What should be answered before an AI pilot?
What the process is, exactly; where it actually breaks; and whether a competent person with a checklist would fix it. If a checklist would do, it beats a model on cost.
Blog
AI will not fix your process, it will speed it up
No. AI runs a broken process faster and at scale, and because it produces plausible output even when the process underneath is wrong, the problem is harder to see than it was with older software. Map the process first: who does what, in what order, and which steps exist only out of habit, because some can simply be removed. Then aim AI at the points where the process actually breaks and a competent person with a checklist would not cope.
AI will not fix your broken process. It will run it faster, at scale, with confidence.
That sentence is most of what a small or mid-sized organisation needs to know before signing anything this year, and it is the sentence the sales deck will never contain.
The oldest mistake, reprinted
I have been watching organisations buy technology for forty years, and the most reliable failure has never changed: automating a process nobody understood. The database that faithfully stored the chaos it was given. The workflow tool that hardcoded a workaround someone invented in 2019. The CRM that three teams use three different ways, which is to say, not a system at all but three opinions with a licence fee.
AI is this mistake's biggest opportunity in decades, because AI is better than any previous technology at hiding the problem. Old software failed loudly when the process underneath was wrong; you hit the edge cases and things visibly broke. AI does something worse. It smooths over the cracks and produces plausible output anyway. A broken process plus AI does not look broken. It looks fine, faster, right up until you audit the results.
If your quoting process produces inconsistent quotes, AI will produce inconsistent quotes in seconds instead of days. If nobody agrees on how enquiries should be qualified, the model will make up its own mind, fluently, differently each time. Speed is the only thing that changed, and speed multiplies whatever it touches. Errors included.
What the process work actually looks like
I worked with a community foundation that manages millions in charitable funds and needed a new system for the whole grant-making operation. The temptation was to start with software demos. We did not look at a single product until the process itself was mapped: who does what, in what order, what gets decided where, which steps existed for a reason and which existed out of habit.
Two things happened, and they happen every time. Several steps turned out to be unnecessary, so they were removed rather than automated, which is the cheapest fix in all of technology. And the requirements that emerged were so clear that choosing a system became almost boring. The tender was won on fit, not on demo theatre. Years later the system is still in use, because it fits how the organisation actually works rather than how a vendor imagined it might.
None of that was AI work, and that is the point. It is the work that decides whether any technology, AI included, will pay for itself or just accelerate the mess.
The three questions before the pilot
Before you pilot any AI tool, answer three questions in writing.
What is the process, exactly? If it takes more than a page to describe, or if two people describe it differently, that is your project, and it costs thinking rather than licences.
Where does it actually break? Automating the steps that already work is where most AI spend quietly dies. The value is at the failure points, and you cannot aim at them without naming them.
Would a competent person with a checklist fix this? Sometimes yes, and then a checklist costing nothing beats a model costing plenty. When the honest answer is no, the volume is too high, the pattern-matching too complex, that is precisely where AI earns its keep. I recommend it often, on exactly those jobs.
The uncomfortable summary
The organisations getting real value from AI right now are, almost without exception, the ones that already understood their own processes. The technology rewards the discipline it did not create. Everyone else is buying speed and calling it progress.
Fix the process. Then, and only then, make it faster. The order is not optional, and no model on the market changes it.
Questions people ask
Which organisations are getting real value from AI?
Almost without exception, the ones that already understood their own processes. The technology rewards a discipline it did not create.
Not sure where to start?
Most clients begin with a conversation. No pitch, no hard sell.
Just a straightforward discussion about where you are and whether I can help.
Book a free 30-minute call