The context gap: good AI advice for the wrong company

Two minutes, now

Ask it something only your company could answer.

Open whichever assistant you use most and ask one of these. Read the answer and then ask yourself the one question underneath. That is the whole test and it settles the argument faster than the rest of this page will.

Pick one and paste it in.

“How do we decide whether to take on a client who is a bad fit but pays well?”
“What does good work look like in my team, and what would get sent back?”
“Who needs to be involved before this goes to a customer, and in what order?”

Then ask yourself: could this answer have been written for any company in my sector? If it could, your context is not in the system, and every answer you get from here on will be well-reasoned advice for a company that is not yours.

Most people expect the answer to be wrong. It is usually worse than that, because it is right. Right and general, which reads as authoritative and gives you nothing to act on that you did not already know.

Out in the wild

Three answers you have already had.

Each of these is genuinely good advice. Underneath each one is the thing it would have needed to know to be useful to you.

Start with a pilot in one team, measure for a quarter, then decide whether to scale.

What it did not knowThree of your teams have been doing this since February, and your last pilot ran for five quarters because nobody was allowed to end it.

Align stakeholders early and secure executive sponsorship.

What it did not knowYour executive sponsor is the person who says yes to everything and then quietly does not turn up, and everyone in the company has already routed around that.

Establish clear criteria for what constitutes acceptable output quality.

What it did not knowYou have those criteria. They live in one senior colleague's head and get applied on Thursdays, and the last two attempts to write them down stalled at the second meeting.

The pattern is the same each time. The advice describes a process that exists on paper somewhere, and your company runs on the version that people actually do, which nobody has written down and which is the only part that would make the advice worth anything.

Defining it

When everyone has the same models, the difference has to come from somewhere else.

Two years ago the advantage was access. Whoever had the better model, the better tooling or the earlier contract was ahead. That window has closed for most companies, because your competitor buys from the same vendor and pays roughly the same price.

Harvard Business Review made the argument directly in February 2026: once everyone has the same models, the same AI-enabled tools and the same vendor ecosystem, organisational context becomes the thing that separates companies.1

Context is demonstrated execution: the workflows teams actually follow across systems. Rohan Narayana Murty, in Harvard Business Review, February 20261

Read the word demonstrated. Your stated process is a document somebody wrote in 2023. Your demonstrated process is the order people actually involve each other in, the exceptions that make somebody pick up the phone, the judgement calls that repeat every week and that nobody has ever had to explain out loud. That second one is the asset, and in most companies it exists only in people.

Three routes

Everybody picks one of these. Each one has a price.

All three of these are defensible and I have watched all three work for a while. The third one is heavier than the other two and it is still a choice. What matters is choosing on purpose, because most companies arrive in one of these without ever deciding.

A

Keep it to drafting

Use it for first versions and rewriting. Judgement stays with people. Nothing about your company goes into anything.

What it costsYou get the smallest available gain, and your people fill the gap themselves in personal accounts, which is the same exposure with none of the oversight. Fine as a deliberate position for a year, expensive as a permanent one.

B

Feed it everything you have

Point it at the drive, the wiki, the handbook, the old decks. Let it read the company and answer from that.

What it costsYour stated process goes in, so the answers get more confident and stay wrong in the same way. Plus you have now put a lot of material somewhere, and the question of who may read what arrives later and louder than you expect.

C

Write down how the work is really done

Capture the demonstrated process. Who decides what, which exceptions trigger which action, what good looks like here and who says so.

What it costsReal time from people who have none, and somebody has to settle the disagreements that surface once you try to write down what good means. That conversation is the reason this route gets postponed, and it is also the entire value of it.

Which one is your company on today? Pick the honest answer, not the one in the plan.

All four readings, if you would rather not click:

A. Check what your people are doing in their own accounts before you decide this is working. That is usually where route B is already happening without you.
B. Take one answer it gave you last week and ask the person who owns that process whether it is how they actually work. The gap between those two is your project.
C. The trap here is writing the version that sounds good. Get it from watching the work rather than from asking people to describe it, because those produce different documents.
All three. Normal, and worth naming out loud. Different teams have chosen differently and nobody decided that. Pick one for the next two quarters and say so.

The mechanism

Three reasons your context stays out of the system.

The valuable part was never written down, because it never had to be

How you decide whether a client is worth the trouble. What makes an estimate optimistic. Which customer gets a phone call instead of an email. All of that is carried by people who learned it by being there, and it works fine that way until you need to hand it to something that was not there.

What is written down is the version for outsiders

Handbooks, process documents, the onboarding deck. Those describe the company as it would like to be seen, which is a reasonable thing for them to do. Point a model at that material and you get advice built on the official story, which is precisely the story that has never governed how anything actually gets done.

