In short: pick early customers for fit and learning, not logo value. Agree a written success test up front, work closely for six to twelve weeks, and convert the ones that succeed into paying customers, case studies and references.
Key points
- A good design partner has the problem badly and now, a named owner, resembles your next fifty customers, and will let you talk about it.
- Avoid the famous logo with no owner, the friend who says yes to anything, and nine-month enterprise procurement.
- Start with a named list of 30–50 organisations, each with a specific reason to talk now.
- Use a one-page agreement covering problem, success test and date, commitments, terms, and what happens if it works or doesn't.
- Paid pilots, even at a small fee, filter for seriousness; a free AI pilot often gets no attention.
Who makes a good design partner
- Has the problem badly and now. The urgency is visible: a deadline, a cost, a regulator, a competitor.
- Has someone who owns it. A named person with authority to try something new and time to give feedback.
- Looks like your next fifty customers. A partner too unusual teaches you to build for one.
- Will let you talk about it if it works, even anonymously.
Avoid the famous logo with no owner, the friend who'll say yes to anything, and the enterprise that needs nine months of procurement before you can learn anything.
Finding them
Start with a named list, not a market. For a small team selling to institutions, that's account-based selling at its most basic: 30–50 named organisations, researched one by one, each with a specific reason to talk now.
The warmest routes are usually the ones you built in Parts 4 and 6: your investors' portfolio companies, your first hundred community members, and their introductions.
Web3: your design partners might be curators, market makers, issuers whose asset you integrate, or wallets and front-ends that distribute you. An asset issuer has a reason to want a new use for their token — that's a partnership with a built-in incentive to promote you.
AI: your design partners are teams with a workflow you're replacing. Get access to real data early; a model that works on your test set and fails on their tickets is the most common way AI pilots die.
The one-page agreement
| Section | What it says |
|---|---|
| Problem | Their words, one paragraph |
| Success test | A measurable outcome and the date it's checked |
| Their commitment | A named owner, weekly time, access to data or systems |
| Your commitment | Roadmap priority, a direct line to the team, support level |
| Terms | Free, discounted or paid during the pilot; price after |
| If it works | Conversion to paid, a case study or reference, logo use |
| If it doesn't | What both sides learn, and a clean exit |
Paid pilots, even at a small fee, filter for seriousness. An AI pilot that costs nothing often gets nothing in attention.
Running it
Weekly check-ins with the owner. Ship something visible every one or two weeks. Log every piece of feedback in the evidence log from Part 3. At the success-test date, review honestly together: did it work, and what would make it a no-brainer to pay for?
What you get out
A product shaped by real use. Your first case studies. References for investors. And a clear picture of who your ideal customer actually is, which is usually narrower than the one in the deck.
The checklist
- [ ] Criteria for a good design partner, written down
- [ ] A named list of 30–50 candidates
- [ ] A one-page agreement with a success test and a date
- [ ] A price after the pilot, agreed before it starts
- [ ] A case study or reference ask built in
Common questions
How do I choose design partners for my startup?
Pick for fit and learning, not logo value: a partner with the problem badly and now, a named owner with authority and time, that looks like your next fifty customers, and that will let you talk about it if it works.
What should a design partner agreement include?
The problem in their words, a measurable success test with a date, their commitment, your commitment, pilot terms and price after, what happens if it works such as conversion and a case study, and a clean exit if it doesn't.
Should early pilots be free or paid?
Paid pilots, even at a small fee, filter for seriousness. An AI pilot that costs nothing often gets nothing in attention.
What is the most common reason AI pilots fail?
A model that works on your test set and fails on the customer's real tickets. Get access to real data early.