For the COO Guide

AI transformation for COOs: a practical operating plan

AI arrived through technology functions, and the consequences land in operations, which is why AI transformation belongs with the person who owns the operating model. This is what you own, how to plan the first 90 days, what it really costs and how to move from pilots to something that runs on a Monday morning.

Let's talk
Ownership01

What the COO owns in AI transformation

You own the processes AI is supposed to improve, the people whose work it changes and the operational risk when it goes wrong, so if anyone should hold the whole picture it is you. The evidence keeps pointing the same way: BCG found that only 26 per cent of firms have embedded AI into a broader transformation while high performers are around seven times more likely to have redesigned workflows, and workflow redesign is operations work. We set out why the accountability defaults to you, and why ownership so rarely does, in AI accountability lands on the COO.

90-day plan02

A 90-day AI plan for the COO

A durable first quarter is narrow on purpose: it builds one thing well rather than surveying everything AI might do. Each phase has an owner, a deliverable and a decision, so progress is visible to the board.

  • Days 0 to 30: map and choose

    Inventory where AI already touches your processes, then choose one where a wrong answer is recoverable and a person still owns the decision. Deliverable: a shortlist with a costed business case. Decision: which single use case to build.

  • Days 31 to 60: build the curated set and the controls

    Stand up the curated data the use case needs rather than waiting for the whole estate to be clean, and give it an owner, a decision boundary and a measurement baseline. Deliverable: a working capability in test with its governance defined. Decision: go or no-go to production.

  • Days 61 to 90: run, measure, sequence the next

    Put it live behind human ownership, measure the outcome against the baseline, and let what you learn name the next process to fix. Deliverable: a measured result and a sequenced roadmap. Decision: where the second use case goes.

The data question that sits under all three phases is set out in fix the data for everyone, not just for AI.

Operating model03

Designing the AI operating model

The return comes from redesigning how the work is done, not from the tool, and the peer-reviewed evidence is blunt about the boundary: in Organization Science, people working inside the tasks AI handles well were faster and higher quality, while outside that boundary AI users were 19 per cent less likely to produce a correct answer. So the operating model has to define, per process, where AI helps, where it stops, who checks and how an exception is handled. We cover the shape of that work in AI operating model design.

Cost04

TCO and adoption-cost control

The licence is the cheapest part. The model or token cost is typically 8 to 15 per cent of the three-year total, and the rest sits below the waterline in data readiness, integration, governance, change and a recurring run cost the business case rarely names. Scope it as an operating model change and you can control it; scope it as a software purchase and you will underestimate it by 40 to 60 per cent. The full picture, layer by layer, is in the real cost of AI for a COO, and a costed assessment is what our Breathe engagement produces.

Delivery05

Moving from pilots to production

Most AI stalls between an enthusiastic pilot and anything that runs in the business, because the pilot was never scoped with governance, measurement and an operating model in mind. The way through is to build one deployable capability with those built in from day one rather than bolted on afterwards, which is what our Flow engagement does, and to redirect the capacity it frees into work you can measure rather than letting it evaporate.

Governance06

Governance, accountability and decision boundaries

Governance at the scale of one use case is three things: a named owner, a defined response when the system is uncertain, and a record that someone weighed it. Get that right on one process and it becomes the pattern for the rest. In regulated firms it also satisfies the accountability the rules already assume, and it is far easier to establish for a curated capability than for an estate. We wrote about the supervision failure mode, where the model even flagged its own doubt and nobody acted, in the AI warned them twice.

Leadership07

When fractional AI leadership is useful

If AI has become your problem without ever becoming your remit, a fractional Chief AI Officer gives you senior leadership embedded in the team to hold strategy, governance and delivery together, without the cost of a full-time hire. It suits businesses of 100 to 2,000 people, especially in regulated sectors, and it is exactly what our Grow engagement provides. We compare it with the alternatives in what a fractional Chief AI Officer costs.

The guide08

More in the COO guide

Questions09

Questions COOs ask us

  • Who should own AI transformation?

    AI transformation is an operating model change, so it belongs with the person who owns the operating model, which in most businesses is the COO. In regulated firms the FCA already puts technology systems under SMF24, the Chief Operations function, so the accountability lands there whether or not anyone assigns it. The mistake is leaving strategy with a technology function while the consequences land in operations.

  • What should a COO's 90-day AI plan contain?

    An inventory of where AI already touches your processes, one high-value process chosen for redesign, a costed business case that includes data, integration, change and run costs, a named owner with a decision boundary for each use, and a measurement baseline so you can prove the return. It is a plan to build one thing well, not a survey of everything AI might do.

  • How should a COO measure AI value?

    Against a baseline you set before you start, on the process you own, in the numbers your board already tracks: cost per transaction, cycle time, error rate, capacity released. Time saved is not value until the process is redesigned to collect it, so measure the collected outcome rather than the tool usage.

  • How much does enterprise AI adoption really cost?

    The licence or token cost is typically 8 to 15 per cent of the three-year total. The rest sits below the waterline: data readiness, integration, governance, the productivity dip during change, and a recurring run cost the original business case rarely mentions. Scope it as an operating model change, not a software purchase, or you will underestimate by 40 to 60 per cent.

  • When is a fractional Chief AI Officer appropriate?

    When you need senior AI leadership embedded in the leadership team but do not need, or are not ready for, a full-time hire. It suits businesses with 100 to 2,000 people, especially in regulated sectors, that want strategy, governance and delivery held together by one accountable person a few days a week.

Every engagement is led by Dan Simms, our founder, a technology leader with 30 years in regulated sectors including CTO and CIO roles.

Contact 10

Talk through your AI plan

There's no pitch deck, just a conversation about the processes you own and where AI genuinely fits, so tell us what you're weighing up and we'll reply within one working day.

Get in touch

We'll respond within one working day. Or email hello@oxygenbubbles.com directly.

Or reach us another way

Oxi is our AI assistant. It answers questions about our services straight away, and anything it cannot answer comes to us.