Key takeaways
- When attendees of a McKinsey alumni panel were asked where their organisations lag on AI, the top answers were operating model and adoption and change, not technology or data.
- A panellist cited McKinsey research that only about 6% of organisations see disproportionate EBIT from AI. Most treat it as a personal copilot.
- Value leaks through three gaps: from pilot to production, from a faster task to a redesigned workflow, and from time saved to money in the P&L.
- The advice that works: pick a handful of anchor use cases with high value and low risk, win visibly, and use the wins to fund the rest. Tool rollouts and proof-of-concept sprays rarely deliver.
- Track ROI inside the AI tool, with both activity and outcome measures, so you can see where adoption is failing.
Most organisations are using AI, but very few are transforming with it, and the reason is leadership rather than technology. That was the thread running through a McKinsey alumni panel I joined this month, Path to AI transformation, with a consulting chief AI officer, a chief AI and transformation officer at a global information company, and a moderator who is also a McKinsey alumnus. As a former McKinsey consultant who now runs a consultancy building AI systems, I recognised almost every point. These are the lessons worth sharing, with the speakers kept anonymous.
Key definition: A chief AI officer decides where an organisation applies AI, how the value is captured and who owns it. Adoption is whether people actually use the new way of working, not whether the tool is installed.
Where do organisations say they are struggling with AI?
The moderator ran a live poll with the six domains of McKinsey's Rewired playbook: strategy and use cases, technology, data, operating model, talent, and adoption and change. Attendees said they were strongest at strategy, solution design and finding use cases. They said they were weakest at operating model and ways of working, and at adoption and change management.
Even the moderator was surprised: "I was not expecting operating model and adoption and change to be the two we're struggling the most at." The thinking is in place. The doing is not.
Why do most companies use AI without transforming the business?
Because they treat it as a productivity tool for individuals. One panellist cited McKinsey research, presented at a recent investor conference, that only about 6% of organisations are seeing disproportionate EBIT from AI. McKinsey publishes its wider findings in its annual State of AI survey. The rest give employees chatbots and copilots, which is necessary but leaves the major workflows of the business untouched. In the panellist's words: "We're using the technology, but we're not transforming the business."
Where does the value leak between a pilot and the P&L?
The chief AI and transformation officer named three gaps, and they compound:
- Pilot to production. It is easy to demo something impressive. It is hard to make it work with real data, legacy systems, guardrails and risk controls.
- Deploying technology versus changing how the business runs. Making a task 30 to 50% faster only moves the bottleneck unless the whole workflow is redesigned.
- Productivity versus business outcomes. Employees saving 20 to 30% of their time does not reach the P&L by itself. That capacity has to be deliberately redeployed: into revenue, customer experience, or growing without adding headcount.
Her conclusion: "AI transformation is fundamentally a leadership challenge."
What does a chief AI officer actually do?
Less technology than you might think. Both panellists described a job that is mostly priorities and people:
| Responsibility | What it means in practice |
|---|---|
| Use cases | Find where economic value sits for this business: supply chain, yield and throughput in manufacturing; pricing, promotion and fill rates in retail |
| What to stop | Free up time and budget for the few things that matter |
| Build or buy | Build very little from scratch; buy where the problem is already solved |
| Talent | The scarcest skill is not technologists but people who can redesign a business process |
| Adoption | Assign owners, set governance, bring sceptics along |
| Technology | Understand model choice, accuracy, reliability and cost well enough to make the calls |
One panellist described a "bimodal" organisation: people who lean into every new tool, and people who worry about their relevance. Managing the space between the two is the job. The other put it this way: "AI is moving at the speed of technology; organisations move at the speed of humans." She joked that her assistant had turned the whole organisation chart into meeting invites, many of them with "alignment" in the title.
How fast should you go, and how broad?
Neither as broad as possible nor as slow as possible. The consulting chief AI officer described it as a portfolio:
- Find a handful of anchor use cases that are high value, lower risk and faster to implement.
- Organise the rest of the programme around them.
- Get visible wins early, and use them to fund and energise everything else.
What he has seen fail, every time: rolling tools out broadly to "see what happens", spraying proofs of concept, running innovation exercises without a business target, and technology-first plans such as "fix all our data first, then do AI". Data still matters. The moderator described a client spending almost all its transformation time turning messy data into usable data products. But data work has to serve a named business outcome.
On where value shows up first, the panel's view was that customer service and customer engagement move first because they are easier to imagine, and supply chain (forecasting, demand planning, scheduling) is the next big wave.
How do you prove ROI on an AI programme?
Track several measures for every use case, some about activity and some about outcomes, and track them inside the AI tool itself. The panellist's example was a demand forecasting rollout at an industrial company. The tool showed that an entire region in Asia was rejecting every AI forecast. That visibility started the right conversation: was it the forecasts, the process, the technology or politics? Failures happen in technology, process and people, and where they meet, so granular tracking is how you find out which one is breaking.
Why do people resist, and what do you do about it?
The chief AI and transformation officer described three situations, each needing a different response:
| They say | What is really going on | What helps |
|---|---|---|
| "Why change at all?" | They do not believe the case for change | Bring the data and the facts |
| "Right goal, wrong answer" | They agree on the objective, not the solution | Work through the options together, in small groups |
| "Fine, but not with my budget" | They agree on both, but fear losing resources | One-to-ones, and be clear about what stops and what they gain |
She credited a McKinsey habit for this, the obligation to dissent: the courage to say that a shiny pilot will not change productivity, or is not ready for production. The other habits the panel carried from McKinsey were answer first, explaining technical work to boards and sales leaders in terms of outcomes, and structured thinking.
What does this mean for search and AI visibility?
One audience question was about GEO, optimising to be cited by AI chatbots rather than ranked by Google. The panel's view: people increasingly ask a chatbot instead of searching, chatbots give a specific recommendation with reasons rather than a list of links, and hallucinations mean credible, checkable sources matter more than ever. If your company is not a source that an AI model trusts and can cite, it is increasingly invisible.
What should you do next?
The best advice of the session was practical. Do not wait for an AI title. Take ownership of one AI-shaped business outcome in the job you already have: pick a problem that matters, form a small cross-functional team, and take it from hypothesis to production. Learn what broke and what surprised you. The other panellist's advice was simpler: stay curious, and try the new models yourself, because the technology changes in weeks, not years.
The moderator closed with a line borrowed from James Carville: it's all leadership, stupid.
Frequently asked questions
Do smaller companies need a chief AI officer?
Usually not as a full-time role, but someone has to own the same decisions: which few use cases to fund, what to stop, and who is accountable for adoption. In a mid-sized company that is often the CEO or COO, with outside help on the technology.
Where should a company start with AI?
With one or two anchor use cases tied to a number the business already tracks, such as response time, forecast accuracy or revenue per head. Avoid rolling out tools broadly with no target, or fixing all your data before you start.
How long until AI shows up in the P&L?
Only after the workflow has been redesigned and the saved time has been redeployed. A faster task alone moves the bottleneck. Measure adoption and outcomes from day one so you can see where value is leaking.
Is the 6% figure from published research?
It was cited on the panel from McKinsey research. McKinsey's annual State of AI survey is the public source for its findings on AI high performers.
About the author: Tengo Meskhi is CEO of Conceptdigital. He was previously at McKinsey and QuantumBlack, and Deputy CIO at Bank of Georgia.
If you want help choosing your anchor use cases and getting the first one into production, we are happy to talk it through. Book a 30-minute call.