Signs an AI consultant is overselling, and what an honest pitch sounds like

The signs an AI consultant is overselling mostly show up in the order things happen: a savings figure before anyone has watched the work, a product named before the problem is, a list of opportunities with nothing marked as not worth doing, and “fully automated” attached to anything that touches money. Each one is a result being sold before it has been measured. The fix is not to distrust AI. It is to ask what the claim is based on, and listen to the answer.

I sell AI automation, so read this list as applying to me as much as to anybody else.

The test underneath all of them

Every sign below is a version of one question: did this person look at how your business actually runs before telling you what it needs?

An honest pitch before that point is mostly questions an AI consultant should ask you. It asks who does the repeated work, in which systems, how often, and what goes wrong. It can say what kind of job usually pays back and what usually does not, but it cannot say which of those your business is. A pitch that already knows the answer on the first call is describing somebody else’s business, or nobody’s.

Seven signs, and the sentence you will hear

A savings figure before anyone has watched the work

“Businesses like yours save a big chunk of their admin time.”

A percentage or payback period quoted before the process has been observed is a sales number. It may be true of a business somewhere. It cannot be true of yours yet, because nobody has counted your hours. The honest version is conditional: if the order entry takes as long as you describe, a build of this size pays back in a stated time, and here is what I need to check that.

Ask where the figure comes from. “Another client” is fine if they can say how that business differs from yours. “Industry research” says nothing about your office.

The product arrives before the problem

“What you need is an AI agent on your website.”

When the recommendation is a product rather than a job, the consultant is selling what they build. Chatbots and agents are easy to demonstrate in a meeting, and the repeated work in the back office is not, so pitches drift toward the website. Whether your business needs an AI chatbot or an AI agent at all is a separate question from whether it needs one first, and the difference in scope between the two is set out in the AI chatbot vs AI agent comparison for business.

The honest order is the job, then the tool. A consultant who asks what happens between an enquiry arriving and a job being booked, before they mention any product, has the order right.

Every process on the list is a candidate

“We found fourteen opportunities for AI in your business.”

A long list with no negative recommendations is written to sell the next engagement. Most businesses have a few repeated jobs that pay back and several that do not, usually because the hours are too low, the process lives only in one person’s memory, or the exceptions have not been counted. A consultant who never says “not this one” has not done the ranking. The part that says what to skip belongs on any list of AI readiness assessment deliverables, and it is the part to read first.

“Fully automated” where money or a promise is involved

“It runs itself. Nobody needs to check it.”

For moving data between systems on a fixed rule, that can be true. For anything that sets a price, commits a date, answers a complaint or sends an invoice, it is the claim to push back on hardest. A language model composes plausible text, and a plausible number is not a correct one. An honest design keeps a person approving wherever money or a commitment leaves the business, and the pitch says so without being asked. If the build handles customer records, PIPEDA applies too, and the consultant should raise it.

Ask: which step in this build has a human approval, and what happens when the model is wrong?

“Our own AI” that nobody will name

“We use our proprietary AI platform.”

Most AI builds for small businesses run on a public model and a workflow platform. That is fine. What matters is that you can name them, that the subscriptions are held in your company’s name, and that somebody else could take the build over. A consultant who will not say which model or platform is underneath is either protecting a markup or making it hard to leave. Both are worth knowing before you sign. The accounts line is one of the rows in how to compare AI consultant quotes, and it is the one most often left blank.

Urgency borrowed from the headlines

“Your competitors are already doing this. You cannot afford to wait.”

Statistics Canada counted 19.2% of Canadian businesses using AI to produce their goods or services in the second quarter of 2026. Most of your competitors are not doing this yet. Urgency is a pressure tactic when it replaces a reason. The real deadline in most small businesses is whatever the repeated work is costing each week, and that is a number you can measure rather than a feeling to be sold.

No small first step

“The platform only makes sense if we roll it out across the whole business.”

A sound build, whether your office is in Toronto or Thunder Bay, starts with one job, running on your real records, that you can see working before you pay for the next one. A pitch that needs everything at once has no point at which you can check it. It also has no point at which you can stop, which is usually the reason for it.

What an honest pitch sounds like

The same conversation, from somebody who is not overselling, sounds different in small ways:

• “I do not know yet. Here is what I would need to measure.”
• “That one is real but it does not pay back at your volume. Leave it.”
• “This part stays with a person.”
• “The subscriptions go on your card, in your name.”
• “Start with the one job. If it works, we talk about the next.”

None of those sentences is exciting, which is part of how you can trust them. A pitch built to impress in the meeting is built for the meeting.

Three questions that sort one from the other

You do not need technical knowledge to test a pitch. Three questions do most of the work:

1. What would you tell me not to automate? An honest consultant has an answer, or says they cannot know until they have looked. An overseller finds a use for everything.
2. What number do you need from me to size this? The answer should be hours, volumes and systems. If they do not need any numbers, the price was not based on your business.
3. Where does a person stay in the loop? There should be at least one place in any build that touches customers or money.

If you are still at the stage of working out which processes to automate first in a small business, answer that before you take any pitch, because a consultant can only oversell into a gap you have not filled yourself.

Where an assessment fits

Much of the overselling above happens because the diagnosis is skipped and the sale starts at the solution. Paying for the diagnosis on its own, separate from any build, removes the incentive to skip it.

That is what a fixed-fee AI business assessment in Ontario is for. On this site it is $999 over ten business days: I watch how the repeated work is done, count the hours, and write up three jobs worth doing, the tools each one needs, and the ones to leave alone. Any other supplier should be able to price the build from it, which is the point. If the document only makes sense when I do the build, you should discount it the same way you would discount any of the pitches above.