Hello, this is Ryuta Hamamoto from TIMEWELL. This is Part 2 of the AI-native new business development series. I spent more than ten years building new businesses inside large companies, led "CHANGE by ONE JAPAN", a challenger-support programme with over a hundred large companies, and now run WARP ENTRE, an AI-driven development programme for founders and intrapreneurs. In Part 1 I wrote about building the mock while you interview. This part is about turning what you heard into three sheets of business hypothesis.
Series: An AI-native methodology for new business development
New business plans are usually too thick. Read a hundred-page plan and it is rare to be able to say in three lines whose problem it solves, with what, and why nobody can copy it. In the programme I run, we go the other way. Before the thick document, three thin sheets. Lean Canvas puts "what to build" into words. 7 Powers decides "why it can't be copied". The design concept adds warmth to "how it should feel". Only when those three exist have you earned the right to write the thick one.
These three sheets used to take three weeks. With AI, one day. But let me say this up front: what got faster is the writing. The deciding did not get faster by a millimetre. For each sheet, this article separates what you can hand to the AI and what a person has to decide. For a read on where your team stands, the AI literacy check takes a few minutes.
Summary: Run the nine Lean Canvas blocks with an AI in ninety minutes, but keep "filled in" and "validated" apart by having a person narrow things down on the strength of interview quotes. Use 7 Powers' definition, Power = benefit × barrier to imitation, to have the AI score all seven from 0 to 5, then pick the one or two you can build early. For the design concept, choose three concept words, a font pair and four colours, compress them into a 300-character spec, have the agent produce HTML mocks, and pick from ten by comparison. The output of each sheet is the input of the next.
Why the three sheets are one flow
First, how the three relate. Lean Canvas is Ash Maurya's tool from Running Lean for seeing a whole business on one page: nine blocks covering problem, customer segments, unique value proposition, solution, channels, revenue streams, cost structure, key metrics and unfair advantage1. The nine steps I laid out in The complete guide to building a new business are this canvas rearranged into launch order.
7 Powers is Hamilton Helmer's classification of competitive advantage from his 2016 book of the same name. He defines Power, the condition for durable excess returns, as customer benefit multiplied by a barrier to imitation, and sorts it into seven types2. The ninth Lean Canvas block, unfair advantage, usually fills up with something vague like "know-how others don't have". 7 Powers is the tool for taking that vagueness apart. I went deep on it in 7 Powers: the seven forces behind durable advantage.
The design concept adds emotion to those two sheets of logic. The canvas decides what to build and Power decides why you win, but neither decides what a customer feels in the first three seconds. Worldview, tone, words, colour, type. Deciding those and turning them into a moving picture is the design concept. As I wrote in Differentiation when features can't differentiate, in an era when you cannot win on features, this step can itself be the unfair advantage.
The output of each sheet is the input of the next. The canvas's unfair advantage and customer segment feed the 7 Powers diagnosis; its unique value proposition and channels feed the design concept. Keep that order and each later sheet exposes the weaknesses of the earlier one. When you do 7 Powers and cannot write a single barrier, it means the canvas's unfair advantage was empty. Then you go back to the canvas.
Looking for AI training and consulting?
Learn about WARP training programs and consulting services in our materials.
Sheet one: Lean Canvas, into words in ninety minutes
The biggest trap of running a Lean Canvas with an AI, stated first: the AI fills all nine blocks, plausibly. A filled canvas looks like a validated canvas. What the AI filled in is a hypothesis. As Part 1 said, an AI customer hypothesis can only be tested against a customer. So before you run the canvas with an AI, assume you have at least a handful of verbatim interview records. If you do not, go back to Part 1 before this article.
Given that, three tasks can be handed to the AI. The first is generating candidates for narrowing the customer segment to a single person. Narrowing to one face, "Tokyo, thirties, dual-income, one-year-old child", is easier when three candidates sit side by side. Have it produce three people across six axes: demographics, occupation and role, values, daily routine, information sources, and receptiveness to new services. A person chooses, and the criterion is "the one I can describe in the most detail", the one closest to the interview records.
The second is listing degraded alternatives. The proof that a market exists is that people spend time or money on clumsy substitutes. "List ten clumsy alternatives this persona uses today for this problem, and score each 1–10 on time, money and mental cost." Anything scoring nine or above is your real competitor. Not the companies in your own industry.
