Do You Need an AI Consultant? An Honest Guide for Small Businesses

Somewhere between "we should probably do something about AI" and an actual working system, most small businesses hit the same wall. You have read the articles. You have tried ChatGPT. Someone in accounts is quietly using three tools nobody signed off on. And the honest summary of where you stand is the one owners keep repeating in forums: six AI subscriptions, no coherent workflow.

At that point the question arrives: do we need to pay someone? No consultant can answer that for you without a conflict of interest, so here are the criteria themselves — when outside help earns its fee several times over, when it is money you will regret, what the going rates are in 2026, and how to tell a strategist from a reseller with a slide deck.

What an AI consultant actually does

"AI consultant" covers at least four different jobs, and the most common way to waste money is to buy one when you needed another.

The strategist. Works top-down. Maps your workflows, scores them for automation potential and risk, and hands back a prioritised sequence with a business case attached. The deliverable is a document and an order of operations. They do not build anything.

The implementer. Builds the thing — an automation in Make or n8n, a customer-facing assistant, a searchable knowledge base over your documents. The deliverable is a working system that you then own and maintain. Good implementers are often terrible strategists, and will happily build whatever you asked for even if it was the wrong workflow.

The trainer. Runs workshops, builds a shared prompt library, and gets the team past the "I tried it once and it made something up" stage. The deliverable is adoption. This is the cheapest kind of help and the most often skipped.

The engineering shop. Fine-tuning, integrations into your ERP, data pipelines. Engagements start around €20,000 and go up steeply. For a business under 100 people this is almost never the right first purchase.

What none of them can do is tell you which tools will matter in eighteen months, or care about your margins more than you do. Treat anyone who promises either as a salesperson.

Four signs outside help will pay for itself

1. Your bottleneck is a process, not a tool. If quoting a job takes three days because the information lives in four places and two people's heads, no subscription fixes that. Somebody has to redesign the process first, and that is genuinely hard to do from inside it.
2. You can name a repeated, measurable task. "Our two admins spend about eleven hours a week re-typing supplier invoices into our accounting system." That sentence is a business case: eleven hours at roughly €22 an hour across 46 working weeks is about €11,000 a year, so a €4,000 build pays back inside five months and keeps paying. Vague ambitions do not survive contact with an invoice; numbers like these do.
3. You handle regulated or client-confidential data. If you are processing health records, legal files, candidate data or anything else that attracts a regulator's interest, the cost of getting classification and data handling wrong under GDPR and the EU AI Act dwarfs a consulting fee. Paying someone who has read the legislation properly is cheap insurance.
4. You have tried twice and adoption died both times. Two failed attempts is a pattern, and patterns rarely break without an outside push. Usually the problem is not the tool: it is that no single person owned the workflow and the old manual route was still available.

Five signs you do not need one yet

This is the more common situation, and it is worth being blunt about it.

1. You have not written down your repeating tasks. A consultant's first two days will be spent extracting that list from you, at €1,000 or more per day. You can produce the same list in an afternoon with a spreadsheet and the team.
2. Nobody on the team uses AI daily yet. Buy three seats of one good assistant for sixty days first. At roughly €20–25 per user per month, that is less than a tenth of a consulting day and it tells you more about your team's appetite than any assessment will.
3. Your actual question is "which tool should I use?" That is a two-hour exercise, not an engagement. Our guide to auditing your AI tool stack walks through the same scoring a consultant would use, and it will also tell you which of your current subscriptions to cancel.
4. You cannot name the metric that would move. If you do not know whether success looks like hours saved, quotes sent, churn reduced or errors cut, nobody you hire will know either. They will pick a metric that makes the project look good.
5. You are a team of five or fewer with little process debt. At that size the owner plus a weekend plus a decent framework usually beats an external engagement, because you are the process and you can change it on Monday.

