanks with what they know from previous sales jobs or just make it up on the spot… They will have a hard time building trust and often even erode it when caught out.

Now imagine that same untrained salesperson drafting every email, writing every proposal section, and sending every follow-up across your entire pipeline. The output is still “fine.” It is just not yours. And the damage compounds. Quietly, daily, across every touchpoint your market sees.

No one would do this when hiring a new salesperson. But most companies do exactly that when they implement a new AI tool! The problem is that AI workflows or agentic systems can scale almost infinitely. So, instead of potentially eroding trust over 10 conversations a day, an AI prospecting agent could reach hundreds of potential buyers in just minutes. AI is very powerful, but sometimes we treat it like this all-knowing black box that will magically transform our business, and when it doesn’t, we tell ourselves that the tool wasn’t good and we just need to find a better one.

But the problem isn’t a tool problem. It’s a lack-of-context problem.

Subscribe now

The Mistrust Is Not The Problem. It Is The Signal.

A lot of CEOs I talk to are excited about AI, but they are (often privately) reluctant and hesitant. They will not say it at a board meeting, but they tell me privately: they do not fully trust their AI to represent them accurately.

So they hesitate to scale it further.

Most AI consultants treat that hesitation as a failure of nerve. “You need to adopt AI or your competitors will beat you to it.” That advice is not only wrong but dangerous. There is a reason the CEO is feeling mistrust. (And if you’re feeling this too, I suggest you lean into this gut feel more.) They have seen the generic output. They have watched AI polish off the rough edges that make their business different.

Hesitating to scale that further isn’t avoidance because of nerves. It is good judgment with the wrong frame. You do not have a trust problem. You have a context problem. Once the context is in place, the trust is earned — and scaling becomes obvious.

The 60% Gap

Boston Consulting Group’s The Widening AI Value Gap: Build for the Future 2025 report surveyed 1,250 senior executives across more than twenty-five sectors. The findings define exactly what is at stake.

Only 5% of companies qualify as “future-built” for AI — systematically generating substantial value across functions. Another 35% are scaling and beginning to see value. The remaining 60% are laggards: minimal returns despite real investment.

The gap is not small. Future-built companies achieve 1.7x revenue growth versus laggards, 3.6x three-year total shareholder return, and 1.6x EBIT margin. In areas where AI is applied, leaders see twice the revenue increase and 40% greater cost reductions.

And BCG is explicit about why the 60% are stuck. It is not budget. Many laggards have significant AI tool budgets. It is not the technology. They are using the same models everyone else is. BCG’s phrase: “they don’t yet have the proper capabilities for scaling AI in place.” What they mean by “capabilities” is the foundational layer — operating model, data foundation, methodology, governance — without which AI cannot compound.

One critical part of that layer is what I call the AI Context Layer, or when I help my clients install their, I call it the Trust Cortex.

What The AI Context Layer Is

The AI Context Layer is an infrastructure layer that provides your AI with the knowledge, background information, standards, guardrails, and other context it needs to not only accurately reflect your company, products, and achieved outcomes, but also your beliefs and values, the things you stand for, etc. It allows the AI to talk like you, sound like you, and act like you.

It is not software or a platform. Not a single document or a single tool. Not something you buy. It is a layer your company builds over time (with an initially more intense ramp-up as you build the structural construct and the must-have foundational assets) that makes every AI tool, agent, and workflow you deploy yours reliably.

It contains a combination of things that make your company’s output recognizably yours: your codified value proposition, your voice and tone, your operating standards, your structured expertise, and your methodology. Everything that currently lives in your head, your founder’s head, your senior salesperson’s head, and the ten years of customer conversations you have never written down.

With it, your tireless, infinitely scalable contributor finally has the briefing your new salesperson never got. The output stops being “fine.” It starts being unmistakably you — across 5000 emails, 500 proposals, every customer touchpoint AI now reaches. AI stops scaling your gaps and starts scaling the thing that made your business work in the first place.

This is the difference between Scattered AI and Scaled AI. Scattered AI runs on tools without context — productive, busy, generic. Scaled AI runs on tools the company has trained. Same models. Different outcomes.

The specific components of the Trust Cortex — what I call the Foundation Five — are the subject of the next article in this series.

