Sometimes you don't. An honest consultant will say so to your face. If the problem is one tool and one task, drafting emails, say, or summarizing paperwork, a subscription plus an afternoon of practice covers it. Where a consultant starts earning the fee is when work crosses systems. Your inbox, calendar, CRM and invoices all have to cooperate. Customer data is on the line. Or you already tried to wire it up yourself and the thing didn't stick.
What does an AI consultant actually do?
The job title is vague enough to hide behind. Here's the concrete version. A good consultant starts by mapping how your business actually runs: where the hours go, which tasks repeat, what software already holds your data. Then they pick the right level of technology for each job, and no, it isn't always an agent. Some problems want a dumb automation. A few genuinely call for machine learning or an agent acting on its own. Plenty need nothing but a better process, and a decent consultant will say that in the first call.
The build comes after the map, not before. Design the workflow, wire it into the tools your team already uses, test it against real work, teach your people to run it. Anyone who skips the mapping and goes straight to tools is selling you software and calling it advice.
Can you just use ChatGPT and skip the consultant?
Often, yes. One chat subscription covers most of what a solo operator needs: drafts, summaries, research, a first pass at nearly anything written. Working consultants admit this openly, the good ones anyway.
DIY holds up under three conditions. The task has one trigger, a document arriving, an email landing. Nothing sensitive flows through it. And somebody on your team has a few spare hours a month to fix it when it wobbles, because it will wobble.
Where it breaks down is systems talking to each other. A chat window handles tasks; it will not watch your inbox overnight, update the CRM, and book the follow-up call by itself. We've watched owners wire that pipeline together on a weekend, get it working in testing, and then two weeks later a vendor changes something and it fails quietly and nobody has time to debug it. The tool was cheap. The stalled workflow was not.
When is it worth bringing someone in?
Watch for these signals:
- You can name the workflow that is costing you. "We lose two evenings a week to invoicing" is a consulting problem. "We should probably use AI" is not one yet.
- You tried the DIY route and it didn't stick. That is a failure of available attention, not intelligence, and it doesn't fix itself.
- The work touches data you can't afford to leak. Patient records, financials, anything with a compliance rule attached.
- The subscriptions keep piling up while the work stays where it was.
None of those describe you? Keep experimenting on your own. The experiments teach you exactly what to ask for later. And if the real goal is getting better at the DIY route, that's a job for coaching, not consulting.
How do you spot a consultant you should not hire?
The title has no license or exam behind it, and a lot of agencies rebranded as AI experts overnight. The red flags are depressingly consistent:
- A confident total price before anyone has looked at your business. A real quote follows discovery. A quote produced first is a guess dressed as a price.
- A strategy deck as the final deliverable. You're left holding a roadmap and none of the road.
- Commissions on the tools they recommend. Ask directly whether they earn anything from a vendor.
- A build only they can operate. If your team never learns to run it, you bought a dependency.
- A plan that starts by replacing the systems your team relies on. Rip and replace is where most AI projects die.
One question filters most of this: what will be measurably different in 90 days? A builder answers in specifics. A deck seller changes the subject.
What does AI consulting cost?
There's no honest universal number. That's exactly why the question gets asked so often and answered so vaguely. Hourly billing, fixed project fees, monthly retainers, all three exist, and the figures quoted online span an order of magnitude depending on who's selling.
The sequence matters more than the model. Price comes after discovery and attaches to a defined outcome, or it's a guess. Our own consulting works that way. No set price, because every business is different; a 15-minute call scopes what your team needs, what we'd build, and what it costs. When consulting is the wrong answer, the call ends with us saying so.
What should a first engagement look like?
Small and bounded. One workflow, a few weeks, a result you can check. The best first builds are almost always the unglamorous ones: scheduling, email handling, lead follow-up, the project-management grind. Work nobody will miss doing by hand.
Whatever gets built lands inside the software your team already uses, with a person in the loop approving what ships. When the first workflow has held up under real use for a month, you've earned the right to pick a second. That pace looks slow from the outside. It's the pace that survives contact with an actual business.