Article

What Does an AI Consultant Actually Do? (And How to Tell If You Need One)

By Zac RobertsMay 21, 20267 min read

Owners of small service businesses get cold-pitched by AI consultants every week now. The pitches all sound similar: "transformation," "agentic workflows," "we will future-proof your firm." The price tag rarely matches what you actually receive. This article is a plain-English description of what a useful AI consultant does on a real engagement, what a bad one charges you for instead, and three questions that tell you whether you need one at all.

It is also the article we wish more prospects had read before talking to us, because half the time the honest answer is "you do not need a consultant for this, you need an afternoon and a free trial."

The job, in one paragraph

A useful AI consultant maps one or two of your workflows that cost the most operator time, evaluates whether an existing tool can handle them, and either implements that tool against your real data or builds a small custom service that does. The deliverable is a workflow that runs faster on Monday than it did on Friday. Everything else — slides, maturity models, "AI strategy" decks — is upholstery.

What a good engagement actually looks like

A real engagement for a five-to-fifty-person service business breaks into four phases, takes between three and six weeks end-to-end, and produces a measurable change in one workflow. Not the whole business. One workflow.

Phase 1: discovery (one week, fixed-fee or free)

The consultant sits with the partner or owner and writes down the three workflows that cost the most weekly time. They watch one of those workflows happen, end-to-end, with the operator who actually does it. They time it. They write down which steps could be automated and which steps require human judgment for legal, ethical, or relationship reasons. They pick one workflow to attack first.

If a consultant skips watching the real operator and instead asks for "a process document" — they are going to build the wrong thing. The process document is what the owner thinks happens. The actual workflow is what the operator does. They are never the same.

Phase 2: tool evaluation (3 to 7 days)

For the chosen workflow, the consultant evaluates two or three off-the-shelf AI products against the firm's real data, on a 60-minute test each. They write down what each product covers and what it does not. The output is a one-page recommendation: buy the tool, build a thin layer in front of it, or build a small custom service. The 80/40 framing we describe in the prior article in this series is the underlying rule of thumb.

Most consultants skip this phase because the recommendation might be "buy a $40/month tool and you are done," which is hard to charge a five-figure fee against. A good consultant tells you the boring answer if the boring answer is right.

Phase 3: implementation (2 to 4 weeks)

Now they actually ship something. If the recommendation was a tool, they configure it against your real systems, write the operator-facing documentation, and run a two-week shadow period where the tool runs alongside the old workflow so the operator can compare outputs. If the recommendation was a custom build, they write a thin service — typically a foundation-model call plus a handful of integrations into your existing software — and run the same shadow period.

A custom build for a single workflow at a small service business is not the six-figure project AI agencies pitch. Done well, it is two to four weeks of focused work, sized to your exact workflow, and lives on top of your existing stack rather than replacing it.

Phase 4: handover (one week)

The consultant trains the operator, documents the failure modes, hands over the credentials and the codebase or the tool admin, and writes a short post-engagement note describing what to monitor. Then they leave. A good consultant is not interested in becoming a permanent vendor. They want to be the person you recommend to your peers, not the person whose retainer you cannot get out of.

What a bad engagement looks like

Bad AI engagements share a few tells. None of them are subtle once you know to watch for them.

The first deliverable is a slide deck about "AI readiness" or a "maturity assessment." Nothing ships. The deck is the deliverable.
The proposal describes "transformation across the organization" but cannot name a single workflow it will measurably change in 30 days.
The consultant insists on a 6-to-12-month engagement before any code or configuration runs in production.
There is no mention of evaluating off-the-shelf tools first — the assumption is that custom is the answer before they have seen your workflow.
The pricing is per-seat, per-month, or otherwise structured to grow with your headcount rather than to be paid off when the workflow ships.

Three questions that tell you whether you need a consultant at all

Most small service businesses do not need an AI consultant. They need ninety focused minutes with a free tool. Before you sign anyone, answer these three questions honestly.

Question 1. Can you describe the workflow you want to automate in two sentences?

If you cannot, the bottleneck is workflow clarity, not AI. Spend a week documenting the workflow with the operator who runs it. You will either solve the problem without AI or have a much smaller, cheaper engagement to scope.

Question 2. Have you spent 60 minutes evaluating two off-the-shelf products on your real data?

If not, do that first. Most workflows at small service businesses are 80 percent or more covered by an existing product. You do not need to pay a consultant to discover that. You need a free trial and an afternoon. Only when both products fail the test does a consultant earn their fee.

Question 3. Is the workflow operator-critical and irreplaceable to your business?

If the workflow touches money, legal commitments, or core client relationships — yes, an experienced consultant pays for themselves on audit trail and failure-mode design alone. If the workflow is internal productivity (scheduling, summarization, drafting), a tool with a shadow period and a careful operator usually suffices. Save the consultant fee for the workflows where mistakes are expensive.

How Sensara handles this in practice

Our first consultation is free and the goal of it is to answer the three questions above honestly. If the answer is "you can do this yourself with a $40 tool," that is what we tell you, and we point you at the tool. If the answer is "you need custom, and here is roughly what it would cost," we say that. About a third of our intake calls end with us recommending a tool and not selling an engagement at all. That is fine — the calls that turn into engagements turn into engagements for the right reason.

If you want that conversation on your own workflow, send a short note through the form on this site. Include one workflow and roughly what it costs you per week. We will reply with which side of the consultant-vs-do-it-yourself line we think you are on, with the same honesty we use internally.