What Happens If You Don’t Build One

The gap between the laggards and those scaling is widening exponentially, not narrowing. The companies building the context layer now are compounding two things at once: their AI advantage and their differentiation. The companies still cycling through tools are compounding neither. The math won't get any more forgiving next year.

You do not have to build this with anyone in particular. You do not even have to call it the Trust Cortex. But if you are a founder-led B2B company sitting somewhere in that 60%, you have to build something that does this job — or you stay in the 60% while your competitors do not.

Foundation first. Scale second.

But then, six months later, you sit down to update the value proposition and messaging — and discover it now exists in dozens of projects as a static PDF as well as in nine slightly different versions across five workflows. Your product offering and positioning shifted, but your marketing and sales output still includes bits and pieces of the old positioning, confusing buyers and eroding trust.

Now you realize, you have a problem. A big, hairy monster problem, because besides the value proposition, there are dozens of documents that will require constant maintenance to remain current. This is what hits most companies around month six of their AI journey. Not a tool problem. Not a model problem. A knowledge organization problem.

Subscribe now

The Misconception Most CEOs Hold

Most CEOs assume their AI problems are retrieval problems. They think that by feeding their AI too much context, the AI will get confused. Granted, an AI will have an easier time finding a specific piece of information in 50 documents compared to 800 documents. But this isn’t the limiting factor anymore if structured properly. For example, Claude Sonnet 4.6 can hold roughly 1 million tokens (or 1,500 pages of text) in a single conversation. That’s more than most $5–50M companies have in foundational knowledge across their entire go-to-market function. Loaded correctly, AI finds what it needs. So, that’s not the problem.

The problem is somewhere completely else: We have always organized company-specific knowledge into composite documents, such as style guides, sales playbooks, onboarding handbooks, and so on. Specific documents for specific processes executed by humans, which is fine if you are working at human speed. They work fine for a single conversation.

But if you want to build AI workflows, automations, and agentic AI systems, you will need to build something that is scalable and robust. Something that serves as the single source of truth, powering more than a dozen workflows. Something that gives details on who you are, who you serve, and what you sell, but also knows your point of view, the issues you take a stance on, your special processes, customer case studies and testimonials, evidence for the claims you make, and so much more. And something that is easy to maintain and update, so you can truly scale. Because you cannot scale with AI if the context AI works with can’t be maintained easily.

I call this “something” the AI Context Layer (or Trust Cortex^TM).

The Four Types Of Knowledge

Imagine all of your organizational knowledge (whether it is explicitly documented in composite documents or locked in people’s minds) as a huge ball of tangled yarn.

To build an AI Context Layer (or Trust Cortex^TM), you first need to untangle this ball of yarn so you can make it AI accessible. At TrustLeader, we deploy a proprietary methodology to do this, consisting of 5 pillars: Extract, Codify, Structure, Implement, and Amplify. I will go into the steps and frameworks within this methodology in future posts, but today I want to tease apart the four main knowledge types in your organization, looking at them through the lens of the purpose they serve. This is important because each type requires s a different home, a different owner, and a different update rhythm so they can be accessible to AI in an optimized way.

1. Declarative Knowledge

The facts, definitions, and reference material AI needs to know what is true about the company.

This is the relatively stable and most often the already explicit knowledge — the things you have decided are true and that you commit to holding by. Owned by leadership or subject matter experts (like product managers): the people who have the authority to decide what the company stands for.

Examples:

• Your ICP definition and buyer persona profiles

• Your product and service descriptions

• Your pricing structure and packaging

• Your approved claims and the evidence behind them

• Your corporate voice guide

• Your brand guardrails — vocabulary you use, vocabulary you don’t

• Your value proposition

• Your pain and payoff pairs

• Your competitor landscape and differentiation

• Your mission, vision, values

• Your industry definitions and the category vocabulary your team has agreed on

• Your regulatory boundaries and compliance constraints

2. Procedural Knowledge

The step-by-step guidance AI needs to know how the company does what it does.

This is your operational know-how, but codified. Owned by the people who run each operation — sales ops owns the sales process, marketing ops owns content production, customer success owns the renewal playbook.

