On Tuesday 20 October I spent the day at the CMO & CAIO Inspired Summit at the Hilton London Syon Park. The speakers came from large organisations: retailers, banks, a publisher, a hotel group, travel and security firms. I went to hear what’s working, and I shared three takeaways on LinkedIn afterwards. This is the longer version, with more of what was said in the sessions.
Thanks again to Inspired Business Media for a day packed with good speakers.
Where the room was
One speaker, quoting McKinsey’s research this year, set the tone. 90% of CEOs believe AI will deliver a return next year, 92% of CFOs worry that it won’t, and only 6% of companies get 5% of their EBITDA from AI. What that 6% have in common, the speaker said, is that they re-engineered and automated their processes.
My impression was that these organisations are past the stage of giving everyone a tool, and past the stage of running lots of pilots. They’re now in “show me the return”. It’s budget season, and the CFO wants to know where the money went. The mood was sober, with plenty of lessons learned and portfolios cut back.
Workflow automation stood out as the one area where the return is easy to count, because you can measure the time saved.
1. Redesign how decisions are made
The strongest numbers of the day came from a travel business of about 400 people. A quote for a group tour took an experienced person three to five hours of judgement: the right product, the margin, the season and what they knew about the destination. They put together around 30,000 of those quotes a year.
They started with a pilot on 5% of enquiries. It cost £50k and brought in £2m of new revenue in six months, at a 25% margin. It now runs across the whole sales process on a pipeline of about 25 agents, and a quote takes 230 seconds.
The speaker was candid that the first approach didn’t scale. “I was automating workflows, when really I should be automating decisions.” A process, in their words, is a sequence of imperfect decisions. So the team stopped asking staff “what are your pain points?” and started asking “how do you decide?” The rules, know-how and history behind the best salespeople’s quotes went into what the speaker called a decision infrastructure, which the agents use to decide the way the experts do. Every step can be traced, and when an expert retires or leaves, what they knew stays in the business.
A chief data officer made the same point from the reporting side. They described the Monday meeting most of us know: 20 minutes arguing about whose number is right, 35 minutes of slides and 5 minutes deciding. Their answer was to retire the rituals (the monthly deck, the dashboard checked before every call, “final v7”), because the same old ritual with AI is just a faster ritual. The aim is a meeting that spends most of its hour deciding.
The tool rollouts were described flatly by comparison. One speaker likened rolling out Copilot and expecting benefits to rolling out Microsoft Office and expecting digital transformation.
2. Fewer, bigger bets, each tied to a line in the P&L
A digital leader from a large retail group described cutting about 43 AI initiatives to 4, deliberately bigger ones. Each had to move cost to serve, operating cost, working capital or revenue. The benefits came in ahead of the business case, but the costs were out by a factor of two or three, mostly in places nobody had priced: regulation, retraining, data storage, training and hiring. Asked what they’d do differently, the answer was data first, fewer initiatives and more education for colleagues.
Another speaker’s test was straightforward. If you can’t say which line of the P&L an initiative will improve, it’s a proof of concept, so call it one and decide up front what you want to learn from it.
A publisher added a lesson from a social commerce trial, where each team turned out to be asking the experiment a different question: profit, new audiences or top-line sales. The fix was to agree at the start what the test is for and how long it runs. And a panellist from a bank warned that proofs of concept left running without being properly secured become an easy target for attackers.
3. Foundations and people decide whether it works
Clean data and business context came up again and again. The chief data officer’s team rebuilt 25 years of data across eight countries before putting AI on top. Then they gave the AI the context a new starter would need: what each term means, plus the threads, tickets, meeting notes and decision documents that explain why things are the way they are.
Bad data with AI produces what they called fluent nonsense. A wildly wrong answer gets caught and a mildly wrong one doesn’t, and at least the old errors looked like errors. “New garbage with AI looks like a perfect sentence, with perfect grammar.”
Their safeguard was the most practical idea I heard all day. The automated weekly report deliberately carries ten absurd figures, and in the meeting the question is whether the data is wrong or whether nobody caught it. Either answer teaches them something.
On people, several speakers admitted they’d invested more in the technology than in their staff. One panel described a new bottleneck: when ten developers using AI produce the work of twenty, the person approving that work is overwhelmed. A security panel described the same pressure on approvals. When an agent is waiting on a person, they feel obliged to approve faster, or hand the approval to their own AI, and the check “degrades back to an automated process”.
Governance has to keep pace too. A speaker from local government described public bodies spending two years writing an AI strategy before testing a single use case, while staff got on and used their own tools. The security panel said the same about trying to block AI: people find their own route, often by emailing documents home to use the tool they wanted.
What this means for a smaller business
Most of the SMEs I work with are a stage or two behind the organisations in that room, and I think that’s an advantage. They can skip the expensive detours the speakers described, the tool rollouts and the pilot sprawl, and start where the return was found.
If I were choosing a first step, it would look like this:
- Pick one process that relies on judgement, such as pricing, quoting or deciding which work to take on, and tie it to a figure the owner already watches.
- Ask the people who do it well how they decide, and capture that before building anything.
- Cost it honestly, including the time your people will spend learning it and checking its output.
- Put light governance around it from the start, and test the checks, for example with a deliberate error.
That’s close to what I mean when I talk about building a brain for a business: capturing how your best people decide, so the business keeps that knowledge and can build on it.
One summit of large organisations is a strong signal rather than proof, but the direction was consistent across the day. If you’d like to talk through where that first process might sit in your business, I’m happy to have that conversation.