An anonymous real example
A team used generative AI on complex bid documents, including commercial spreadsheets, pricing and costings. They got into difficulty because work requiring explicit calculations and checks was being treated like a drafting task. In a separate bid-team trial, AI was used to pull requirements from tender documents. Incorrect extractions created extra checking work. Producing the answer faster had not made the whole task faster or more dependable.
The assumption worth revisiting.
A bid mixes very different kinds of work: persuasive writing, interpreting requirements, calculating a price and committing the business to delivery. One tool and one review method do not automatically suit all four. The familiar efficiency question is ‘How much faster can we submit?’ The more useful commercial question is ‘Which parts of this promise can we demonstrate are right?’ An error in the first interpretation can propagate through a beautifully consistent proposal. A missed requirement can make a bid non-compliant. A costing error can turn a win into work the business cannot deliver at the expected margin. Those are possible consequences of the mechanism, not a claim that every AI-assisted bid suffers them.
Are you making the proposal faster—or making a better promise?
Give the capacity a purpose.
Put some of the drafting time saved into joining up Bid, Finance, Legal and Delivery around the assumptions that determine the deal. Review the economics and the promise before celebrating the submission speed.
How this fits the capacity allocation →
Use AI to challenge your first interpretation
Explore this question without assuming our first interpretation is right: Are you making the proposal faster—or making a better promise? Offer competing explanations, identify assumptions and suggest evidence we could obtain from the people doing the work.
Treat the answers as possibilities to test with people and evidence.
Go deeper into the evidence.
Organization Science / INFORMS · Peer-reviewed field experiment
In a P&G innovation experiment involving 791 professionals, AI-assisted individuals matched unaided teams on proposal quality. AI teams were 9.2 percentage points more likely than unaided individuals to produce a top-decile solution.
One company and early innovation tasks, not realised growth. The AI-team versus unaided-team top-decile difference was not statistically significant. The final analysis excluded non-randomised participants; 776 participants supplied complete post-task data.
video podcast · 60 min 19 sec · Start at 37:58
Ethan Mollick: How to build an AI-first organization
Sana · Strange Loop
A researcher explains an experiment that challenges assumptions about individuals, expertise and teams. Start at 37:58 and ask what a team must add beyond a capable individual with AI.
Listen or watch for: If AI helps individuals cross expertise boundaries, what additional value should a human team create?
About this resource
One study’s author discussing a specific experiment, not proof that every task needs a cross-functional team. The page warns its AI transcript may contain errors.