AI works in the pilot. Then it meets the business.

You've probably watched this happen, or you're about to. An AI tool lands beautifully in one team, faster answers, cleaner output, a genuine lift. Everyone's encouraged. Then it's scaled across the business and the magic thins out. The natural conclusion is that something went wrong with the technology, or the rollout, or the training.

Usually it's none of those. The tool met the business and the business wasn't ready for it in the way that actually matters.

That sentence is worth unpacking, because "AI readiness" has quietly come to mean only half of what readiness is.

The half everyone's working on

Ask most enterprises what their AI readiness work looks like and you'll hear a familiar list: data infrastructure, model selection, governance, security, integration, vendor evaluation, change management. A centre of excellence. A steering committee.

Call this supply-side readiness — getting your systems ready to deliver AI into the business. Can the stack run it? Is the data clean? Is it secure, governed, integrated? All of it necessary. All of it being done.

There's a second half, and it's almost entirely missing.

The half almost no one is

Call it demand-side readiness — whether the business itself is ready to be understood and acted on by AI well enough to get value from it.

The simplest way to hold the two apart is as two questions:

Supply side asks: can we run AI here?

Demand side asks: is there a coherent business here for AI to run on?

One quick clarification, because the word "demand" pulls people toward marketing. This has nothing to do with demand generation, pipeline, or adtech. It's literal, the demand the business places on AI: what you're asking it to understand and act on. Supply is what you put in. Demand is what the business asks of what you've put in.

A word for the demand-side condition and what it actually means

Here's a piece of new vocabulary, introduced slowly because it's worth getting right: structural commercial legibility.

Strip it back. Legible simply means readable. So this is about how readable your commercial system is and to three readers at once: your leadership, when they make strategic calls; your functions, day to day; and now AI, which can only act well on a business it can read clearly.

Here's what illegibility looks like in practice and you'll recognise it. Ask Sales, Marketing, and Customer Success to describe your customer. You'll often get three different answers: each true to the team that gave it, none reconciled with the others. The business can't be read cleanly, because it's quietly saying three things at once. That's an illegible commercial system and it's exactly the thing AI inherits.

One distinction prevents a lot of confusion here. People hear "legible" and reach for "coherent" but they're not the same, and the order matters. Coherence is whether the parts agree. Legibility is whether they can be read at all. You can't resolve the disagreement between those three customer definitions until you can first see them laid side by side. Legibility comes first; coherence is what it makes possible. Most transformation work skips straight to coherence and stalls because the legibility underneath was never there.

So the demand-side question, said more precisely: is the business legible enough for AI to read it and produce something coherent when it acts?

Why we only ever talk about the supply side

Three reasons and none of them is that it matters more.

It sells. The supply side has vendors, products, budget lines, a whole market built to keep the conversation there. No one sells "a more legible operating model" off a shelf.

It's safer to raise. Supply-side work points at the technology, which is external to most leaders. Demand-side work points at the business itself, its structure, its decision rights, its internal disagreements. Much harder to bring up, especially from the outside.

It's easier to show. You can put an architecture diagram on a board slide. "The business is more legible than it was" is harder to evidence, those metrics are still being built.

So the supply side does all the talking, because it's the easier conversation, not the more important one.

What it costs to do one without the other

Two things tend to follow.

The dashboards get sharper; the decisions don't. Put better tools on an illegible business and you get a crisper picture of the same confusion, those three customer definitions, now tracked in real time. More precision, same disagreement about what to actually do.

Then the agents inherit it. Ask an AI agent to act for the business, qualify a lead, price a deal and it absorbs whatever the business encodes. Three theories of the customer? It picks one, or blends them where no one can see. The agent isn't wrong; the business was never aligned. Only now the misalignment compounds at machine speed.

None of this is a case for slowing down on the supply side. It's a case for not doing it alone because doing it alone is what's produced the gap between AI spend and AI value that the McKinsey State of AI 2025 report finds in roughly half of organisations: flat or negative returns.

What the demand-side work actually is

It's structural, not cosmetic. Functions describing the business in compatible terms. Decision rights made explicit. Evidence that travels. Strategy and execution staying joined up as the business changes.

And underneath all of it, one thing the supply-side conversation rarely reaches: a clear, structured account of the business itself, not the data it holds, but the business the data is about. What you sell, to whom, why you win, how you make money, what your priorities are, how those become day-to-day decisions, where the trade-offs sit. Picture what you'd hand a senior hire on day one to bring them up to speed and now imagine it expressed clearly enough that an AI system could use it as the ground for everything it does.

That isn't a documentation job. You're not writing down the business you already have, because the one you already have is scattered across functions, kept in people's heads, and contradicted by its own systems. The work is to make it readable in the first place.

One line holds the whole idea:

We don't ask AI to figure out the business. We give it a structured understanding of the business.

The question to put next to the one you're already asking

You almost certainly have a supply-side readiness effort under way, stack, data, governance, deployment. Good. Necessary. Set one more question beside it:

What does our demand-side readiness look like?

If the honest answer is "we don't have any," that's not a failing almost no one does yet. But the gap is real, and it compounds quietly, until every pound spent on the supply side just buys a more expensive version of the business you already had.

The next era of B2B value won't go to the businesses with the best AI. It'll go to the ones legible enough for AI to compound.