The owner of a small professional services firm is three months into a six-month AI engagement that is not delivering. She has raised the concern twice informally. Each time the vendor’s project lead has reassured her, promised attention, and changed nothing material in the work that has come back since. She is now sitting on a contract she has not reread, a stack of emails she has not collated, and a fairly uncomfortable sense that the next move could either fix the engagement or harden both sides.
This is the position a meaningful share of AI buyers reach in the second half of a first vendor engagement. The failures are predictable in shape, the recovery paths are narrower than they are for an enterprise buyer, and the sequence of moves available to an owner is short enough to write down. Knowing the sequence in advance is the difference between escalating cleanly and either over-tolerating a failing engagement or torching a relationship that could still have been recovered.
What is a failing AI engagement, in practice?
A failing AI engagement is one where the agreed work or outcomes have materially drifted from the contract, and informal feedback has not corrected it. Four shapes account for the bulk of cases. Scope drift, where the vendor keeps adding work the brief did not contain. Outcome underperformance, where the tool does not deliver. Technical issues, where the system breaks repeatedly. Communication degradation, where contact quality drops.
The four failure modes are not equally likely at any point in the engagement. Scope drift and communication degradation tend to surface early. Outcome underperformance becomes visible once enough usage has built up to test the result. Technical issues can arrive at any time and often cluster around model updates or integration changes. Naming the shape you are seeing is the first useful move because each shape has its own conversation.
Why does it matter for your business?
It matters because the cost of a failing AI engagement is rarely just the invoice. Gartner research has flagged that a large share of AI projects fail to deliver expected value, with poor implementation and unrealistic expectations the dominant causes. For an owner-operated business the consequences compound. Cashflow is tighter, the founder’s attention gets absorbed by recovery, and the team’s confidence in further AI work drops sharply.
The legal and commercial reality also sits with the buyer in ways that surprise some owners. The Air Canada chatbot ruling at the BC Civil Resolution Tribunal established that the business deploying an AI agent is responsible for its outputs in the same way it is responsible for an employee’s. UK contract law builds in an implied duty of good faith in some B2B relationships under the Yam Seng line of authority, but the practical consequence is still that you have to assert your rights, document the breach, and follow the process. Owners who tolerate a failing engagement for too long usually do so because the alternative looks more costly. The alternative is rarely as costly as it looks.
Where will you actually meet it?
You will meet it as a four-step ladder, with documentation discipline running through every step. Step one is a written summary of the issue to your named contact, with a specific response asked for by a specific date. Step two is a meeting that includes a more senior person on the vendor side. Step three is a formal notice under the contract’s dispute resolution clause. Step four is termination.
The escalation is sequential by design. Each step gives the vendor a clean chance to engage, and each step builds the paper trail that protects you if the relationship has to be ended. Step one is usually a single email, three or four paragraphs, with a response asked for within seven to ten working days. Step two is a video call with an agenda tying issues to commercial impact. Step three usually involves a commercial solicitor for a one-hour review before the notice goes out. Step four is the formal letter, the data return, and the post-termination housekeeping. Engagements that recover commonly do so between steps one and two. Those that end cleanly tend to do so between steps three and four.
When to ask vs when to ignore
Ask the question, in writing, the third time the same concern comes back unaddressed. Two informal raises and a third one in writing is a cleaner cadence than four informal raises and then a sudden formal letter. The written summary at step one is a request for a specific response, not yet a legal escalation, and vendors who can recover the engagement will often do so once the question is in writing.
Ignore the urge to escalate over a one-off slip. A missed call, a late deliverable, a single technical glitch, a slow week, these are not failure modes. They are normal noise in any commercial relationship and treating them as breaches hardens the vendor unnecessarily. The pattern to escalate on is the second occurrence of the same shape after the first informal raise, or any single occurrence of an issue with material commercial impact such as a regulatory breach, a customer-facing outage, or unauthorised data use.
What the salvage pattern looks like, when an engagement does recover, is also worth naming. Three features show up consistently. An explicit scope renegotiation, written down rather than verbally promised. A named-individual reassignment on the vendor side, usually a more senior project lead. A mutual acknowledgement, in writing, of what has not worked and what will change. Engagements that recover share these three. Engagements that limp on without them tend to end up at step three or four anyway, with more sunk cost.
Related concepts
Two pieces sit closest to this one. The failed AI engagement cost decomposed post puts numbers against a real burn and is worth reading before any termination decision. The post-burn vendor diligence post covers the retrospective review you run once an engagement has ended badly. This post is the live-escalation playbook between those two.
The reviewing an AI vendor contract post and the exit clauses and switching costs post are the contract-side pieces you will want open on screen at step three. The managing an AI vendor relationship as an owner-operator post covers the relationship discipline that prevents many of these situations arising in the first place.
A note on legal advice. This post is informational, not legal advice. Any escalation that reaches step three or step four should run past a qualified commercial solicitor before the notice goes out. Pre-action discipline materially affects the recovery you can achieve, and the hour of paid advice at the front of step three is the highest-value spend in the whole sequence. If you want to talk through where an engagement of yours has reached and what the proportionate next move looks like, book a conversation.