The third is the Fermi estimate of market size. Have it produce TAM, SAM and SOM with the assumptions and arithmetic shown. What matters is not whether the number is right but which assumption is shakiest. Follow up with "name the single most uncertain assumption in this estimate and propose a way to check it", and you know what to research next.
A person decides three things: the top three problems, the one person, and the price. The problems in particular are not the AI's candidates but the ones the interviews showed to pass three tests: frequency, pain, willingness to pay. Do they hit it at least monthly? Does recalling it make them wince? Are they already paying for a degraded version? Only problems that pass all three become businesses.
The prompt pattern for the first draft is below. The point is to hand over the interview records first.
Using the interview records below (already extracted, with quotes) as evidence, draft the nine blocks of a Lean Canvas.
# Rules
- Attach the supporting interview quote number to every block entry
- Any block not supported by a quote is prefixed "HYPOTHESIS:" with one line on what to ask to confirm it
- At most three problems; rate each on frequency, pain and willingness to pay as ○ △ ×
- If unfair advantage cannot be written, write "TBD". Do not fill it in for the sake of it
# Interview records
[paste the Part 1 extraction]
Write "do not fill it in for the sake of it" and the AI honestly returns "TBD". That "TBD" tells you what to do next.
Sheet two: 7 Powers, deciding the moat
The iron rule for using 7 Powers on a new venture: you do not need all seven now. Each Power has a season in which it can be built. What a venture can build at launch is mainly counter-positioning (a new model incumbents will not copy because it would damage their own business) and cornered resources (people, patents, data or rights others cannot get). Network effects need seeding early; switching costs come mid-stage; scale economies, branding and process power grow from expansion onwards2.
Inside a large company there is a peculiar twist here. Counter-positioning is an advantage born of an incumbent being unable to act "rationally". But a corporate new-business unit is part of that incumbent. You are proposing a model that cannibalises your own core business. That is the intrapreneur's pain and, at the same time, the biggest opportunity. Territory the existing divisions "rationally" avoid is one of the few places an internal venture can compete on equal terms with an outside startup.
What you hand to the AI is the diagnosis. For each of the seven, have it score current strength, future buildability and the first action to take, from 0 to 5. With scores, you argue priorities on substance instead of feel. The prompt pattern:
Diagnose my business against 7 Powers, one at a time
(Scale Economies / Network Economies / Counter-Positioning / Switching Costs / Branding / Cornered Resource / Process Power).
Output a three-column table for each Power:
- Current strength (0–5, one-line reason)
- Potential to build in future (0–5, why)
- First move if we build it (one action)
Finish by recommending the top two Powers to build at our current stage, with reasons.
I make the decision. Recommend, do not decree.
# Business context
[Lean Canvas summary and three competitors, one line each]
A person decides which Power to grow as the main one. Then have the AI push one level harder on it: "What benefit does this Power create for customers? Why can competitors not copy it, or not want to? If that is vague it is not a moat, so press hard. Finally, point out, without deference, the weakest point in my barrier." Many people find they can write the benefit but not the barrier. That is the right outcome. If you cannot say why it cannot be copied, there is no moat yet. Back to the canvas's unfair advantage block, and rewrite.
Once moat candidates exist, check them against VRIO: value, rarity, imitability, organisation. All four ticked means a strong moat. Even with imitability ticked, if the organisation to exploit it is missing, the resource is wasted. In corporate new business the most common failure is exactly that last one. Access to the parent company's assets is supposed to be the advantage, and the new-business unit lacks the authority to use them. If the diagnosis shows this, internal alignment becomes the job, ahead of the business plan.
Sheet three: design concept, adding warmth
The third sheet is the one most business developers skip. "We'll get a designer later." In the AI era the order has flipped. You do not have to be a designer to conduct the design. Designers used to be the cooks who translated feeling into form. Now you set the menu and the coding agent serves the dish in seconds. Your role moved from the one who draws to the one who decides.
You decide four things, in order. First, concept words: exactly three. One and you get lost; five and no direction is set. Pick from at least two vocabulary groups, such as warmth, trust, energy and refinement. For an AI meal-planning app, say: watchful, calm, unpretentious. Those three words are the compass for fonts, colours and copy alike.
