When the research kills your favourite idea — designing a roster from data instead of taste
I was designing the character line-up for a miniatures campaign aimed at a specific fan community. I had a set of archetypes I liked and that fit the theme beautifully. Before sculpting anything, could I check whether the audience actually wanted them — and would I accept the answer if it said no?
It said no, and the correction was the whole value. I'd picked characters that served the *theme*; the audience buys the ones that represent *them*. Almost none of my elegant thematic picks appeared in the actual demand data. The bigger find came from a number I nearly skipped: a quarter of that community doesn't fit any catalogue category at all — which turned a side offering into the one aimed at the largest single segment.
The research is done and it's now the basis for the line-up. What's deliberately not here are the specifics — it feeds a campaign that hasn't been announced. This entry is about the method, not the product.
The cheapest sculpt is the one you don't make
A miniature is expensive before anyone can buy it. Concept, high-resolution sculpt, print-readiness, test prints, revisions. Choosing the wrong subject doesn't cost you a sketch — it costs weeks.
So before any of that, one question: does the audience want this, or do I just like it?
What I had, and why it was wrong
I'd built a set of archetypes around the theme. They were coherent, they were fun, and each one justified itself within the fiction. I was pleased with them.
Then I went looking for actual demand data and found a peer-reviewed academic survey of the community, run internationally, with over ten thousand accumulated participants. Not a forum poll. Not vibes. A primary source with methodology attached.
Almost none of my archetypes appeared anywhere near the top of what people actually identify with.
The mistake has a clean name: my picks were theme-led. I'd chosen characters that served the story. But in a community where people buy a figure because it represents them, demand doesn't follow the theme — it follows identity. Those are different questions and I'd answered the wrong one confidently.
The fix wasn't to throw out the theme. It was to invert the order: take the genuinely in-demand subjects and dress them in the theme's roles. Now the most-wanted subjects each get a distinct part to play in the fiction. Coverage and coherence at once, instead of coherence alone.
My favourites survived — as flavour, not as the foundation. That's a much better place for them, and I wouldn't have moved them there on my own.
The number I nearly skipped
The ranking table had two rows I initially read as noise: "Other" at about 24%, and "Hybrid" at 14%.
Those look like data-cleaning leftovers. They're not. Somewhere between a quarter and a third of this community doesn't map to any clean catalogue entry — by choice. No standard line-up will ever contain them.
That reframed an entire part of the offer. There's a bespoke tier in the campaign that I'd been treating as a premium curiosity off to one side. It isn't. It's the tier aimed at the single largest segment there is — the people who will never see themselves in anyone's standard range. It moved from a footnote to something that needs to be prominent and well-explained.
That's the finding I'd have missed by only reading the top of the table.
The risk that only shows up if you look
Buried in the same research: some subjects in this community are closed — created and owned by individuals, with rules about commercial use.
Selling a commercial miniature based on one without permission wouldn't just be a legal exposure. It would be a reputational one, in the exact community you're asking to fund you, on the day you're most visible. The worst possible time to find out.
So the rule went into the project's constraints before any design work started: the base line-up uses only open, natural subjects. Closed ones need verified licensing or they don't happen.
Nobody would have caught that from taste. It came out of reading the sources properly.
Honest about what the data isn't
The write-up carries its own caveats, because a finding without limits is a liability:
- The percentages measure which subjects people identify with — not intent to buy a miniature. Those correlate. They are not the same thing, and I'm not going to pretend the survey answers a question it never asked.
- The data is from 2020. The top of the ranking has been stable since, which is worth something, but it's still not this year.
- Real purchase intent needs a different exercise: benchmarking comparable campaigns. That's a separate job, and it's queued.
The method, which is the actual point
This is the pattern I now run before committing production time:
- Find a primary source, and rank sources by quality — peer-reviewed above aggregator, aggregator above forum consensus, forum consensus above my own instinct.
- Separate confirmed fact from inference, in writing, and mark which is which.
- State what the data can't tell you, in the same document, at the same size.
- Let it overrule you. Research you only accept when it agrees isn't research.
Every one of those steps is a habit borrowed from how I run the workspace — the same rules about citing sources and separating fact from inference that stop an AI system from inventing a confident answer turn out to stop me from doing it too.
Step four is the hard one. This one cost me a set of archetypes I liked. It saved the weeks I'd have spent sculpting them.