AI ROI pretotyping
What is AI ROI pretotyping?
The cheapest evidence about an AI bet is the evidence you collect before you build it. Pretotyping is how you get it in days, not quarters.
The short answer
AI ROI pretotyping is the practice of testing whether an AI use case will create measurable value before you build it. Instead of committing engineering budget on a business case, you run fast behavioural experiments, often with no production AI at all, to prove real demand, adoption, and payoff, then fund only the bets the evidence supports.
- Pretotyping was created at Google by Alberto Savoia and is taught at Stanford.
- It answers “are we building The Right It?” before engineering answers “are we building It right?”
- Tests use real human behaviour, clicks, sign-ups, usage, willingness to pay, not opinions.
- Exponentially has run the method across 4,000+ experiments for enterprise teams.
Prototyping vs pretotyping
A prototype answers “can we build this, and does it work?” A pretotype answers a cheaper, earlier question: “if we built this, would anyone actually use it, and would it pay off?” For AI bets that second question is the expensive one to get wrong.
Pretotyping deliberately fakes the hard, costly parts so you can measure real behaviour fast. A workflow can be run by a human behind the scenes, an agent can be simulated, and a landing page can test demand, all before a model is fine-tuned or an integration is built.
Common AI pretotyping techniques
A few proven patterns do most of the work on AI bets:
- Wizard-of-Oz, a human quietly performs the AI’s job so you can measure whether the output is actually valued and used.
- Concierge, you deliver the outcome manually, end to end, to learn what “good” looks like before automating it.
- Fake door, you advertise the AI capability and measure real demand before any of it exists.
- Pinocchio / mechanical, a non-functional stand-in that tests whether people will adopt the workflow at all.
Why it beats the AI business case
The stakes are well documented. RAND finds more than 80% of AI projects fail, roughly twice the rate of non-AI IT projects, and McKinsey’s 2025 survey shows that while 88% of organisations use AI, only 39% report any enterprise EBIT impact. Business cases built on assumptions, adoption rates, time saved, error reduction, are exactly where that money leaks.
Pretotyping replaces the riskiest assumptions with data. If a fake-door test shows nobody clicks, or a concierge run shows the output isn’t trusted, you’ve saved the build cost and the opportunity cost.
That is why pretotyping is the engine inside AI portfolio governance. It makes evidence cheap enough that the board can demand proof before capital without grinding delivery to a halt. The result, across our work, is teams avoiding spend on the bets that were never going to pay off.
Where it fits the AI Bets Audit
In a two-week AI Bets Audit, pretotyping is step three. After we map and rank your portfolio of bets, we pretotype the one to three highest-stakes ones to replace their biggest assumptions with behavioural evidence, so the fund / fix / kill decision rests on data, not debate.
Sources
- 95% of enterprise generative-AI pilots produce no measurable P&L impact. MIT NANDA, State of AI in Business 2025
- More than 80% of AI projects fail, roughly twice the failure rate of non-AI IT projects. RAND Corporation, 2024
- 88% of organisations use AI, but only 39% report any enterprise-level EBIT impact. McKinsey, The State of AI 2025