Second, fonts: choose the one mood that best fits the three words and fix a Japanese-Latin pair. Fonts are clothing. Dress wrong for the occasion and even a good person is not trusted. Third, colour: one primary, one accent, two neutrals, distributed 60/30/10. Equal proportions always shout. Avoiding pure black and pure white removes half the amateur look on its own. Fourth, adopt exactly one current design trend. You need not take them all, but consciously deciding which to adopt and which to reject is what professionals do.
Once decided, condense into a 300-character spec: the canvas's value proposition and channel, the three words, the fonts, the colours, and "the feeling we want in the first three seconds". Those 300 characters are the brief for the agent.
From the 300-character spec below, build the hero section of a landing page as a single index.html (Tailwind CSS via CDN, no build).
Structure: hero, a grid of features, a CTA. Produce three options, each with one line on which of the three words it expresses most strongly.
# 300-character spec
[value proposition / target customer / three concept words / font pair / four-colour palette (hex) / the feeling for the first three seconds / one adopted trend]
Screenshot the three options, send them back to the chat, and ask "against the three words, what is the single weakest point?" Three rounds. In ten minutes you have a real HTML hero. The core principle is diverge wide, decide fast. Rather than an hour polishing one option, five minutes producing ten and choosing by comparison. Not only do you see six times the options in the same hour; you decide before attachment sets in. Polish a single option and sunk cost makes retreat impossible. Comparison suits human instinct.
The moving hero from this third sheet also becomes the base of the live mock kit from Part 1. The picture you show in the next interview is now derived from canvas, moat and concept.
A one-day timetable, and where large companies get stuck
The timetable for running all three in a day: Lean Canvas for ninety minutes in the morning; 7 Powers for seventy minutes and the design concept for ninety in the afternoon; time in between to reread the interview records.
| Time | Step | Hand to the AI | A person decides |
|---|---|---|---|
| 9:00–10:30 | Lean Canvas | Three persona candidates, ten alternatives, Fermi estimate | The one person, top three problems, price |
| 10:30–11:00 | Reread records | Blocks unsupported by quotes | What to ask in the next interview |
| 13:00–14:10 | 7 Powers | Seven scores (0–5), weak points in the barrier | The main Power to grow, VRIO's organisation tick |
| 14:30–16:00 | Design concept | Three options × three rounds of HTML, weak-point critique | Three words, fonts, colours, the adopted trend |
| 16:00–16:30 | Cross-check | Contradiction detection | Back to the canvas, or forward |
The last thirty minutes matter most. Lay the three sheets side by side and hunt for contradictions. Is the canvas customer "an overworked section head" while the concept word is "playful"? Is the main Power a cornered resource that never appears in the canvas's solution block? Hand all three to the agent and ask it to "list every contradiction between these documents, quoting the passages", and it catches what you missed.
Large companies get stuck in two places. One is the canvas's revenue block: pulled along by the existing price list, a small price like "30,000 yen a month" cannot get through internally. The other is 7 Powers' organisation: no authority to use the parent's assets. Neither is a framework problem; both are alignment problems. And that is the point of three thin sheets. With a hundred-page plan, a senior manager does not know where to object. With three sheets, the objection lands somewhere specific: "that price won't fly", "we can't let you use that asset". A specific objection can be negotiated.
In WARP we run the three sheets on participants' own themes, in a single day. As I wrote in How to build a business in the AI era, when "being able to build" gets cheap, what is valuable is deciding what to build. The programme structure is on the WARP page.
Wrapping up
Words with Lean Canvas, moat with 7 Powers, warmth with the design concept. With AI, the three sheets take a day. But AI only sped up the writing. Narrowing to one person, to three problems, choosing the Power to grow, choosing three words: the deciding is done by a person, grounded in interview quotes.
If you do one thing tomorrow, take the canvas you have and write, in each of its nine blocks, "which interview and which statement is the evidence for this?" The blocks where you cannot write it are where you go and listen next. If you want to run the three sheets on your own theme with us, let's talk through a consultation.
Footnotes
-
Lean Canvas (Ash Maurya, LEANSTACK / Leanspark). The book is Running Lean (Ash Maurya, O'Reilly, 3rd ed., 2022) ↩
-
Hamilton Helmer, 7 Powers: The Foundations of Business Strategy (2016). For the definitions of the seven and which stage each is easiest to build in, see 7 Powers: the seven forces behind durable advantage ↩ ↩2
