What does an AI consultant actually do all day?

“AI consultant” might be the most abused job title of the decade, so I understand the skepticism. A lot of what wears the label is someone reselling hype at an hourly rate. Since it’s a service I offer, here’s what the legitimate version of this work involves, partly so you can evaluate me and mostly so you can evaluate anyone.

Mostly, it’s listening

The first real task isn’t technical at all. It’s sitting with how your business runs (who types what into which system, where the waiting happens, which job everyone hates) and finding the places where today’s AI helps. Not where it demos well. Where it helps.

That distinction is the whole job. AI right now is spectacular at some tasks (summarising, drafting, extracting data from messy documents, triaging the repetitive stuff) and unreliable at others in ways that aren’t obvious until you’ve been burned. A consultant’s value is knowing which is which before you spend money finding out.

Then: prototype fast, on your real work

The next step should be a working prototype on your actual data within days: your real invoices, your real emails, your real mess. Not a slide deck. Slide decks are how AI consulting gets a bad name; a prototype either saves your bookkeeper an hour or it doesn’t, and everyone can see which.

Small businesses have an advantage here they rarely realise: you can adopt this stuff faster than the big guys, precisely because you don’t have a committee. A good consultant leans into that — small experiments, fast verdicts, keep what works.

Then: wire it in, or say so if it shouldn’t be

A prototype that lives in a browser tab dies in a browser tab. The lasting value is integration, with the AI step embedded in the tool your team already uses so nobody has to remember to use it. Sometimes that’s configuration; sometimes it’s a small piece of custom software around the AI; sometimes the right recommendation is an off-the-shelf product that already exists, and the engagement ends with me naming it.

Which brings up the part of the job that should reassure you most: a legitimate AI consultant says “don’t use AI for this” constantly. Some processes need a rule, not a model. Some need ordinary software with no AI in it. Some need a hire. If every problem you describe gets “AI can do that!” back, you’re talking to the hype reseller.

Questions that sort the real ones from the resellers

Ask what they’d try first in your business. A real one answers with something small and boring, not a transformation roadmap. Ask what AI is currently bad at. A real one answers instantly and specifically. Ask what happens if the prototype doesn’t work. A real one has a cheap answer, because their process is built to find that out early.

And ask what it costs to find out if there’s anything worth doing in your business at all. Here, that conversation is free. Worst case, you leave knowing which of your workflows AI can’t help with yet, which is worth knowing too.