Do you need an AI consultant?

For a lot of people reading this, no. If you have a couple of free hours, you do not mind reading about software, and your week has two or three obvious problems rather than twenty vague ones, you will do a decent job of this yourself. Paying someone would be buying a service you can perform.

That is worth saying at the top, because every incentive here runs the other way. AI consultants and strategists was the second fastest-growing job title in the United States on LinkedIn's Jobs on the Rise 2026 list, behind only AI engineer. A great many people started calling themselves this very recently, and few of them will tell you when you do not need one.

I run assessments for a living, so treat this as advice from an interested party and weigh it accordingly — here is what I actually do. The rest is the test I would apply from your side of the table.

When is sorting it out yourself the right call?

Do it yourself when time is the thing you have and money is the thing you do not. If three of these four are true, you should almost certainly not be hiring anyone:

• You have the hours and do not resent them. Two to four hours of research, plus an hour per tool to set up. If you would read about software on a Sunday anyway, that barely counts as a cost.
• Your problems are the common ones. Meeting notes, drafting replies, cleaning up transcripts, first drafts, scheduling. Solved problems with well-known answers, and free tiers generous enough to test properly.
• You only need one or two things fixed. The value of outside help scales with the number of decisions. One decision is not worth a fee.
• You will accept a good-enough answer. The second-best note-taker, adopted on Tuesday, beats the best one on the market, adopted never.

There is also a cheaper first step almost nobody takes: before buying advice or software, spend a fortnight pushing whichever general assistant you already pay for much harder — your real documents, your real inbox problem. A fair share of "which tool should I buy" questions dissolve at that point.

Useful numbers rather than dramatic ones, and they set a ceiling on what anyone can honestly promise. If a consultant says they will hand back fifteen hours a week, ask which tasks disappear and how they got the figure.

When does paying for outside help actually pay off?

When your own time is worth more than the fee, and the research has been on your list for months without happening. Postponement is the real signal, not ignorance. Five situations where it stacks up:

1. You are at capacity. The research hours do not exist as spare time. They come out of billable work or out of your evening.
2. You have decision paralysis, not a knowledge gap. You have read the comparisons, you have fourteen tabs open, and you have decided nothing in six weeks. More information will not fix that — a filter will.
3. Your hour is expensive. At $150 an hour, four hours of research costs $600 in foregone work before you install anything.
4. Your situation is unusual. Client data with confidentiality obligations, an awkward existing stack, or a team who all have to adopt the same thing. Generic advice fails fastest here.
5. You have already bought things that did not stick. Subscriptions you pay for and never open are cheap evidence that the self-serve route is not working for you.

How do I work out whether it is worth the money?

Arithmetic, not enthusiasm. Four numbers, and you can do it on an envelope before you get on a call with anybody.

If payback runs past three months, the recommendation has to be unusually good. And if a provider will not attach an hours-returned figure to each recommendation, you cannot run the sum at all — which is its own answer.

[Steve — add a short example here of a caller you told not to book, and what you sent them instead.]

What should I watch out for in the AI consulting market?

Three things: vague deliverables, borrowed authority, and claims nobody has to substantiate. The money is arriving in this market considerably faster than the standards are.

The vendor layer underneath the advice is messy too. Of the thousands of vendors marketing "agentic AI", Gartner reckons only around 130 offer anything genuinely agentic — the rest is chatbots and automation rebranded at a premium — and predicts more than 40% of agentic AI projects will be cancelled by the end of 2027 on cost, unclear value or weak risk controls. A consultant repeating that marketing is passing on someone else's press release.

MIT's "GenAI Divide" study, published in July 2025, found 95% of generative AI pilots produced no measurable impact on profit and loss despite $30–40bn of investment. A sole trader adopting a transcription tool is a different animal from a bank rebuilding claims processing, so do not over-read it — but it should temper anybody's promises, mine included.

The legal floor is low as well. The US Federal Trade Commission launched Operation AI Comply in September 2024 with five enforcement actions against businesses making AI claims they could not support, including one selling a "robot lawyer". Regulators only reach the worst of it; everything short of fraud is left to you.

A consultant who has never talked anyone out of anything is not advising. They are selling.

What should I ask before hiring an AI consultant?

Five questions, which separate the people doing the work from the people who watched the same videos you did:

1. What exactly do I receive? A named list of tools with prices and setup times, or a strategy document? Ask to see a redacted example. If none exists, you are the first one.
2. How many hours a week does this return, and how did you get that number? Vagueness is disqualifying — it is the only figure that justifies the fee.
3. Do you take commission on anything you recommend? Affiliate income is not disqualifying. Undisclosed affiliate income is.
4. What happens if you find nothing worth changing? Someone who measures outcomes has an answer ready. A vague reply tells you how often they check.
5. Who have you done this for who looks like me? A consultant whose examples are all 400-person companies will prescribe like it, and those tools will drown you.

What if you are somewhere in between?

Most people are, and the way out is to size the problem before deciding who solves it. If your week leaks two hours, do it yourself. If it leaks eight, the arithmetic changes and outside help pays for itself inside a month.

Either way, doing nothing is the expensive option. Nearly 60% of US small businesses now report using AI in their operations, and 58% say they use generative AI specifically, up from 40% a year earlier. The question is no longer whether you use these tools, but whether you chose them deliberately or by accident.