Every company has two processes. The one in the handbook, and the one that survives contact with a customer on a Friday afternoon. Paul Musters

Writing it down means settling arguments

The moment you try to describe what good looks like here, two people discover they have disagreed about it for four years and have been working around each other successfully the whole time. That is uncomfortable, it takes a meeting nobody scheduled, and it is the actual work. Most context projects stall exactly there, and then get reported as a tooling problem.

The company-level version of the same gap

Anonymous, six questions, three minutes. Your level, the cost of it, and what the next level gives you.

Do the Culture Level scan

Level by level

Route C has a name in our model. It is Level 3.

01Campfire60%

02Wild West25%

03Blueprint10%Context gets written down here

04Engine4%

05Ecosystem1%

Share of companies per level. The five AI Culture Levels we use with clients, from everyone doing their own thing to AI being part of how the company runs.2

In the five AI Culture Levels we use with clients, Level 3 is the level where the way of working gets documented and the good method becomes the normal method. That is the same act as putting your context into a system, arrived at from a different direction. About one in ten organisations is there.

Which is worth knowing before you plan this as a technology project. The barrier sits somewhere else than the retrieval setup or the vendor. Somebody has to sit with three people and write down what they actually do, and that has to be worth their week.

Where a measurement fits in

An Operating Profile in use. Personality type and AI level in one profile, with the agents that fit it.

There is a smaller version of context that gets skipped in the rush to the company-wide one. How this particular person works: what they need to see before they can decide, how they like to be given a draft, where they are strong and where they routinely hand things off. That is context too, it changes the quality of the answer immediately, and it does not require a committee.

So the profile carries two things: how somebody thinks and works, and their level with AI. The agents get built on that, so the first useful context in the system is the person in front of it. The company-wide version is a bigger project and it goes better once people have seen what the small one does.

About emaho

emaho measures one Operating Profile per person: personality type and AI level in a single profile. On that we build a personal set of AI agents that fit how that person works, inside the tools they already use. Fifteen minutes to complete, first profile free, built for companies between 20 and 500 people.

See your own profile, free Find your AI Culture Level

Fifteen minutes per person. No credit card, no strings.

The answer

Begin on a Friday afternoon.

Pick one workflow that matters and watch it happen twice. Watch, rather than ask, because people describe the official version when you ask and they do the real one when you are just sitting there. Write down the order, the exceptions and the moments somebody made a call.

Then get the judgement calls out of the heads that hold them. Two hours with the person everybody checks with, and a list of the twenty questions they get asked most and how they answer them. That list is worth more than any handbook you own and it takes an afternoon to make.

Put that in front of the model before you put anything else in. A short, accurate description of how the work really goes beats a drive full of documents about how it is supposed to go, and it comes with a fraction of the questions about who may read what.

And keep it alive. Context written once is out of date within a year, in the specific way that makes people stop trusting the system. One person, one hour a quarter, checking whether it still matches what happens.

Curious what your assistant said?

Run the two-minute test at the top and send me the answer you got. I am collecting these, and I will tell you what I would have expected and what it says about where your context lives.

Message me on WhatsApp

If you do nothing

Advice that is right in general and wrong in this building.

The visible cost is decisions made on generic advice. Somebody asks, gets a reasonable answer, and acts on it. It is defensible in a meeting and it fits any company, which means it does nothing about the specific thing that makes your business work or the specific thing that is wrong with it.

Then there is the erosion of trust in the tool. People try it on something that matters, get an answer that is confident and beside the point, and quietly go back to doing it themselves. After three of those you have a licence nobody uses and a leadership team wondering where the productivity went.

And the one that compounds. If the advantage now sits in whose context is written down, then every quarter you do not write it down is a quarter somebody else does. That gap is not visible from inside, because from inside everything works the way it always did.

Asked and answered

The context gap, asked and answered

Why is AI advice always so generic?

Because the model knows your industry and not your company. It has read everything ever written about how businesses like yours should work, and nothing about how yours actually does. Harvard Business Review put it plainly in February 2026: once every company has the same models and tools, organisational context is what separates them.

What is organisational context in AI?

It is how the work really gets done: the order people involve each other in, which exceptions make somebody pick up the phone, what counts as good enough here and who decides that. Rohan Narayana Murty describes it in HBR as demonstrated execution, the workflows teams actually follow, which is different from the process written in a handbook.

How do I test whether my company's context is in the system?

Ask your assistant something only your company could answer, such as how you decide whether to take on a client who is a bad fit but pays well. Then ask whether that answer could have been written for any company in your sector. If it could, your context is not in there, and it takes two minutes to find out.

How do you find out whether your company's way of working is written down anywhere?

Try to hand one workflow to somebody new and see what you can give them. emaho reads organisations on five AI Culture Levels, and writing down how the work is really done is Level 3, where about one in ten sits. That is the same act as getting your context into a system, arrived at from a different direction.

Should we just upload all our documents to the AI?

