The short answer
A good AI consultant starts with a business bottleneck and often talks you out of things. Be wary of anyone who leads with technology, cannot say what they would not build, or whose deliverable is a strategy document rather than a working system.
What AI consulting should actually deliver
The category is crowded and poorly defined, so it is worth stating plainly what the work should produce.
A useful AI consulting engagement produces a decision about where AI creates leverage in your specific business, and then a working system that proves it. Not a survey of the technology landscape. Not a roadmap. Something that runs.
The order matters. The engagement should begin with the workflow and the business outcome, and only then select the technology. Any engagement that begins with a tool and looks for somewhere to apply it is running backwards, and it usually ends with something impressive that nobody uses.
The questions that separate useful from expensive
What would you tell me not to build?
This is the most revealing question you can ask, and it is worth asking first. A consultant who cannot name a situation where their own service is the wrong purchase is selling rather than advising. Anyone genuinely experienced has a list of things that sound good and do not work.
What is the first thing that will be different, and when?
The answer should be specific, small, and soon. A named workflow, a measurable change, a timeframe in weeks. If the first deliverable is a strategy document or a discovery phase with no working output, you are buying an opinion.
Who owns what when we stop?
The accounts, the data, the configuration, the phone number, the content. Agree this at the start, in writing. It is trivially easy to settle before work begins and genuinely painful afterwards.
What breaks, and what happens then?
Every automated system fails sometimes. The important question is what the failure mode looks like. A system that quietly stops capturing leads is dangerous in a way that a system that loudly errors is not. Ask what the fallback is and who gets told.
Can I see something you built for a business like mine?
Not a logo wall. A working thing, with a real outcome, that you can inspect. Small consultancies often have less to show than large ones and better work to show it in.
Signals worth taking seriously
- Leading with the technology. If the pitch opens with a model, a platform, or an agent framework rather than your business, the fit is wrong.
- No stated limits. Any honest description of AI includes what it should not do. A pitch with no boundaries has not thought about failure.
- Guaranteed outcomes. Nobody can guarantee a ranking, a lead volume, or a revenue number. Confidence about method is reasonable, certainty about results is not.
- Deliverables that are documents. A roadmap is not a system. If the engagement ends with a deck, you have bought a plan you still have to execute.
- Replacing software that works. Proposals that begin by ripping out functioning operational tools are usually solving the consultant's problem, not yours.
- Vagueness about cost. Especially usage-based components. Ask what a peak month looks like, not an average one.
When you do not need a consultant
Some of the highest-return work in this space costs nothing and requires no help.
- Your Google Business Profile is inaccurate. Fixing hours, categories and service area is free and often outperforms months of paid work.
- You have never counted your missed calls. Do that before buying anything. The number decides whether any of this is worth it.
- You want to try a language model on your own workflow. Doing this yourself for a few weeks is genuinely instructive and will make you a much better buyer.
- The real problem is capacity, not leads. More inquiries do not help a business that cannot service the ones it has.
Does local matter for AI consulting?
Less than for a trade, more than for pure software.
The technical work does not require proximity. What proximity buys is context: understanding that your demand spikes with the first freeze, that your customers search a certain way, that the competitor two suburbs over is already bidding on your terms.
Our own research on this market found that seasonal demand swings in Minnesota trades are extreme and predictable, and that local intent searches behave very differently from expertise searches. A consultant who does not know that will build you something correct in the abstract and mistimed in practice.
For what we found, see why service plus city pages fail and AI Overviews and local search.
Questions people ask about this
01What does an AI consultant actually do?+
They should identify where AI creates leverage in your specific workflows, then build and prove a working system. The engagement starts with the business outcome and selects technology afterwards, not the other way around.
02What is the best question to ask an AI consultant?+
What would you tell me not to build. Anyone genuinely experienced has a list of things that sound appealing and do not work. A consultant who cannot name a case where their service is the wrong purchase is selling rather than advising.
03How much does AI consulting cost?+
It varies too widely by scope for a single figure to be honest. What matters more is the structure: whether there are usage-based components, what a peak month costs rather than an average one, and whether the deliverable is a working system or a document.
04Do I need an AI consultant at all?+
Often not yet. If your Google Business Profile is inaccurate, you have never counted your missed calls, or your constraint is capacity rather than leads, those come first and none of them require a consultant.
05Should an AI consultant be local?+
The technical work does not require it, but local context does matter. Knowing how demand moves seasonally in this market, and how local searches behave differently from expertise searches, changes what gets built and when it ships.