Back to all articles
APIs as a ProductPart 09
Rate limitingAPI productsAPI monetisationPlatform strategy

APIs as a Product, Part 9: Access, Rate Limits and Pricing

Janahan Thavendran3 min read

TL;DR

  • Keep the sandbox open and make production access earned through certification and a commercial agreement.
  • Set rate limits per partner and per type of call so orders and payments are protected before bulk work, and make limits visible to partners.
  • Charge for an API when usage costs money without creating revenue; decide with infrastructure cost, support time and partner revenue on the same page.
In this article

APIs as a Product, Part 9 of 10 · Running an API as a business.

Picture lunchtime on a busy day. A partner starts a full catalog refresh for every merchant it serves. It is a perfectly valid use of the API. It is also happening at the exact moment diners are placing orders, on the same platform, through the same endpoints. Order confirmations slow down for everyone else.

Nobody has done anything wrong. The business just hasn't decided whose traffic matters more at noon.

That is the moment an API stops being a technical asset and becomes a business you have to run. Three questions sit underneath it: who gets in, how much they can use, and whether they pay.

Who gets access

We split access into two doors. The sandbox was open: any partner could get a key, read the docs and start building without a meeting. Production was earned: a partner went live after passing certification and signing a commercial agreement.

That split did two jobs. It kept the first hour friction-free, which is when developers decide whether you are worth the effort. And it meant that every partner touching real merchants and real money had shown they could handle failures, not just the happy path.

Limits are a product decision

Seen that way, rate limits stop being a single number set by engineering. They become a set of rules about what the platform protects first.

Order and payment calls come first, because a slow order is a lost sale for a merchant and a hungry customer. Bulk catalog work comes second, pushed towards quieter hours or paced so it can't crowd out orders. Limits are set per partner and per type of call, not as one global ceiling, so one busy partner can't slow down the rest.

The other half is telling partners what the limits are. A limit they discover from a wall of errors feels like an outage. A limit that is documented, returned in response headers and visible in the portal is just a rule they can design around.

Whether to charge

The answer depends on what the API is for.

Our API existed to bring orders to merchants. Every partner who integrated was a new sales channel, and the business earned from those orders. For an API like that, charging partners for access slows adoption of the very thing that makes the business money.

Charging makes more sense when usage costs you money without creating revenue: heavy data exports, high-volume reads, or a guaranteed level of service that needs its own capacity. In those cases, a price is a way to make the trade-off visible, not just a way to earn.

Whatever you decide, decide it with the P&L open. Infrastructure cost, support time and the revenue each partner brings in belong in the same conversation.

A rate limit is a pricing decision made in engineering. The only mistake is making it by accident.


Related reading: per-tenant quotas and a shedding order for when everything arrives at once are in Platform at Scale, Part 11: One Tenant Can't Sink the Ship.