Examples:

• Your sales process, from inbound lead to closed deal

• Your qualification framework and disqualification triggers

• The canonical procedure for writing a blog article — and the smaller overlays for specific blog types

• The proposal-drafting process

• The customer onboarding sequence

• The renewal and expansion playbook

• Escalation rules — when AI hands off to a human, and to whom

• The content brief template and how to fill it out

• The discovery call structure and question flow

• The objection handling decision tree

3. Episodic Knowledge

The records of specific things that have happened, which AI draws on as proof, pattern, and precedent.

This is your institutional memory. Owned by whoever captures and curates the record — typically sales ops, marketing ops, or customer success.

Examples:

• Win/loss records — every deal, why it was won or lost, by which competitor

• Customer case studies and success stories

• Customer testimonials and quotes

• Sales call recordings and transcripts

• Discovery call notes and themes

• Customer support tickets and resolution patterns

• Past blog articles, with metadata and performance data

• Past proposals — which ones won, which ones lost

• Past email campaigns and their results

• Product feedback from customers

• Implementation notes from past client engagements

• Specific quotes from prospects that signal pattern shifts

4. Meta Knowledge

The information about the knowledge itself, which tells AI and the team what to trust, what is current, and who owns it.

This is the layer most companies skip entirely. Owned by whoever runs the knowledge architecture — at smaller companies, an operations lead; at larger ones, a knowledge ops or AI ops function.

Examples:

• Document owner — the named person accountable for each artifact

• Last updated date

• Update cadence — quarterly, annually, ad hoc

• Authority level — who has approval rights to change the artifact

• Status — draft, approved, deprecated, under revision

• Version number and change log

• Source attribution — where the knowledge came from

• Confidence level — committed position, working hypothesis, exploratory

• Dependencies — what other artifacts reference this one and would be affected by a change

• Permissions — who can read, who can edit, who can approve

• Retention rules — when knowledge expires

Most companies dump all four types into one place — usually a shared drive or a wiki — and treat them as interchangeable. They aren’t. The shape of the knowledge tells you how to handle it. Get this taxonomy right, and your AI capability compounds. Get it wrong, and every new workflow you build inherits the mess.

Sorting By Type Is Only The Start

Understanding the four types of knowledge buckets is only the start. Once you extract and codify your knowledge, you will need to provide an appropriate structure for your knowledge to live in. Finally, you need to build a system that stays organized as the knowledge base grows. And that’s where most companies fail.

Six considerations determine whether your knowledge base scales or whether it quietly fragments:

1. Schema discipline — once you have 500 records, restructuring is painful

2. Cross-system relationships — how Notion, Airtable, Drive, and your CRM relate as one system

3. Naming and identifier conventions — persistent, predictable, versioned

4. Canonical vs. synced sources — when information lives in two places, which is authoritative

5. The serving layer — how knowledge reaches workflows (manual, pre-loaded, automated)

6. Update propagation — how a change in the source reaches every workflow that depends on it

Each one is worth its own article. I’ll write those soon, but the point for today is that Scattered AI is what happens when you treat all knowledge as the same thing — when your brand guide, your sales process, your win/loss data, and your governance information all live in one composite document or one undifferentiated shared drive. It works for a while… until it doesn’t.

You cannot scale your AI if you cannot scale the context your AI runs on.

Scaled AI runs on a context layer containing knowledge that has been sorted by purpose, placed in homes that fit its shape, owned by people who can keep it current, and structured to compound rather than fragment.

The four types are how you start moving from one to the other.

Leave a comment

However, what played out over the next 10-12 years was like going from watching an exciting race turn into a massive car pileup… in slow motion. It was painful to watch companies trying to make sense of this new world.

Ten years later, in 2021, McKinsey reports that 70% of digital transformation initiatives failed to meet their objectives. Gartner found that only 48% met or exceeded their targets. Globally, failed transformation efforts cost businesses an estimated $2.3 trillion a year.¹

But we do what we always do in business. We declared the lesson learned. We moved on.

Now, with the excitement and hype around AI, I can’t help but feel a sense of déjà vu when I talk to CEOs about their AI strategy. I see the same mistakes being made now that caused Digital Transformation projects to fail.

