Selling AI to a Business That Was Burned Before

To sell AI to a business that was burned before, stop selling and start diagnosing. A prospect who has already paid for an AI project that failed is not sceptical about AI. They are sceptical about a specific thing that went wrong, and until you know which thing, every proof point you offer lands on the wrong objection. There are three common failure types, they need three different responses, and one of them means you should walk away. This is one of the least discussed parts of making money with AI, and increasingly one of the most common.

Why the standard pitch fails a business that was burned before

The instinct with a burned buyer is to differentiate. We are not like the last vendor. Our approach is different. Here is a case study.

This fails for a structural reason: the last vendor said all of that too. Every claim you can make about your competence is a claim they have already heard, believed, and been wrong about. Repeating it does not distinguish you, it puts you in the same category.

Worse, it skips the thing the buyer actually needs, which is an explanation of what happened to them. They have usually not been given one. The vendor left, the project was quietly shelved, and nobody did a post-mortem. You are talking to someone carrying an unexplained failure and a budget conversation they lost internally.

Offer the explanation first. It is the only thing you have that the previous vendor did not.

The three failure types

Almost every failed AI project a small or mid-sized business has experienced falls into one of these.

Ask three questions and you will know which one you are in:

1. When it was working the way it was supposed to, what did it do?

2. Who used it, and for how long before they stopped?

3. Is it still running?

Question one separates wrong-problem from the others: if they cannot describe a useful outcome even in the best case, the project was scoped badly from the start. Question two separates no-adoption: a tool that ran for three weeks and then went quiet is an adoption failure regardless of what anyone says about quality. Question three catches vendor collapse, because a tool that still runs and is still used was never the problem.

Selling into a wrong-problem failure

This is the easiest of the three, and the one people handle worst by rushing.

The buyer's real fear is spending again on something that turns out not to matter. So the response is not a better demo, it is a smaller commitment with a number attached to it. Find the thing that costs them measurable hours or money now, agree what a fix is worth, and scope the first piece of work to be smaller than the previous failure.

Make the success measure theirs and make it boring. Hours saved per week on a named task beats any capability claim. If you cannot find a number, that is information: you may be about to repeat the previous vendor's mistake.

Our notes on telling a client their AI project idea will not work are the flip side of this conversation, and worth having ready, because burned buyers sometimes come back with a version of the same bad idea.

Selling into a no-adoption failure

Harder, and the most common. The previous project probably worked fine.

The mistake is treating this as a training problem. It almost never is. Adoption failures happen when a tool asks people to change where they work: open a new tab, learn a new interface, remember a new step. Under pressure everyone reverts, and no amount of onboarding survives a busy Tuesday.

Two things move this buyer:

• **Show where the tool lives, not what it does.** If your answer is "in the tools they already have open", you are addressing the actual failure. If it is "in our dashboard", you are proposing the same project again.

• **Ask who stopped using it first.** There is usually one team or one person, and the reason is specific and knowable. Finding it out loud, in the meeting, demonstrates more than any case study.

Be honest that adoption is a shared risk rather than something you can guarantee. A vendor who promises adoption is making the exact promise the last one broke.

Selling into a vendor collapse

This buyer does not doubt AI or their own judgement. They doubt you will still exist in eighteen months, and if you are a small agency or a solo operator, that is a reasonable concern rather than an objection to overcome.

Answer it structurally, not emotionally:

• Where does the work live if you vanish? Their infrastructure, their accounts, their repository, with their credentials.

• What is the exit? Write it into the contract before they ask.

• What are they locked into? Name the dependencies honestly, including the model provider, because a provider price change is a risk they now know is real.

Saying "here is what happens if I get hit by a bus, and it is written into the agreement" is worth more than any amount of reassurance. It also converts your smallness from a liability into a demonstration of exactly the transparency the last vendor lacked. The related pricing conversation is covered in explaining AI costs to a client, which matters here because unexplained costs are how a lot of vendor relationships ended.

When to walk away

There is a fourth pattern that looks like the other three and is not. Some businesses were not burned by a vendor. They ran a project with no owner, changed the requirements three times, never freed anyone's time for it, and then blamed the tool.

The tell is that nothing about how they buy has changed since. Same absent sponsor, same fuzzy outcome, same expectation that the vendor will supply the internal will.

You will fail there too, and you will be the second name on their list of AI companies that did not deliver. That is expensive in a small market where buyers talk to each other. Decline, and say plainly why: without a named owner with time allocated, no vendor makes this work.

What to do differently in the proposal

Whatever the failure type, three changes to how you write the proposal:

1. **Reference their failure explicitly.** Name what went wrong last time and how this is structured differently. Pretending it did not happen makes them think you did not listen.

2. **Make the first milestone genuinely small and genuinely useful.** Not a pilot that proves nothing. One real thing, in production, in weeks.

3. **Define what failure looks like.** A stated kill criterion is the strongest trust signal available to you, because no vendor who intends to over-promise will offer one.

This is a longer sale than a greenfield one, and worth it. Burned buyers who do commit tend to stay, because they have now compared two vendors rather than one. The broader positioning work sits in selling AI services to local businesses and, once you have a win, writing a case study that sells your AI service, where a recovered project is unusually persuasive material.

One piece of context worth carrying into these meetings: your prospect is not unusual. MIT's Project NANDA report on the state of AI in business, published in 2025, found that around 95% of generative AI pilots produced no measurable impact on profit and loss. Whatever happened to your buyer happened to almost everyone. Saying so, plainly, does more to lower the temperature than any reassurance about your own track record.

FAQ

Should I criticise the previous vendor?

No. It reads as a sales tactic and the buyer often chose that vendor personally, so criticising them criticises the buyer's judgement. Describe the failure mode structurally without attaching a name to it.

How much discount does a burned buyer expect?

Usually none. What they want is a smaller first commitment, which is not the same as a cheaper one. Discounting signals you expect to under-deliver.

What if they will not tell me what went wrong?

Ask what the tool was supposed to do rather than why it failed. People discuss scope more freely than failure, and the scope usually tells you which of the three types you are in.

Is a free pilot a good idea here?

Rarely. The previous project probably had a pilot phase that went fine, so another one repeats the pattern that already misled them. A small paid engagement with a defined outcome is a stronger signal.