Most companies don't need an AI strategy. They need someone who can say no.

Cognizant put out research last week called Plug-and-Play AI Is a Myth. The title is doing a lot of work, but the finding underneath it is more interesting: the biggest blockers to AI adoption aren't technical. They're strategic.

No clear plan. No way to measure ROI. Regulatory fear. I'd add one more: no one in the building who can tell the difference between a useful AI project and a waste of six months.

The gap between buying AI and getting value from it

ModelOp's latest benchmark has a stat that keeps coming back to me. 67% of enterprises now have between 101 and 250 proposed AI use cases. But 94% have fewer than 25 in production.

That's a 10% conversion rate from idea to production. If your sales pipeline looked like that, you'd fire someone.

Goldman Sachs looked at the same question from the other end. Only 10% of S&P 500 teams have actually measured AI's impact. 70% mention it on earnings calls. The gap between "talking about AI" and "knowing what it's doing" is enormous.

And Workday found that 40% of time saved by AI tools gets eaten by correcting errors, rewriting outputs, and checking results. The tools are generating work, not removing it.

The bottleneck isn't the models. It's specs, taste, verification, and change management. The stuff that sits between "we should use AI" and "this is actually working." I've written more about why generic AI fails inside enterprises in Plug-and-Play AI Is a Myth.

What a fractional AI CTO actually does

A traditional CTO manages infrastructure, security, and engineering. An AI CTO does something different.

They look at your business, your data, your team, and your regulatory environment — and they tell you which two or three AI projects are worth doing. More importantly, they tell you which twenty aren't.

Most mid-market companies — say £10M to £200M in revenue — can't justify a full-time hire for this. Nor should they. The role is intensive during the build-out phase, but it doesn't need 52 weeks a year once the foundations are in.

A fractional AI CTO typically works two to four days a month across a 6-to-12-month engagement. You get someone making architecture decisions, evaluating vendors, and killing bad ideas early. Without the permanent headcount.

Why the Big 4 model doesn't work here

Deloitte just launched GenW.AI. McKinsey, BCG, and Accenture all have AI offerings. But these engagements are built for organisations with a billion in revenue and the IT departments to match.

Mid-market companies often don't have a CTO at all, let alone one with AI expertise. When a Big 4 firm delivers a 200-page strategy deck, there's frequently nobody in the building equipped to execute it.

The fractional model inverts this. Instead of a document, you get a practitioner who sits inside your decision-making. Someone who'll be in the room when the vendor demo happens and can tell you what was real and what was theatre.

The problem with "AI advisory" on LinkedIn

The role has become fashionable. LinkedIn is full of people offering fractional AI CTO services who were doing social media marketing 18 months ago.

Here's what I'd look for if I were hiring one.

First: have they actually built anything? Not advised on it. Built it. Written code, managed data pipelines, shipped something to production. If their experience is slide decks, they'll struggle when the work gets real.

They also need to know your industry. AI in financial services looks nothing like AI in healthcare. Regulatory constraints, data structures, risk tolerance — they vary dramatically. Someone who's worked inside your sector, not just consulted to it, understands things a generalist won't.

The most underrated quality is the ability to say no. A good fractional CTO will kill more ideas than they greenlight. You're paying for judgement, not enthusiasm.

And they need to be current. GPT-5.5 dropped this month. DeepSeek V4 launched with a trillion parameters. Gemini 3.1 Flash-Lite is outperforming models that cost ten times more to run. Qwen 3.5 runs on a phone. If they can't tell you the practical differences between these — and which ones matter for your use case — they're not current enough.

How we run it

At Exponential Partners we work in three phases, and the fractional CTO model maps across all of them.

It starts with training. We run one-to-one sessions and small group workshops — tailored, not off-the-shelf. Before anything else, we find out where people actually are: what tools they're using, where the friction is, what they've tried that didn't land.

The most common thing I hear before a session is "the models just aren't that good." In most cases that reflects a gap in tool choice, workflow design, or approach — not a genuine ceiling.

From there we move into strategy. We work function by function, workflow by workflow — not top-down with a pre-set answer. The goal is to build a view on where AI creates the most value for your specific business, which tools belong in regulated environments, and which don't. In financial services, that means getting into KYC, compliance workflows, onboarding, and internal knowledge work. The strategy runs alongside the training, not as a separate workstream.

Then comes transformation. We start with the opportunities that have the highest value and lowest risk, and deliver those first. These aren't short-term experiments — they're lasting changes to how teams work. If something needs to be purpose-built, we can take it from strategy through to delivery. We're building an asset management platform for a family office right now, so that capability sits in-house.

One managing director at a Tier 1 financial services firm put it this way after a session: they'd been using Microsoft Copilot for months and hadn't appreciated what was actually possible. After the session, their leadership team felt confident making informed choices on AI partners, vendors, and strategy.

The honest caveat

This model isn't right for everyone. If you're a 50-person company with no data infrastructure and no technical staff, a fractional CTO can't fix that alone. You need foundational work first — data hygiene, basic tooling, someone internal who can own the day-to-day.

And if you're already running AI projects successfully with clear ROI, you probably don't need one. The fractional model is for the companies stuck in the middle: ambitious enough to have 100 AI ideas, but without the technical leadership to figure out which three to actually build.

CrewAI's latest survey says 100% of enterprises plan to expand agentic AI usage in 2026. Whether that expansion produces results or joins the 90% of use cases that never reach production depends on one thing: whether someone with the right technical judgement is steering.

If you're not sure whether this model fits your situation, message me. I'll tell you honestly.