Ontology. A word you should get familiar with.

Every AI programme has to answer one question before it spends a pound: what are we pointing this at? The strong answer is a number — one metric that matters, the number that would change the business if it moved. But a number on its own is a target, not a plan. What turns it into a plan is a map: you put the metric at the centre and draw everything that moves it. That map is what we mean by an ontology of your business. The word is heavier than the idea.

It's easier to see than to define, so here is one being built.

A mid-sized services firm picks a single metric: the time from a qualified enquiry to a signed contract. Today that runs at around forty days, and it caps how much new revenue the business can take on. So it goes at the centre of the map.

What moves it? Four things, mostly: how quickly a tailored proposal gets drafted, how many rounds of clarification the client needs, how long legal review takes, how long pricing sits waiting for sign-off. Each of those has its own inputs. Proposal drafting depends on someone assembling a bespoke document from past proposals and the specifics of this brief. Clarification rounds depend on how well the first scoping call captured what the client actually needed. Legal time depends on how standardised the contract is. Pricing sign-off depends on an approval chain with three names on it.

Draw that out and something becomes visible that a list of AI use cases never shows you. The proposal-drafting node is intelligence work — the research, synthesis and drafting that a capable model can now do in an afternoon under a person's direction, rather than over two days. That's where the technology clearly belongs. But the map also shows that automating it, on its own, barely moves the forty days — because the contract still waits in legal, and pricing still waits on three signatures. The drafting was never the whole constraint.

That is the point of the map. It keeps focus singular — one metric, not a scorecard — while keeping the picture whole. Without it, you get one of two failure modes. Start from tools and you get sprawl: a pilot in every function, none of them tied to a number anyone cares about. Start from a single task and you get a point solution: faster proposals, bolted onto a process that still takes forty days. The ontology is what sits between those two — a single point of focus, drawn out to everything that feeds it.

It also shows why moving the metric is rarely a technology problem alone. Read the map again and the levers fall across five layers: vision (choosing the metric in the first place), governance (who signs off pricing, how standard the contract is), technology (the drafting), culture (whether the team trusts a model-drafted proposal enough to send it), and process (how scoping is run). Pull only the technology lever and the number twitches, then settles back. That is the single most common reason AI pilots disappoint: the effort touched one layer, and the other four held the old shape in place.

None of this needs new AI knowledge to start. It needs the map. If you are commissioning AI work for your team, the ontology is the first thing to ask for — before any tool, any platform, any pilot. The metric at the root, the drivers drawn out, the layers named. A provider who opens with tools has skipped the only step that tells you whether the tools will matter.

Start from the metric. Draw what moves it. The tools come last, and by then they're obvious.