How can AI help with business model design for a startup?

AI drafts the canvases from the evidence report and the strategy rows, and regenerates them when the strawman changes, so the four value propositions, the model and the monetization sheet stay consistent with each other and with the field. What it does not do: hold the unit economics honestly, know what the floor can absorb on a Saturday, or decide a price. Those still need a spreadsheet and a person who has run a profit-and-loss statement.

Can AI do unit economics and pricing?

AI drafts the canvases from the evidence report and the strategy rows, and regenerates them when the strawman changes, so the four value propositions, the model and the monetization sheet stay consistent with each other and with the field. What it does not do: hold the unit economics honestly, know what the floor can absorb on a Saturday, or decide a price. Those still need a spreadsheet and a person who has run a profit-and-loss statement.

What parts of business design still need a human?

AI drafts the canvases from the evidence report and the strategy rows, and regenerates them when the strawman changes, so the four value propositions, the model and the monetization sheet stay consistent with each other and with the field. What it does not do: hold the unit economics honestly, know what the floor can absorb on a Saturday, or decide a price. Those still need a spreadsheet and a person who has run a profit-and-loss statement.

With AI · Business Design · Product Thinking

§The 's conclusion said AI made the middle nearly free. Strategy's conclusion said it generates the ladder and cannot cut it. Business design sits between the two: more of it is draftable than strategy, and the part that is not draftable is the part that decides whether the company lives.

§ 15.1What gets drafted#

§The canvases. Four value proposition canvases from the pages, with jobs, pains and gains pulled from the quotes and cited to interview codes, in minutes. The business model canvas from the strategy rows and the technical path recommendation. The monetization sheet's first column, the fronts that apply, from the three piles. In 2022 this was two days of a facilitator's time and the canvases drifted from each other. Now they are generated from one operating brief and regenerated together when a row changes.

§Consistency. This is the real gain. When the strawman's row changed the pilot restaurant, the feasibility sheet, the model's key activities and the list's owners all had to change with it. Regenerated from the brief, they did. Written by hand, one of them would have been missed, and the miss would have surfaced as a contradiction in a partner meeting.

§The first pass at the list. A model reading the canvases and sheets will extract the claims and propose numbers, owners and dates. The claims are usually right. The numbers are usually the category's benchmarks, which is a fine starting point and a dangerous ending one.

§ 15.2What does not get drafted#

§The unit economics, honestly. A model will produce a sheet in seconds, and it will balance. That is the problem. The sheet that matters is the one where a person who has run a profit-and-loss statement looks at the cost line and says: that is not what a server's glance costs, that is not what support costs per hundred users, that inference line assumes a price that will not hold when the model changes. Casado and Bornstein's observation about AI margins is now a line item every AI product carries, and it is exactly the line a generated sheet gets wrong most often, because the model has no idea what its own inference costs the company at scale.

§The operation's limits. No model knows what the floor absorbs at nine on a Saturday. It knows what interviewees said, and this whitepaper has spent two chapters on why what they said in a quiet room is not what happens. of delivery is learned by standing in the operation or by doing the job by hand, and chapter 5's discovery loop and concierge test cannot be prompted.

§The price. A model will propose prices, tiers and packaging that look like the category's. The three-number conversation with eight members of the segment, what is acceptable, what is expensive, what is prohibitive, is a human conversation, and the smallest market test is a real page with a real yes button. Pricing decided by generation is pricing decided by benchmark, and the benchmark is what the incumbents charge for a that is not yours.

§Which cost was invented. The point economy would have survived any generated business design, because it was internally consistent and the category has point economies. Six strangers filling in a form for nothing killed it. The gave-freely pile is built from watching people, and no draft contains it.

§ 15.3The operating brief, again#

§Every artifact in this whitepaper's template pack can be regenerated from one brief: the segments and their quotes, the strategy rows, the technical path, the decisions made and the open items with owners. The person maintaining the brief is doing the job the seventh whitepaper describes at length. In business design specifically, the brief must hold the numbers that were tested against internal data and the rows that were contradicted, because a regenerated sheet that reverts to the pre-data assumption is worse than a stale one.

§ 15.4What comes out#

§The list, owned. The sixth whitepaper asks how to test its rows in market once the product exists: the validation loops first, then the growth loops. This one asked whether the thing was worth building at all, and wrote down, with numbers, what would have to be true.