Improving Signup Without a Rearchitecture

Big Cartel's sign-up flow required a shop subdomain, a shop name, an email, and a password. The subdomain was uniquely awkward for a store platform: your subdomain is also your login for Big Cartel, not your email — a twenty-year-old architectural decision we still live with today. Sellers who didn't have a shop name ready, or who kept hitting "subdomain taken," dropped out of signup.

The subdomain-taken problem was worse than it should have been because Big Cartel historically never deleted accounts except for legal reasons. Old, dormant, unpaid accounts sat there for years, locking up desirable subdomains — things as generic as soap.bigcartel.com — that new sellers wanted.

We made these changes gradually and deliberately between 2023 and 2026, watching the metrics at every step. But in 2026, after social login shipped, Customer Support started flagging a spike in reports. Around the same time, Voice of Customer submissions mentioning account or login problems rose, and we saw a rise in the number of accounts being created under the same email address in a short window.

Once we saw the spike, the root cause wasn't really a mystery to name at a high level: a Big Cartel account's login is the shop subdomain, not an email address or username, a decision made twenty years ago that the whole system was still built around.

One faction of the team pushed hard to use this moment to fully re-architect signup and sign-in around email as the account identifier. That is probably the more correct long-term fix. But it wasn't simple: we had millions of existing accounts built on the old system, plus an already-established baseline of sellers who intentionally run multiple shops under one email. Rebuilding login around email would mean building an account switcher so those sellers could move between shops, plus a real migration path for every legacy account, and it would take weeks if not months to do properly.

At the time, the team's priority was paid subscriber growth, and we were mid-flight on a set of other experiments in the same area of the product. So the harder, more correct fix went on the roadmap, and I went looking for what we could do immediately.

The initial theory was that something in the social login handoff was creating orphaned records. Rather than accept that, I went and read the actual Voice of Customer entries.

One seller's complaint was that when they tried to login they got the error "shop does not exist." That phrasing was specific enough to reproduce. I opened the sign-in form, typed an email address into the shop URL field, and got back exactly that error. That was the smoking gun: the sign-in flow still asked for a shop URL that sellers no longer consciously set or remembered, because signup had stopped asking for it.

Walking the flow surfaced a second gap. Forgot Shop Password required already knowing your shop URL — and because URLs were now auto-generated rather than chosen, many sellers had never registered what theirs was, even after receiving it by email. There was also no "Forgot Shop URL" flow at all. A seller who couldn't recall their URL had no supported path back into their own account, so creating a new one looked like the only option.

The engineer who had proposed the account switcher pushed back hard on the lighter-weight approach. I built the case to leadership that, given current roadmap priorities, the incremental fix was the right call — and that it bought us time to design the account-switcher migration properly later rather than rushing it under pressure from an active incident.

My designer used AI tooling to move fast on the shop-URL detection fix specifically, prototyping and helping ship a production-ready version within about an hour. That speed is part of why the team could resolve this in days rather than committing to the multi-week rebuild path.