Can AI actually solve your business problem?

Before asking whether a consultant can help, work out what kind of problem you have. Most disappointment in this field comes from applying AI to a problem class it does not touch.

Four classes. Find yours.

Class one: the same thing, many times, with clear rules

Repetitive work with a predictable shape. Answering a question that has one correct answer. Moving information from one place to another. Sorting incoming things into categories. Sending something at the right moment.

This is where AI genuinely delivers, reliably, today. It is also the least glamorous class, which is why companies skip it in favor of something more impressive and then conclude the technology underdelivered.

The two conditions that matter: it happens often, and being wrong occasionally is survivable. High frequency justifies the build. Low blast radius makes it safe to start there. Ranking candidates.

Class two: something nobody has time to do at all

Distinct from the first class and frequently more valuable. Not work being done inefficiently, but work not happening.

Follow-up nobody gets to. Renewals that lapse silently. Customers who went quiet two years ago. Inquiries arriving outside working hours. Reviewing your own data for patterns nobody has looked at.

The reason this class pays so well is that the comparison is against nothing. You are not making a person faster, you are causing something to exist. Every business has a list like this and almost nobody has worked it, because it is never the most urgent thing and never will be.

Class three: the process itself is broken

Now it gets uncomfortable. If work is unclear, disputed between departments, or genuinely badly designed, automation will not fix that. It will encode the confusion and execute it faster and more consistently, which is worse.

The tell: if you cannot describe the process in a few sentences without saying "it depends who is doing it", you have a process problem wearing a technology costume.

The right sequence is to fix it on paper first. Do that and you will sometimes find the fix was the whole answer and no build is needed, which is a good outcome that no vendor will suggest to you.

Class four: the constraint is somewhere else

The class where honest consultants should decline the work.

If you do not have enough customers, that is a demand problem and automating your operations will not create any. If your product does not do what customers need, no amount of efficiency addresses it. If you cannot hire the skilled people you need, AI does not currently replace a licensed technician or a senior engineer. If your margins are thin because of what you charge, that is a pricing decision.

Automation makes an operation more efficient. It does not create demand, fix a product, or replace expertise. Companies in this class often buy AI anyway, because it feels like progress and the real problem is harder to face.

The test

Ask what your actual bottleneck is, then ask what happens if it disappears.

If removing the bottleneck means you handle more work and make more money, you are in class one or two and this is worth pursuing. If removing it changes nothing because the constraint sits elsewhere, you have found class four and saved yourself a project.

One caution on how people usually frame this. Companies ask whether AI suits their industry, and industry is the wrong variable: the four classes above cut across sectors, and two businesses in the same trade can sit in different ones.

What a good consultant does with each class

Class one and two: rank them, value them, tell you what to do first.

Class three: tell you to fix the process, and say so plainly even though it is a smaller sale.

Class four: tell you the honest answer, which is that your problem is not this. Somebody who takes the work anyway is selling rather than diagnosing, and it is the question worth asking directly: what have you told a client not to automate?

Where consulting adds value versus where it does not

If you already know which class you are in and which processes matter, you need a builder.

If you cannot tell class one from class three, that is exactly the useful thing an outside pass provides, and it is hard to do for yourself, because you are inside the operation and the process looks normal from in there. The readiness test.

Our own diagnostic work is largely this sorting exercise applied across a whole operation, with values attached and an order to act in. It includes telling people when their constraint is elsewhere, which is a short engagement and the correct one. How it runs.