Is it better to scale AI across your business or go deep in one function?

Go broad first. In aibl’s survey of 755 UK mid-market leaders, spreading AI across several functions returns 57%, against 49% for going deep in just one. Only 14% of UK businesses have reached real scale, running AI across three or more functions with a proven return.

Adoption is common. Scale is rare.

Most UK businesses have started with AI somewhere. Getting a return from it everywhere is a different problem entirely. Starting is now the easy part, and most firms have done it. Turning that into a return that compounds across the business, rather than sitting in one team, is the hard part.

In aibl’s survey of 755 UK mid-market leaders, just 14% have AI running in three or more functions with a return. That’s the definition of scale used throughout this report. Everyone else is still working it out.

The reality is that starting with AI is now the easy part. Most firms have done it. Scaling it, so the return compounds across the business rather than sitting in one team, is the hard part.

The four squares: where your business actually sits

Map your business on two axes: how many functions use AI, and whether any has gone all the way to scaled. Four groups fall out, each with a very different return, from 21% at the bottom to 62% at the top. Most UK businesses sit in the weakest of the four.

Stalled firms use AI in one or two places with nothing scaled. They report 21% measurable ROI, the weakest position on the map. Siloed firms go deep in a single function but stay narrow. They post 49%.

Scattered firms spread AI across three or more functions without deep scale anywhere, and they report 57%. Scaled firms combine breadth and depth and reach 62%, rising to 72% once two or more functions are genuinely scaled.

Most UK businesses sit in the Stalled square. A little AI in a function or two, nothing pushed further. Moving out along either axis roughly doubles the return.

Why breadth beats depth

Broad-but-shallow beats narrow-but-deep, and that’s the finding that cuts against instinct. In aibl’s survey of 755 UK mid-market leaders, Scattered firms that spread AI across functions report 57% measurable ROI, against 49% for Siloed firms that went deep in just one place.

Depth confined to a single part of the business doesn’t travel. A brilliant AI rollout in finance, with nothing happening elsewhere, still leaves the rest of the business running the old way. The return stays local.

Breadth spreads the habit of using AI, the governance that goes with it, and the confidence to keep going. That’s worth more, on average, than mastering one function in isolation. But it isn’t the final word. Firms that combine both, breadth across the business and real depth in at least two functions, clear 72%. Breadth gets you moving. Depth, once you’ve got the breadth, is what pushes the number higher.

Depth still pays, once you have it

Depth on its own, without breadth, underperforms. But once the breadth is there, going deeper within a function pays sharply, and the gap between shallow and deep use is one of the largest in the whole dataset. This is the second half of the sequence, not the first.

Basic tools, retrieval systems and single-task agents, return 29% measurable ROI. Agents running a full workflow return 56%. Agents working across departments return 83%. Each step deeper in agentic capability roughly doubles the payoff of the last.

So the sequence that works is breadth first, then depth. Get AI genuinely used in several functions. Then push the strongest ones further, towards agents that handle a full workflow rather than a single task. Trying to do the reverse, going deep in one place before anywhere else has started, is the Siloed trap.

This is where the AI adoption roadmap is worth reading alongside this piece. It sets out the stages; this article explains why breadth needs to come before depth inside them.

Why the gap isn’t really about adoption

None of this is a story about UK businesses falling behind on AI. Adoption is genuinely widespread, and national data backs this up. UK business AI use has nearly tripled since 2023, but that usage has widened steadily without ever deepening into real scale.

The honest read is that starting is common and getting a return from scale is not. That separates a Scattered business from a Scaled one. It’s a governance and capability question more than a tooling one. Our AI business strategy guide covers how to sequence that work once you know which square you’re in.

What to do this quarter

Count the functions where AI is in real daily use. Then count how many you’d honestly call scaled, meaning it runs with a measurable return rather than just switched on. Those two numbers place you in one of the four squares, and they tell you whether your next move is breadth or depth.

That tells you which square you’re in. If you’re Stalled, the next move is breadth. Get a second and third function using AI properly before you go deeper in the first. If you’re Siloed, don’t add more depth yet. Spread what’s working before you push it further.

Breadth first, depth second, both together last.

Frequently asked questions

Read the full research

This is one finding from State of UK AI Adoption 2026, aibl’s benchmark of 755 UK mid-market leaders, in partnership with Executive Summary. The full report maps every function against the adoption stages, with the moves that take a business from Stalled to Scaled.

Read the full State of UK AI Adoption 2026 report →