It is a real option and it has a specific failure mode. Handbooks and process documents describe the company as it would like to be seen, so feeding them in makes the answers more confident while leaving them wrong in the same direction. It also creates a question about who may read what, which arrives later and louder than most people expect.

What is the difference between stated and demonstrated process?

The stated process is the one in the document. The demonstrated process is what survives contact with a customer on a Friday afternoon: the shortcuts, the exceptions, the person everybody checks with before sending. Almost all of a company's real advantage lives in the second one, and almost none of it is written down.

How do I capture how my company really works?

Watch one workflow happen twice instead of asking people to describe it, because those produce different documents. Then spend two hours with the person everybody checks with and write down the twenty questions they get asked most and how they answer them. That list beats any handbook and takes an afternoon.

How do you know whose way of working is worth writing down first?

Start with the people everybody else checks with, and those are not always the senior ones. emaho measures one Operating Profile per person, personality type and AI level in a single profile, which shows who is carrying method that exists nowhere but in their head.

Why do context projects stall?

Usually at the moment somebody tries to write down what good looks like here, and two colleagues discover they have quietly disagreed about it for years. That conversation is the actual work and nobody scheduled it. It then gets reported upward as a tooling problem, which is easier to explain and impossible to fix.

Is context a technology problem or an organisation problem?

Organisation, with a technology component. The retrieval setup is the easy half. The hard half is that somebody has to sit with three people and write down what they actually do, and that this has to be worth their week. Buying a better platform does not move that.

Can our AI advice ever be a competitive advantage?

That is the argument HBR makes. When everyone can buy the same models, the difference comes from what only you know: how you decide, what your real risk appetite is, what good looks like in your teams. The companies that write that down get answers their competitors cannot buy.

Where can I read about the other AI challenges around AI and leadership?

This is one of 25 AI challenges emaho documents. The context gap sits next to checking that quietly stops and the gap between what leaders think and what people do.

Where should we start if we have limited time?

With one person rather than the whole company. How a single colleague works, what they need to see before they decide, where they hand things off. That changes the quality of the answers immediately, needs no committee, and it shows people what the larger version would be worth.

How often does organisational context need updating?

Once a quarter is enough, and it needs an owner. Context written once goes out of date within a year in the particular way that makes people stop trusting the system: mostly right, wrong in one place, and you never know which. One person, one hour a quarter, checking it against what actually happens.

How does this relate to AI maturity levels?

Writing down how the work is really done is Level 3 behaviour in the five emaho AI Culture Levels, where the good method becomes the normal method. About one in ten organisations is there. At Level 2 the method lives in individual heads, which is exactly why the advice you get back describes a generic company.

What connects to it

Context has neighbours.

These three come up as soon as you start writing down how the work really goes.

11

Coordination cost

A shared way of working is also what makes review fast. Same document, two different problems solved.

Read this one 07

Shadow AI

Route A has a side effect. When the sanctioned tool gives generic answers, people go and get better ones somewhere you cannot see.

Read this one In the full set 14

Everyone brings their own context

The same gap one level down. Four people, four personal setups, four versions of what the company promised.

Read this one

All 25 AI challenges, and what to do about each

Twenty-five things that break inside a company once people start using AI, each with the research behind it and the level where it starts to bite. This one starts to bite at Wild West, and 8 of the others start there too.

Open the full set

What now

Start with the context you can capture this week.

Writing down how the whole company works is a project. Writing down how one person works takes fifteen minutes, and it changes the quality of the answers immediately.

Write down how one person worksFifteen minutes, shorter than a workshop, and it changes the next answer immediately.
Hand the assistant that pageGeneric advice becomes specific the moment it knows the shape of the company asking.
Add the next person next weekA company context that grows one person at a time stays current. A company-wide document does not.

Start with your own profile, free Ask me anything first Plan a 30-minute call

First profile free · no credit card · built for companies of 20 to 500 · you decide what your team gets to see

Rather start at company level? The Culture Level scan is six questions, three minutes, and asks nothing of you.

Paul MustersFounder of emaho

Paul Musters

Fifteen years of leadership and team development in Dutch scale-ups. That practice now sits in software: one Operating Profile per person, with agents that actually fit. He writes these pages from what he runs into with clients, not from a research summary.

LinkedIn · paul@emaho.world · WhatsApp

Sources and numbers used on this page

Rohan Narayana Murty, When Every Company Can Use the Same AI Models, Context Becomes a Competitive Advantage, Harvard Business Review, February 2026. Source of the argument that once every company has the same models, tools and vendor ecosystem, organisational context becomes the differentiator, and of the line that context is demonstrated execution: the workflows teams actually follow across systems.
emaho AI Culture Levels. Level 3, where the way of working is documented and the good method becomes the normal method, holds about one in ten organisations. Calibrated against BCG 2025 and McKinsey 2025.

The three answers in the recognition block are the kind of advice these tools give, written out in full. They are composites rather than transcripts.