The previous piece was about getting data in. This one is about getting it back out, which is the same problem viewed from the exit and treated with roughly a tenth of the care. Import gets attention because it stands between a prospect and using the product. Export gets built late, usually as a CSV button somebody added during a sales cycle, and then quietly stays that way for years.

That asymmetry is a mistake, and not for sentimental reasons. Export is what buyers ask about when they’re deciding whether to commit, because the answer tells them what happens if this goes badly. A product that can hand back everything you put into it, in a form another system can read, is easy to say yes to. One where the answer is vague makes the decision feel permanent, and permanent decisions get postponed. The team’s instinct — make it hard to leave and fewer people will — has the causality backwards: making it hard to leave mostly makes it hard to arrive.

There’s also a practical reason, which is that export is the feature people use while staying. They pull data to build a report the product doesn’t offer, to hand a file to an auditor, to run analysis in a tool their team already knows. Every one of those is a customer working around a gap, which — as the internal-tooling thread argued about spreadsheets — is a specification you’re being handed for free. But it only works if the export is usable: the CSV with all the columns, not the eight somebody picked in 2023; the full history, not the last thirty days; the identifiers that let a row be matched back to something rather than a display name that isn’t unique.

The gap between “we have an export” and “the export is complete” is where most of the disappointment lives. Attachments and uploaded files often aren’t included. Neither are the things stored as relationships — comments, history, who changed what — because they don’t fit a flat file and nobody decided what to do about that. The result is an export that technically exists and reconstructs maybe sixty percent of what the customer believes they own. Worth knowing which sixty percent, and worth saying so out loud rather than letting someone discover it on the day they need all of it.

At size, this also stops being an interactive feature. A dataset that takes four minutes to assemble can’t live behind a button that holds a browser request open, and the honest version is the queue thread’s answer: accept the request, do the work in the background, tell the person when it’s ready, and hand them something they can download once. That’s more machinery than a button, but the alternative is an export that works in demos and times out for exactly the customers with the most data — which is to say the ones who matter most.

The version of this I’d defend is simple: whatever a customer can put into the product, they should be able to get back out, in a documented format, without asking anyone for help. It costs a few weeks. It buys the ability to answer the hardest question in a sales conversation with one sentence, and it means the people who stay do it because the product is good rather than because the door is stuck.