Working out your AI position without bringing in a consultant

This page gives away a chunk of what we normally get paid for. That is on purpose, and the reasoning is simple enough to state out loud: a company that can work this out on its own was never going to hire us, and a company that tries and gets stuck now knows precisely which part it got stuck on. Both outcomes are better than a vague brochure.

So here is the honest version of how to establish where you stand, using nothing but an afternoon and the people who already work for you.

Start from what you already do, not from what AI can do

The most common way these exercises go wrong is starting with a list of AI capabilities and looking for places to apply them. That produces a list of things that are technically possible and commercially pointless. Start at the other end: what work in this company is repetitive, high-volume, and currently done by someone whose time is expensive?

You do not need to know anything about AI to answer that. You need to know your own operation, which you do.

Two people in conversation. The answers live with the people doing the work, not in a report.

The six questions

Ask these of the people doing the work, not of the management team. The management team knows how the process is supposed to run. The people doing it know how it actually runs, and the difference between those two is where most of the opportunity hides.

One. What do you do more than ten times a week that feels almost identical each time? This finds volume. Anything below that frequency rarely repays the effort of automating it, whatever the tool costs.

Two. Where do you wait for information that exists somewhere in this building? This finds the retrieval problems, which are usually cheaper to solve than people expect and more annoying than management realises.

Three. What gets typed twice? Data moving from one system to another by hand is a signal that two systems do not talk, and often a better first project than anything involving a model at all.

Four. Which decisions do we make on gut feel that we have data about? Careful with this one. The honest answer is sometimes that the gut feel is better than the data, and that is worth knowing before you build something that overrules it.

Five. What did we get wrong last quarter that we could have seen coming? This finds the prediction problems worth having. If nothing comes to mind, that is a genuine and useful answer, and it means the prediction category is not where your value is.

Six. Who here already uses AI without being asked to? Almost every company has someone. Find out what they do, because they have already run your pilot for free and know where it breaks.

A pen over a clipboard. Write the answers down, because the value is in the gaps you find.

The six questions, and what each one is actually for

The right-hand column matters as much as the middle one. Each of these questions has a way of producing a confident wrong answer, and the exercise is only worth doing if you are willing to write down the uncomfortable version.

Score it honestly, not generously

For each opportunity you find, answer three things in writing: how often it happens, how long it takes now, and who would have to change how they work. That third one is the one people skip, and it is the one that determines whether anything actually happens.

A change that saves four hours a week but requires three departments to agree on a new process is a harder project than one that saves two hours and needs one person to change one habit. Volume is not the same as feasibility, which is the whole point of the priority matrix we use for the same job.

Where doing it yourself stops working

Being straight about the limits is more useful than pretending there are none. Three places where a self-assessment reliably runs out of road:

Your data does not look like you think it does. Almost every self-assessment assumes the data behind an opportunity is usable. It usually is not, and finding that out takes technical work rather than conversation. This is the single most common reason a self-made plan stalls after step one.

Nobody internally can say what is technically hard. Your team can tell you what is annoying and what is frequent. Judging whether a given problem is a weekend or a year of work is a different skill, and getting it wrong in either direction is expensive.

The uncomfortable answers do not surface internally. People rarely tell their own management that a process exists only because someone senior insisted on it years ago. An outsider gets told this in the first hour, every time.

What to do with the result

If you get through the six questions and end up with a ranked list you believe in, you have done the useful part. Take the top item, define what proof would convince you it works, and run it small, exactly as described in AI-roadmap maken. You do not need us for that.

If you get through it and end up unsure whether the top three are actually feasible, that is the point where an outside view earns its money, and you will be buying something specific rather than a general assessment. That is a better conversation for both sides.

About this page

The six questions are Radical's own working method, the same opening set we use in an AI-Foto session, written out rather than kept in-house. The feasibility-versus-impact reasoning matches the priority matrix published earlier. Written by Radical's own team; no client data was used.