Guide
Most boards we meet are not behind on AI. They are lost in it, and the two are not the same thing. Here is the order we use with clients, from the fifteen or so AI programmes we have delivered.
By Christopher Fagan, Managing Partner, PEF Consulting Partners
Every week brings a new tool, a new warning and a new competitor claiming to have transformed itself. Inside most organisations the picture is simpler: a handful of people using AI quietly for their own work, a vendor or two circling, and a board that knows it should do something and has no idea what. That is where nearly every client we have worked with started, whatever their size or sector. The ones that got value did not start with a big strategy. They started with an honest look at where they were.
Pick three places in the business where people spend hours on repetitive judgement: reading documents and pulling out the same facts, answering the same customer questions, reconciling the same numbers, drafting the same reports. Measure the time and the error rate for a fortnight. That is your business case, in your own numbers, before anyone has shown you a demo. If no process clears the bar, you have learned something cheap: the right move for now is to wait.
AI amplifies whatever you feed it. If the data behind a process lives in six spreadsheets with three owners, the tool will automate the confusion faster than people could. The readiness question is not whether you have a data lake. It is whether, for the one process you have chosen, you know where the data lives, who owns it and whether you trust it. That question takes days to answer and it decides whether a pilot has any chance.
Governance sounds like the thing you do later. It is the thing that lets you move at all. Write down what AI may do on its own and what needs a person to approve, which data may never leave the business, who signs off a new use, and how you would explain an AI assisted decision to a customer, an auditor or a regulator. One page that people follow beats forty pages nobody reads. The rules in the UK and Europe are tightening, and a business with a clear policy will find the tightening easy; one without will find it expensive.
One process, one named owner, one success measure agreed up front, a few weeks, and a date on which you will stop if it is not working. Pilots fail when they have no owner, no measure and no end. Keep it small enough that stopping costs you nothing you would miss, and honest enough that a result either way is useful.
This is where most AI projects stall, and it is the part nobody puts in the brochure. A pilot that lives in a browser tab is not a business change. Value arrives when the AI step is wired into the finance system, the CRM, the billing platform or the case management tool that people actually work in, with the controls from your policy built into the flow. That is systems delivery, and it is where nineteen years of implementing finance and operational platforms turns out to matter more than any model.
Be careful of advice from anyone with a tool to sell. The useful adviser will tell you when to wait, will test vendor claims against your numbers, and has delivered the integration step, not just the slide deck. That is how we work: a short, independent AI readiness review that ends in a roadmap you can act on, and full delivery when you are ready for it. If you want a quick, private sense of where you stand first, our free AI readiness check takes two minutes and sends nothing anywhere.
Less than most people fear. The expensive mistakes are the tools bought before the problem was understood. A readiness review is a fixed scope, fixed fee piece of work that takes a few weeks, and a first pilot should be sized so that stopping it costs you nothing you would miss.
No. You need someone who owns the decision, data you can trust for the use case you have chosen, and an adviser who can test what vendors tell you. Specialist skills come once you know what you are building.
It rules out starting with the tool. It does not rule out starting. Most organisations have one or two processes where the data is good enough for a first pilot, and fixing the rest becomes part of the roadmap rather than a reason to wait forever.
Data leaving the business through a tool nobody vetted, decisions nobody can explain to a customer or regulator, and money spent on a pilot that was never going to be integrated. All three are governance problems, and all three are avoidable with a one page policy that people actually follow.