It has been four months since the engagement quietly ended. The pilot was technically delivered but never deployed. Your operations lead has stopped mentioning AI at the leadership meeting. Two senior people have privately said they do not think your firm is ready. The next time you propose AI investment to the board, the room will be cooler than it was. None of that was on the consultant’s invoice.
This is the most expensive cost in any AI engagement, and it is the one buyers price least carefully. The consulting fee is the visible number. The real cost of a failed engagement is five to twenty times larger, and the buyer who does not factor it in is comparing the wrong numbers when choosing a proposal.
Why is the consulting fee around a fifth of the real cost of failure?
The fee covers the consultant’s time. The cost of failure covers everything else: your team’s time invested, the productivity disruption during the engagement, the reputational drag that delays your next AI investment, and the opportunity cost of operating without the capability for the 12 to 24 months you have spent on a project that did not produce one. The fee is one of five categories. The other four are usually larger, sometimes by an order of magnitude.
Three published 2024-25 data points anchor the failure rates. MIT NANDA in August 2025: 95% of GenAI pilots deliver zero ROI. Bain in April 2024: 88% of business change initiatives miss original ambitions across 24,000 audited cases. S&P Global Q4 2024: 42% of companies abandoned most AI initiatives, up from 17% in 2023.
The five categories of failure cost
Five categories make up the full cost of a failed AI engagement. Each one has a typical band, and the median total runs £100,000 to £150,000 on a £15,000 to £25,000 headline engagement.
Direct fees are the first and the smallest. The consulting fee for the engagement that did not deliver. £10,000 to £30,000 for most SME-scale strategy or pilot work. This is the only line that gets discussed at board level when the failure is debriefed, and it is consistently the smallest number on the page.
Internal staff time invested is the second. 100 to 150 hours for a strategy engagement, 150 to 250 hours for a pilot that was attempted, costed at loaded internal rates of £50 to £95 an hour. £8,000 to £25,000 per engagement, sometimes more if the engagement was extended trying to recover.
Productivity disruption during the engagement is the third. Senior staff distracted from operational work for the duration. Customer-facing leads pulled into workshops. The CFO running parallel financial models. Across an SME this typically runs 5 to 10% productivity loss for affected teams over the engagement window. £10,000 to £50,000 depending on team size and engagement duration.
Reputational drag is the fourth. After a failed engagement, the next AI proposal faces higher scepticism and a higher approval threshold. The board approves more slowly. Senior staff lose appetite. The delay typically runs 12 to 24 months before another AI investment gets the necessary internal support. Hard to price directly, but it is the gate to the fifth category.
Opportunity cost is the fifth. The cost of not having the AI capability deployed during the 12 to 24 months you spent failing to build it. For a use case worth £50,000 to £200,000 a year in operational gain, that is £50,000 to £400,000 of capability foregone. The number is large because the timeline is long. It is also the number that disappears completely from any post-mortem because nobody has the discipline to count what could have been.
A worked example
The maths sits roughly like this for a typical failed SME pilot. Direct fees £15,000. Internal staff time £15,000. Productivity disruption £20,000. Reputational drag delays the next attempt 18 months. Opportunity cost of capability foregone over those 18 months £75,000 if the use case was worth £50,000 a year. Total: £125,000 against a £15,000 headline.
The £15,000 was the only number on the consultant’s invoice. The other £110,000 sat on the company’s balance sheet as opportunity cost and morale loss, but it was real spend and it has compounded into the next year’s lower probability of trying again.
Integrate.io’s research on data integration projects puts failed-project cost at around 12% of annual revenue for affected firms. For a £3 million revenue SME that is £360,000, which is at the upper end of the band described above.
What this changes about proposal evaluation
The cost of failure changes what “expensive” means at proposal stage. A £15,000 proposal with vague specification, no mid-engagement review, and weak success metrics is the most expensive option in expected-loss terms, despite the cheap headline fee. It carries the highest probability of joining the 70 to 95% that do not deliver value, with an expected loss of £100,000 to £150,000.
A £25,000 proposal with bounded scope, audited data, quantitative success metrics, a mid-engagement go/no-go gate, full client ownership of artefacts, and a written exit report is materially more expensive on day one and materially cheaper on day 365. The expected loss falls because the failure probability falls. The £10,000 of additional fee is buying a £20,000 to £50,000 reduction in expected loss.
This is the right comparison at proposal stage. The wrong comparison is “is this proposal cheaper than the next-cheapest one?” The right comparison is “does this proposal reduce my risk of joining the failure rate?”.
How to make the cost of failure visible at proposal stage
You can put the cost of failure on the table without making the proposal conversation morbid. Three small habits do it.
Build a written estimate of the five categories before you read any proposal. Use the bands above as starting points and adjust for your business. The number does not have to be precise. It needs to be visible.
Ask each consultant under consideration what specifically reduces failure risk in their proposal. A good consultant will name the mid-engagement gate, the success metrics, the data audit, the knowledge transfer protocol. A consultant without an answer is offering the same proposal that produces the failure rate.
Make the comparison the board actually needs. The board approves a £15,000 fee or a £25,000 fee. The board cares about the £100,000 to £150,000 of expected loss reduction or increase. Show both numbers in the recommendation memo.
The failure rate is real. The cost of joining it is much larger than the consulting fee. Pricing the proposal against the cost of failure, rather than against the alternative quote, changes which proposal wins. It usually changes it for the better.
If you would like to talk through how to evaluate AI proposals against the cost of failure, book a conversation.