What AI consulting costs in 2026

European market rates, as a rough map rather than a price list:

• Productised assessment or audit report: €200–€1,500, fixed scope, delivered in days. The lowest-risk way to buy outside judgement.
• Independent consultant day rate: €800–€2,000 per day. Specialists in a regulated niche sit at the top of that range and are often worth it.
• Boutique strategy engagement, two to four weeks: €5,000–€15,000, ending in a roadmap and a prioritised backlog.
• Automation build, per workflow: €2,000–€8,000 depending on how many systems it has to touch.
• Fractional AI lead on retainer: €1,500–€4,000 per month, typically a day or two a week. Sensible once you have several live workflows to keep honest.
• Large-firm transformation programme: €40,000 and up. Not aimed at you, and the deck will say so in the small print.

One rule of thumb keeps most SMBs out of trouble: do not spend more on the diagnosis than you would spend on twelve months of the tools themselves. If your realistic annual AI tool budget is €3,000, a €15,000 strategy engagement has the economics the wrong way round. Buy a fixed-scope diagnostic, act on it, and only then decide whether you want somebody for longer.

Three pricing structures deserve suspicion: payment as a percentage of claimed savings (the savings will be calculated by the person being paid), undisclosed commissions on the tools they recommend, and a twelve-month retainer signed before a single pilot has run.

Seven questions that separate strategists from resellers

Ask these on the first call. The answers sort candidates faster than any proposal.

1. "Which three workflows would you look at first in a business like mine, and why those?" Someone who knows your sector will reference its cost structure within five minutes. Someone who does not will talk about "quick wins" in the abstract.
2. "What will still work if you disappear next month?" You want documentation, named owners, and a system your team can run. Not a dependency.
3. "Show me a before-and-after with real numbers from a business my size." Logos on a slide are not evidence. A measured baseline and a measured result are.
4. "Whose accounts will the tools sit in?" Yours. Always yours. If the automation platform, the API keys and the data live in the consultant's account, you have bought a hostage situation, not a system.
5. "What would you tell me not to automate?" Anyone who thinks everything is automatable is selling. A good answer names something specific — first-contact complaints, anything with legal exposure, work where your judgement is the product.
6. "How do you charge, and what happens in month four?" You are listening for a defined end, a handover, and a price that does not quietly become permanent.
7. "What does the EU AI Act mean for what you are proposing?" If they cannot discuss risk classification or transparency obligations in plain language, they have not read it — and you will be the one carrying the obligation, not them.

The DIY route, and how far it gets you

For most businesses under 50 people, a fortnight of structured owner time gets you most of what a strategy engagement would produce. The sequence:

1. List every repeating task that takes more than thirty minutes a week. Get the team to add to it for three days. Aim for forty lines, not ten.
2. Score each one from 1 to 5 on three axes: how often it happens, how rules-based it is, and how much it would cost you if the output were wrong. Anything scoring high on the first two and low on the third is your starting line.
3. Baseline the top two before you touch a tool. Time taken, error rate, cost per unit. Half of all AI business cases collapse under scrutiny because nobody measured the "before", and you cannot reconstruct it afterwards. Our guide to calculating the ROI of an AI implementation sets out the arithmetic.
4. Run a 30-day pilot with one tool and one named owner per workflow. Close the old manual route for the duration, or people will drift back to it by week two.
5. Review with the numbers in front of you. Keep it, kill it, or escalate it to paid help with a precise brief: this workflow, this baseline, this target, this budget.

That last line is the real argument for doing it yourself first. A consultant handed a brief like that will do better work, faster, for less — and producing the brief is exactly the deliverable you would otherwise be paying €2,000 for. If you want the framework behind the scoring, how to create an AI strategy for small business covers it step by step.

The bottom line

Hire outside help when you have a named, measured problem that the business has already failed to solve twice, or when the regulatory downside is bigger than the fee. Do not hire it to answer "where do we start" — that is the one question you are better placed to answer than anybody you could pay, and the one most likely to be answered with a generic roadmap you will never open again.

Buy the smallest piece of outside judgement that would change your next decision. Fixed scope, clear deliverable, your accounts, your data, a defined end. Then go and run the pilot yourself.