I tried three plays before I found the real problem

Solid Remote Jobs, my side-project job board, is on its fourth strategy in seven months. Three of the first four died on contact with real numbers. I want to walk through all of them, including the one I’m in the middle of right now — because the one I’m in the middle of might die too, and I said in writing what that would look like before I knew the answer.

Play one: ask nicely

The first plan was the obvious one. Email companies, offer to post their remote dev roles for free, build up inventory. I sent 62 cold emails between March and May. One company said yes. The listing went live. A month later, they stopped answering. Listing down.

Sixty-two emails for one listing that lasted a month is not a pipeline. It’s a coincidence with a sales deck.

Play two: flip to the other side of the market

Same June, different idea: stop chasing companies, go straight to developers instead. Build a talent page, let devs list themselves, let companies browse. I shipped it — a /talent page and a /join-talent page, live, real.

I reversed it the same day. Sourcing developers is a different business: high-touch, doesn’t compound, and it quietly turns a job board into a recruiting agency, which is a different company with different economics and a conflict with other work I already do. I’d built the thing before I’d asked whether I wanted to run the thing. Shipping fast caught the mistake fast, at least.

Play three: curate and wait for search

Plan three was patience. Pick a wedge — remote AI/ML engineering, where salaries are high enough that screening eventually has value — and hand-curate ten genuinely good, salary-transparent listings from public sources. Build the SEO page. Let Google send people.

Here’s what four weeks of that actually produced, measured: 31 new users, 43 pageviews, zero week-one retention. The traffic breakdown tells the real story — 15 direct, 14 from one LinkedIn post, 1 from organic search, 1 referral. One LinkedIn post about reviewing 200 job listings pulled 1,996 impressions and 38 clicks, roughly 20 times every other channel combined. Search contributed one visit in a month. For an SEO play, that’s not slow traction. That’s no traction.

The deeper problem wasn’t traffic, though — it was the offer. When someone applies to a listing I aggregated from somewhere else, they’re applying through someone else’s system. I can’t screen them, can’t add anything, can’t tell the company a single thing it didn’t already have. Ten good listings with nothing behind them is a directory with extra steps.

The relaunch: sell the evaluation, not the posting

So in August I deleted the aggregated listings and rebuilt around the actual problem. Companies hiring engineers aren’t short on applicants — they’re buried in them. Nobody’s offering to help with the part that’s actually hard: deciding which of two hundred resumes is worth an hour of someone’s time.

The new model: companies post a role directly, free, salary range required, no exceptions. Every applicant gets a one-page scouting brief — identity, key signals, red flags. The company’s top three get a full report: a six-dimension rated scouting card, a career timeline, a verdict in plain language, no committee-speak. It’s powered by VettLab, a candidate-intelligence product I also run, and the pitch to companies is straightforward: over a thousand dollars of candidate research, free, because you posted a role. Solid Remote Jobs feeds VettLab leads; VettLab is the reason Solid Remote Jobs has something worth posting about. One hard line survived from the talent-page mistake: we only research people who already applied. No sourcing, no outreach to candidates who didn’t ask to be found — that’s a different business, and I already know how that one ends.

The bet I wrote down before I ran it

By September the board had grown, slowly, to 25 followers on its LinkedIn page. Instead of guessing whether content alone was working, I wrote the test down first: three posts a week, every week, for four weeks, zero cold outreach. And I wrote down, in advance, what success and failure looked like — 60-plus followers, a hundred-plus site visits a month, and at least one real company inquiry. One inquiry was the whole bar. If it came in at zero, the verdict was already decided: the bet failed, and outreach comes back.

Four weeks later: 10 of the planned 12 posts shipped (the last week broke — nothing went out for most of a week). Followers went from 25 to 28, well under the 60 target. Site traffic held up reasonably — around 25 visitors a week, better than the old 31-a-month baseline. And the number that mattered most: zero company inquiries. Several hundred people saw the offer. Nobody wrote in.

Along the way the posting data taught me something I didn’t expect, and it cut against what I’d learned back in the SEO phase. Back then, a data-driven post had been the best performer by far. This time, the pattern flipped hard: posts written as a first-person opinion or a real story — the relaunch story, a pointed take on hiring economics — pulled hundreds of impressions each, while straight data dumps, product pitches, and a poll barely moved. Attention follows a person saying something with conviction, not a chart. That’s worth remembering on its own, independent of whether the underlying bet worked.

Honoring a result I didn’t want

Zero inquiries meant the bet failed by its own rules. The tempting move at that point is obvious: find a reason the test wasn’t fair, extend it another month, keep the thing I enjoyed doing (writing) instead of the thing I’d apparently been avoiding (outreach). I’d written the terms down specifically so that temptation wouldn’t win. A number you agree to in advance, before you know the answer, is the only kind of number that can actually tell you you’re wrong. A number you accept after you’ve seen it is just a story you’ve already decided to believe.

So: the content keeps running, in the voice that was actually working. And warm outreach restarts — starting with the people who’d already engaged with the posts, then companies hiring the exact roles the board is built for, found through the same public “who’s hiring” threads where this whole thing started back in March.

Four plays in. Nothing is proven yet. But I know a lot more than I did seven months ago about which parts of this problem are real, and I have a much shorter list of ways to be wrong.