Back to all articles
APIs as a ProductPart 01
API productsDeveloper experiencePartner integrationsPlatform strategy

APIs as a Product, Part 1: Your API Is a Product With Users You Never Meet

Janahan Thavendran4 min read

TL;DR

  • An API has users who rarely tell you anything: partner developers, partner business owners, end customers and, easily forgotten, your own internal teams.
  • Treating it as a product means real discovery with partner developers and asking whether a request helps the next ten partners or only this one.
  • Measure time to first call, time to go-live, integrations still active after three months and partner volume; at Delivery Hero, API adoption tripled and onboarding went from weeks to days.
In this article

APIs as a Product, Part 1 of 10 · Partners, developers and your own teams are users, even when you never meet them.

Picture the developer at a bank who builds the integration that puts your merchants' menus inside the bank's app and sends their orders back to you. You will probably never speak to them. You won't know their name, what they found hard, what they worked around, or what they gave up on. Their code carries real orders into real kitchens, and your only signal about their experience is whether your error logs stay quiet.

That is the strange thing about an API. It has users, often more than the app sitting on top of it. But they rarely open a ticket to say something is confusing. They read your docs late at night, try a few calls, and either push on or quietly pick a different provider. The feedback loop most product teams rely on simply isn't there.

On the platform I worked on, we eventually had more than 50 global and local partners syncing catalog, orders, loyalty and payments through our middleware and open API. Google Pay in Singapore and a bank's mobile app were among them, showing merchant catalogs and sending orders back. Each partner had at least three kinds of user, and they wanted different things.

The developer who integrates wants clear contracts, a sandbox that behaves like production, and errors that say what went wrong. The partner's business owner wants to go live by a date they promised someone else, and wants to know what it will cost them in effort. And the end customer, the person ordering lunch inside a bank app, wants it to work and never thinks about us at all.

There was a fourth group I didn't think of as users at first: our own teams. Loyalty, menus, transactions, catalog and payments were each built by different people. Before we had a common core API layer, every product talked to the others in its own way: point-to-point connections, the same logic written twice, no shared contract. An internal team calling the menu API had its own release cycle and its own deadlines, exactly like a partner. Sitting in the same building didn't mean I knew how they used it.

When the API is treated as plumbing, only the first group gets any attention, and usually only when something breaks. When it is treated as a product, all four get a seat at the table.

What changed when we treated it as a product

Discovery became a real activity. Before each new integration, we spoke to the partner's developers about how their systems handled orders, menus and payments. Most of what we learned never appeared in a requirements document. One partner couldn't accept two orders at the same time. Details like that shaped our roadmap more than any feature request did.

The roadmap had to balance two pulls. Large partners asked for custom behaviour, and saying yes once usually meant maintaining it forever. We learned to ask whether a request would help the next ten partners or only this one. If it was only this one, it went into configuration or into the partner's own code, not into the core contract.

And we needed success measures that weren't just uptime. The ones that mattered were how long a new partner took to make a first successful call, how long they took to go live, how many integrations stayed active after three months, and how much order volume moved through partner channels. When I later owned the developer platform at Delivery Hero, those were the measures we worked to. Over that work, API adoption tripled, developer NPS rose 40 points and partner onboarding went from weeks to days.

None of these show up if nobody owns the API as a product. Engineering will keep it running. But keeping it running and making it worth integrating with are different jobs.

The partners who built on our API were customers. So were the teams down the corridor. Both paid us in effort, not invoices, and both could quietly route around us without telling us why.