Subscribe now

And the data indicates the same: 95% of AI pilots fail to scale (MIT NANDA, 2025).² Only 6% of organizations qualify as AI high performers (McKinsey, State of AI 2025).³ The gap between the companies seeing transformational value and the ones running expensive experiments is not closing — it is widening. BCG found that 5% of firms are now “future-built” on AI. They already see 1.7x revenue growth and 3.6x three-year total shareholder return. The other 95% are losing ground.⁴

Below, I want to share some insights and lessons learned because when I look at what is actually happening inside organizations trying to scale AI right now, I see the same failure modes, the same leadership gaps, and the same structural mistakes playing out in near-identical sequence.

And I think most leaders are ignoring the fix that eventually worked — again.

What Digital Transformation Declared Itself to Be

Initially, the definition of Digital Transformation was as vague as it was ambitious: it was about becoming a different kind of company, not a better version of the same company. That was aspirational and directionally correct, but how did you operationalize something like this as a project with a budget and a deadline?

(Sounds familiar? But I am getting ahead of myself.)

From 2012 to 2015, the definition narrowed somewhat to the use of digital technology to fundamentally change how a business creates and delivers value. But the gap between that declaration and what actually happened inside organizations was enormous.

Digital Transformation meant different things to different people:

• CEOs announced customer experience reinvention.

• CIOs operationalized cloud migration and data infrastructure.

• Innovation teams ran pilots.

All three definitions lived within the same organization, but they were rarely coordinated, measured different things, and were accountable to different people.

Most Digital Business Transformations were spearheaded by IT. The most instructive example of what happens when technology outruns governance and business ownership was GE. Under Jeff Immelt, GE invested an estimated $1 billion into its Predix digital platform — a bet that GE would become a top-ten software company by 2020.⁵ Predix generated minimal financial ROI. GE’s shareholder value, which had peaked at over $500 billion in 2000, had collapsed to around $70 billion by the time Immelt departed. The technology wasn’t the problem. The business model and accountability structure around it were. The “digital shift” had consumed enormous capital while delivering almost nothing measurable to the business.

GE was not uniquely incompetent. It exemplified a specific, widespread failure mode: visionary intent, technology investment, and no clear business ownership of outcomes.

What Successful Companies Did Right To Succeed

In hindsight, it is easy to see that the shift that changed outcomes was exactly what you would expect in and exactly what was resisted for years in practice: Successful Digital Transformation initiatives were business-led, IT-supported. Not IT-led, business-consulted.

The moment a P&L owner became accountable for transformation outcomes, the failure dynamic started to change. The question shifted from “what can this technology do” to “what business result am I accountable for, and how does this technology help me get there?” (Again, remind you of anything?)

Subscribe now

Beyond the ownership shift, three other patterns separated the organizations that eventually scaled from the ones that ran perpetual pilots:

• Workflow redesign before technology selection. The companies that succeeded mapped the business process first, identified the constraint, and then chose technology to remove it. The ones that failed bought the platform and tried to redesign around it. The technology made the decisions that business leaders should have made.

• Learning infrastructure. Documented context, internal retrospectives, and institutional memory of what worked. Each implementation started from a higher baseline than the last. Without this, every pilot started from zero, repeated the same mistakes, and produced no organizational learning that anyone could act on.

• The lighthouse project model. Not a hundred pilots but one high-visibility use case, owned by a business leader, with clear before-and-after metrics. Held up internally as proof of what was possible. Used to build organizational belief and fund the next wave.

For most, digital transformation didn’t end with a competitive moat. It ended with survival. The organizations that came out ahead weren’t the ones with the boldest slide decks. They were the ones that iterated long enough, under enough competitive pressure, that the new way of working eventually became the only way they knew.

The digital natives didn’t transform. They were born into it. Everyone else adapted, slowly and messily, or fell behind.

Where We Are With AI Right Now

So, what does that have to do with AI?

Artificial intelligence is not new. Machine learning, natural language processing, and predictive analytics have been embedded in enterprise software for decades. What changed in November 2022 was the barrier to entry. ChatGPT put a capable AI interface at the fingertips of anyone with a browser and an internet connection. For the first time, businesses didn’t need a data science team, a six-figure implementation budget, or a vendor relationship to experience what AI could do. They just needed to type a question.

