We automated a SaaS directory — here's what we learned

We expected instant publishing to be a convenience. It turned out to be the product. When a founder can submit and see their tool live — with its own indexable page, favicon, and category placement — in under two minutes, the directory stops being a 'channel you try' and becomes part of the launch checklist.

The queue was never protecting quality; it was protecting scarcity. Removing it didn't flood the catalog with spam, because the other systems (verification, ranking) do the filtering the queue pretended to do.

The thing manual directories get most wrong isn't curation — it's maintenance. Catalogs accumulate abandoned products, expired domains, and tools that pivoted away. Every dead listing is a small trust tax on every visitor who clicks it.

Automating this was the highest-leverage decision we made: a scheduled job that re-verifies every domain weekly and flags dead products for removal. It's unglamorous infrastructure, but it's what makes 'every link works' a property of the site instead of a lucky coincidence.

Our takeaway: verification is not a one-time gate. It's a continuous process or it doesn't exist.

Ranking listings by real signals — clicks, upvotes, recency — produced better ordering than any human picks we tested it against, for an obvious reason: it reflects what visitors actually click, not what an editor guesses they'll click.

The catch: engagement data has a cold-start problem and a rich-get-richer tilt. A brand-new listing has no clicks to rank on, and a top-ranked listing attracts more clicks. The fixes were structural: every new launch gets a visibility window, and personalization diversifies which aisles each visitor sees, so the same five tools can't hog every feed.

The lesson we didn't fully expect: once ranking is automated, the marginal gain isn't a better global order — it's a different order per visitor. A founder shipping dev tools and a founder looking for design software should not see the same homepage, and a static directory has no mechanism for that.

Watching taste signals reshape the browse page — category clicks weighting aisles, searches refining matches — made it clear the directory's job isn't to present the catalog. It's to route the right slice of the catalog to each person.

The biggest mistake we could have made was automating everything, including the parts that need a person. Verification, liveness, ranking, matching — software wins at all of them. But the narrative layer (why a category matters, how to think about a tool, honest writing like this post) is still human work, and it reads differently because it is.

That's the actual lesson of the whole project: the old directories weren't manual because they cared. They were manual because nobody built the plumbing. Once the plumbing exists, the human layer gets to be human instead of pretending to be a queue.

Start with verification, not curation. Instant publishing plus continuous checking beats a review queue on every axis that matters to users.

Rank on real engagement from day one — even thin data beats opinion — and design explicitly for cold start and rich-get-richer, because both will show up by week two.

And instrument everything honestly. We show real click counts, not inflated 'views' — because a directory that lies about its own numbers can't be trusted to rank anyone else's product.