An AI company that ships into live stores.

I have spent ten years building on Shopify. Themes first, then apps, then the integration work nobody volunteers for — the webhook that has to fire, the sync that keeps dropping events, the reconciliation script quietly holding an ops team together.

Across dozens of stores I kept finding the same four things. Cards declining and being retried once, badly, or not at all. A Klaviyo account with a welcome series from two years ago and nothing since. A cancel flow that was a platform setting rather than an interface. And somewhere in the back office, a person with a spreadsheet doing a machine’s job every Friday.

Every one of those leaks had a dollar figure attached. None of them were being worked on, because none of them are anyone’s job. The agency writes campaigns. The platform’s support desk answers tickets. The developer builds what is on the ticket.

So we stopped taking general Shopify work and started taking only this: brands doing $3M–$30M where the customer comes back — on a subscription, or on a reorder cycle you can predict — and where the gap between a good business and a great one is almost entirely the plumbing under the retention layer. Somewhere around $3M the same thing happens to all of them: a subscription platform, a 3PL, a second sales channel and eleven tools, and nobody left who can see across the whole of it.

The other thing ten years teaches you is what the work actually is. Almost none of it is marketing. It is integration: taking a fact out of one system, deciding something with it, and writing the result into another. A decline code and the retry window it argues for. A ticket and the order it is really about. A supplier’s PDF and next month’s purchase order. Nobody sells an app for the space between two systems, so that space gets filled with a person and a spreadsheet — and it stays that way, because it never breaks loudly enough to become anybody’s job.

Which is why we describe ourselves as an AI company now rather than a retention agency — and it is a narrower claim than it sounds. Reading an unstructured input and choosing an action is exactly the shape of work models got good at, and it is exactly the work we had been doing by hand across dozens of stores. What has not changed is everything around the model: the API client, the retries, the idempotency that stops a re-run sending a second purchase order, the credential that expires, and the boundary where a person still presses the button on anything touching money. That part is still software, and it is still most of the job.

That is the whole company.

— Founder, Pointerflow founder name and byline to confirm before launch