That moment triggered what most analysts are now calling the Generative AI Era — the broadest, fastest wave of enterprise AI adoption in history. And like every technology wave before it, it comes with equal parts genuine potential and unchecked hype.

Here is where that wave actually stands right now: According to the Gartner® 2025 Hype Cycle for Generative AI, generative AI has passed its Peak of Inflated Expectations and is moving toward the Trough of Disillusionment, a natural phase within a technology’s lifecycle where the gap between what was inflated hype doesn’t match what the technology actually delivers.

Right now, we are with AI, where Digital Transformation was 2012 to 2015 — wide adoption of the idea, thin adoption of the discipline, and cracks already forming in the foundation.

Excitement and subsequent adoption are high: 88% of organizations use AI in at least one business function (McKinsey, State of AI 2025).³ However, only 6% qualify as high performers. These are the few companies that are seeing real, measurable business impact. Roughly 33% are scaling beyond initial pilots. The majority, though, are in the experiment phase, running activity, and calling it progress.

BCG’s research across 1,250 firms adds the dimension that makes this urgent rather than merely interesting.⁴ 5% of organizations are already “future-built” — meaning they are past experimentation, into genuine AI operating capability. They are seeing 1.7x revenue growth and 3.6x three-year total shareholder return compared to their peers. The other 95% are not simply behind. They are approaching AI very differently (see below) and getting stuck in the same virtuous cycle for the top performers, but a vicious one for them. The result: The value gap is widening.

The excitement is real. The investment is real. The potential is real. But the cracks are equally real.

The Parallel Nobody Is Making Explicitly Enough

There are too many parallels between the Digital Transformation era and the Generative AI era to ignore them. Here are four that I found are most instructive to help us avoid repeating the mistakes of the past:

The IT ownership problem → the innovation lab problem. In digital transformation, the CTO and CIO were accountable, and business leaders were consulted. The result was technically successful implementations that solved no business problem anyone cared enough about to fund at scale. In AI adoption today, the same dynamic is playing out with siloed centers of excellence, innovation labs, and AI task forces. Business leaders are excited in theory, but they are not currently accountable for the outcomes.

Platform before process → tool before workflow. McKinsey’s State of AI 2025 found that workflow redesign is the single strongest correlate with real EBIT impact from AI. Yet only 21% of organizations have done it.³ The other 79% selected their AI tools first — and are now trying to redesign work around them. This is the GE mistake repeated.

Change management as an afterthought → AI layoffs before org-wide retraining. Gartner found that employee resistance is one of the most common causes of transformation failure.⁶ The AI equivalent: Deloitte’s State of AI in the Enterprise 2026 (n=3,235 leaders) identifies the skills gap as the #1 barrier to AI scaling.⁷ IBM’s 2025 CEO Study (n=2,000 CEOs) found that 50% of CEOs say rapid AI investment has created disconnected technology across their organizations.⁸ The technology is moving faster than the people, and organizations are not investing in the bridge. The sad thing (and some say a cowardly move): CEOs would rather lay off a large portion of their staff after achieving efficiency and productivity gains, rather than create a culture change, restructure, and retain a large part of their organization to adapt to the new situation.

Shadow IT → Shadow AI. In digital transformation, ungoverned technology proliferated across business units, such as different platforms, different vendors, and different data models, until nothing could connect. Five years later, organizations were running consolidation programs to undo what decentralization had created. In AI adoption today, shadow AI is the direct equivalent. Employees are building and using AI tools outside any governance structure, creating the same fragmented, inconsistent capability at scale. MIT NANDA quantifies it. Gartner flags it as a primary enterprise risk. Most organizations are not taking it seriously until they have to.

The value gaps this creates are compounding and actively widening.⁴ Every quarter an organization spends running ungoverned pilots without business ownership is a quarter the 5% use to extend their lead.

What is important to note, though, is that AI is moving at a breakneck speed. According to Microsoft AI’s CEO, Generative AI will automate most white-collar work by the middle to late 2027. If you are not already building towards this reality, you are falling behind.

