What we expected to work
The plan was the usual one: put the product somewhere visible, write about it, wait. What actually moved things was calling regional travel agencies one at a time and asking what part of their itinerary work was worst — not describing the product until the end of the conversation, if at all.
That's a slow way to find users and it changed the product before we'd built the wrong version of it. The most important thing we heard wasn't a missing feature. It was that agents didn't want a model overwriting their judgment — they wanted a first draft they could edit and keep. That single point shaped the whole drafting design.
What made a message get answered
Anything that could plausibly have been sent to a hundred people got no reply. Anything referencing something specific and true about that person's work usually got one, even when the answer was no.
- Asking about their process, not for their time to hear about ours
- Short — under a hundred words was better than a careful longer version
- An explicit way out, which seemed to increase replies rather than reduce them
- No attachment, no deck, no calendar link in the first message
What was harder than expected
Two things. The first was resisting the urge to pitch early in a call — once you've described the product, everything the other person says afterwards is shaped by wanting to be encouraging, and the call stops producing information.
The second was accepting how long it takes. A conversation is fifteen to thirty minutes plus scheduling, and there's no version of this that produces ten users in an afternoon. It took considerably longer than the build did to reach a point where we knew who the product was for.