← My Thinking

The five layers, and why the order is the whole point

Most AI projects in small businesses don't fail on the technology. They fail because they were built in the wrong order, and the layer that got skipped was load-bearing.

I score every business I audit against five layers. Not because five is a satisfying number, but because these are the five things that have to be true before AI does useful work in a company that doesn’t have a data team.

They build in sequence. Each one is only as useful as the one underneath it. That sounds like consultant framing until you watch what happens when someone skips a rung, which is most of the time.

Layer 1 — Context

The AI knows your business. Positioning, offers, pricing, who you serve, how you talk, what you’ve already decided.

This is the foundation and it is almost always missing. Not partially — entirely. The description of what the business does, for whom, at what price, lives in the founder’s head and nowhere else. There might be an old deck and a website, both out of date and contradicting each other.

Without this layer, everything above it produces generic output. That’s the whole explanation for why AI feels underwhelming to most founders. It isn’t the model. It’s that the model has never been told anything true about the business, so it writes about the industry instead.

Layer 2 — Identity

The AI produces work that’s on-brand. Voice, tone, visual standards, and what “good” means here.

In most founder-led businesses, the standard exists only as the founder’s corrections. There’s a logo and some habits. Ask who besides the founder can say whether a piece of work is on-brand, and the honest answer is nobody.

That’s the constraint that keeps the founder in the review loop forever. You cannot delegate a standard that has never been written down — not to a person, and certainly not to a model.

Layer 3 — Playbooks

The AI follows your process. Your core workflows, documented at outcome level, with the decision rules and handoffs included.

Here’s where it gets uncomfortable. Most businesses have SOPs that describe the official process rather than the real one. The real one includes three workarounds, a spreadsheet somebody maintains privately, and a step that only happens because one person remembers to do it.

Document the official process and you automate a fiction. The audit’s job at this layer is to map what actually happens, tribal knowledge and all.

Layer 4 — Intelligence

The AI reads your business. Data connected, numbers defined, patterns surfaced without you assembling them.

The test I use: how long would it take you to answer “which client was most profitable last quarter?” If the answer is a few hours of pulling from different tools, you’re at a 1. If you couldn’t answer it at all, you’re at 0, and you’re making decisions on instinct plus the bank balance.

Note that this sits above Playbooks, not below. You cannot meaningfully measure a process that runs differently every time. Businesses that start here end up with very clean reporting on a process nobody agrees on.

Layer 5 — Automate

The AI runs the work. Recurring tasks complete without the founder in the loop, and exceptions come to them.

This is the layer everybody wants to buy, and the one that produces the horror stories when bought first. Automation on top of a business the AI doesn’t understand, to a standard nobody has written down, following a process that only exists in someone’s head, generates work that is fast, confident and subtly wrong.

Which is exactly why nobody trusts it, and why it gets quietly abandoned around month three.

Why the order is the whole point

Context makes Identity possible: you cannot define a voice for a business you haven’t described. Identity makes Playbooks safe: a workflow can run perfectly and still produce something nobody wants to send. Playbooks make Intelligence meaningful: measuring an undefined process gives you numbers about noise. Intelligence tells you what’s worth automating. And automation is where the hours finally come back.

Skip a rung and you don’t get four-fifths of a system. You get something that works in the demo and falls over in the third week.

The good news in this is bigger than it sounds. If you score low across all five — and most founder-led businesses I audit land between 0 and 1 — there’s no wasted work to unpick and no wrong turn to reverse. The build order is unambiguous. You start at Context and you go up.

The businesses that struggle most aren’t the ones at zero. They’re the ones who bought Layer 5 two years ago and have been trying to make it work ever since.

Related: what an AI operating system is, and what it costs.

Where does your firm actually sit?

The AI maturity score rates you on all five layers using the same rubric as a paid Third Act assessment: the number, the rung that's stuck, and what the next rung up looks like.

Free5 minutesInstant result
Get your benchmark
Book a call