What the Leadership Shift Actually Looks Like

If you are reading this and recognizing your organization in the potential failure territory above, the fix is not a new AI strategy document. Or different tools. It is a set of structural decisions that most leadership teams are actively avoiding because they require accountability, not just enthusiasm.

Get a business executive with P&L accountability to own your AI outcomes. Not the CTO. Not the Chief AI Officer reporting to the CTO. But a business leader who is measured on revenue or margin impact, and for whom AI is the mechanism, not the mandate. This is the single highest-leverage shift — everything else follows from it or fails without it.

Workflow (re-)definition and (re-)design happen before tool selection. The question your leadership team should be asking is not “what can we do with AI?” It is “where is the constraint in this business process, and can AI remove it?” That question forces specificity. It forces business ownership. And it produces use cases that are actually worth scaling.

Run a few lighthouse use cases, not hundreds of experiments. Defined success metrics. A clear before-and-after that leadership can point to internally. The lighthouse is not the end of the program — it is the proof point that builds organizational belief, demonstrates what business-led AI ownership looks like in practice, and creates the internal case study that funds the next wave.

Learning infrastructure built from the start. MIT NANDA’s diagnosis of why 95% of pilots fail is precise: failure is not the model. It is missing learning infrastructure — no documented context, no institutional memory, no systematic capture of what worked and what didn’t.² Every pilot that runs without this produces no organizational learning. The next pilot starts from zero. This is fixable, but it requires someone to own it.

Governance built in, not bolted on. Deloitte found that only 1 in 5 companies has mature governance for agentic AI — despite adoption surging.⁷ Gartner’s research shows that organizations with AI governance platforms are 3.4x more likely to achieve high effectiveness in AI governance.⁹ The organizations building governance now are not slowing themselves down. They are preventing the consolidation crisis that ungoverned scaling always eventually requires.

The Question Worth Sitting With

68% of senior executives already fear that their AI will fail due to poor integration with core business operations (IBM IBV AI by 2030 Study, n=2,007).¹⁰ While they are right to be cautious, they are making an incomplete diagnosis.

Companies that succeeded with Digital Transformation stopped treating it as a program and started treating it as a new operating model — built iteratively, owned by the business, with enough patience to let it become the way work gets done rather than the thing the innovation team was doing over there.

The organizations that are winning with AI right now made that shift early. They are not running more pilots or buying more expensive tools. They have a business leader accountable for outcomes, a lighthouse use case producing measurable results, a knowledge foundation, and a governance infrastructure that ensures their AI capability scales rather than scatters.

Most leaders ignored it the first time. The ones ignoring it now are betting that the second failure will be cheaper than the first. It won’t be.

— Hannah

Leave a comment

Sources:

1. McKinsey, Common Pitfalls in Transformations —https://www.mckinsey.com/capabilities/transformation/our-insights/common-pitfalls-in-transformations-a-conversation-with-jon-garcia

2. MIT NANDA, The GenAI Divide: State of AI in Business 2025 (Aug 2025) — mlq.ai

3. McKinsey, The State of AI 2025: Agents, Innovation, and Transformation (Nov 2025) — mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

4. BCG, The Widening AI Value Gap: Build for the Future 2025 (Sep 2025, n=1,250 firms) — media-publications.bcg.com/The-Widening-AI-Value-Gap-October-2025.pdf

5. CNBC, Here’s Why GE’s and Ford’s Digital Transformation Programs Failed (Oct 2019) — cnbc.com/2019/10/30/heres-why-ge-fords-digital-transformation-programs-failed-last-year.html

6. Gartner, How to Overcome Employee Resistance to Organizational Change

7. Deloitte, State of AI in the Enterprise 2026 (n=3,235) — deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html

8. IBM Institute for Business Value, 2025 CEO Study: 5 Mindshifts (May 2025, n=2,000) — ibm.com/thought-leadership/institute-business-value/en-us/report/2025-ceo

9. Gartner, Market Guide for AI Governance Platforms (Nov 2025) — https://www.gartner.com/en/documents/7145930

10. IBM Institute for Business Value, AI by 2030 Study (Jan 2026, n=2,007) — newsroom.ibm.com/2026-01-19-IBM-Study-AI-Poised-to-Drive-Smarter-Business-Growth-Through-2030