You have a shortlist of AI consultants and you want to avoid an expensive mistake. Which warning signs should make you slow down, ask more, or walk away? This guide lists the ones that matter most for a small business, explains why each one matters, and says what to ask or do when you spot it.
The short answer: worry less about polish and more about whether they understand your problem, can explain what they would build, and will leave you able to run it without them.
Key takeaways
- A tool pitched before anyone has understood your problem is the most common warning sign.
- A guarantee of perfect results is a reason for caution, not comfort.
- Ask for a customer you can speak to, not just a case study the consultant wrote.
- Check before you sign that you can get your data out and change the system without them.
- If you spot a red flag mid-project, go back to the measures and stop points you agreed.
Which promises should worry me?
A tool before the problem: If the first meeting is about a platform, a product or "an AI agent for your business" before anyone has asked how the work is done today, the answer was decided before they met you. A good adviser starts by asking what outcome needs to change and what the problem costs you now. Some will tell you part of it needs a process fix or plain automation, not AI. That is a good sign.
Guaranteed results: No honest provider can promise that AI output will always be right. These tools make mistakes, sometimes confidently. A promise of flawless output usually means the consultant has not thought about how errors will be caught, or hopes you have not. Ask instead: "How will we spot and fix mistakes, and who checks the output before it reaches a customer?"
"It needs no human review." Closely related. Any claim that you can switch it on and stop checking deserves a hard look. See keeping a human in charge of AI for what sensible checking looks like.
The biggest model for everything: Proposing the most powerful, and usually the most costly, option for every task is a sign of selling, not designing. Ask them to explain why a smaller or simpler option would not do the job.
What if it is not really AI?
Some offers are ordinary automation with a new label, or a thin layer over a public chatbot you could use yourself. That is not always bad: plain automation is often the right answer. The problem is paying AI prices for it, or being told it does more than it does.
Ask: "What would change if the AI part were removed?" and "What exactly does the system do, what information does it use, and how does it reach an answer?" A consultant who cannot explain this in plain English, without retreating into jargon, either does not understand it or does not want you to.
How do I spot borrowed case studies or inflated experience?
| Warning sign | What to ask or do |
|---|---|
| Only case studies they wrote themselves | Ask to speak to a customer whose work was like yours |
| No customer will take your call | Treat the case studies as marketing, not evidence |
| Results described without a starting point | Ask what the numbers were before the work began |
| The person selling is not the person doing the work | Ask to meet who will actually do it, and check their experience |
| A security certificate offered as proof the AI works | It shows how they handle information, not whether the AI suits your task |
| A demo only on their own tidy examples | Ask them to try it on a sample of your real work |
A case study can be accurate and still tell you little. What carries weight is a customer describing results against their own starting point, and saying whether those results held after the pilot ended. Our guide to checking an AI consultant's credentials covers how to do this, and how to tell if an AI demo will work covers testing a demo.
What does lock-in look like, and how do I avoid it?
Lock-in is when you cannot leave, or cannot change anything, without the consultant. The signs are usually visible before you sign:
- They cannot say how you would get a full export of your data, settings and instructions.
- Only they can change how the system behaves.
- Everything runs on their accounts rather than yours.
- Handover and training are vague, or missing from the proposal.
The fix is to agree up front which changes your team must be able to make unaided, and to make knowledge transfer a written deliverable. Our guide to avoiding AI lock-in has the questions to settle. Contract wording about ownership and exit is a matter for a solicitor.
When is a very low or very high quote a warning sign?
Price on its own tells you less than you might think. A low quote may leave out the work that makes AI useful: preparing your data, connecting it to your systems, training your team and checking results. A high quote may include work you do not need, such as a custom build where a feature in software you already pay for would do.
So compare what is included, not just the total. Ask each consultant to list what the price covers, what it leaves out, and what it will cost to run each month once they have gone. Running and checking costs can outweigh the fee over the life of a project. Our guide to what an AI consultant costs gives sourced UK day rates to compare against.
What should I do if I see a red flag partway through?
Do not panic, and do not blame yourself. Most problems can be sorted if they are raised early.
- Go back to what you agreed. If you set measures and a stop point at the start, use them. That is exactly what they are for. If you did not, agree them now.
- Ask the question directly. "What happens when the input is messy?" or "Show me how a mistake gets caught" are fair questions at any stage.
- Secure your position. Make sure you have copies of your data, settings and any instructions used to make the system work.
- Decide. Carry on, change course, or stop. Stopping a project that is not working is a good outcome, not a failure. See why AI pilots stall.
If you think the consultant has not delivered what the contract says, take advice from a solicitor before you act.
Red flags checklist
- They asked about my problem before suggesting a tool
- They explained in plain English what the system would do and what information it uses
- They told me how mistakes would be caught, and did not promise perfection
- They justified why a simpler option would not work
- They offered a customer I can speak to about similar work
- I have met the people who will do the work
- The quote lists what is included, what is not, and the monthly running cost
- I know how to get my data and settings out, and which changes my team can make alone
- We agreed measures and a stop point before work starts
Any unticked box is not automatically a reason to walk away. It is a question to ask before you sign.
Download: red flags checklist (Word)
Common questions
Is one red flag enough to rule someone out?
Not usually. A missing customer reference from a new consultant may have a good reason. A pattern matters more than one gap: a tool-first pitch, no references and a guarantee together are a strong signal.
Should I worry if a consultant says AI is not the answer?
No. It is one of the best signs you can get. It means they are thinking about your problem rather than their sale.
For the full process, start with how to choose an AI consultant and take the questions to ask into your first call.