How do you measure the performance and impact of your API integrations?

In terms of metrics, there should be business outcomes at the root of the API strategy. They may take time to develop, but should show positive signals as the platform is developed and validated. The metrics noted so far are good performance metrics, most are expected, but do not necessarily signal a valuable asset or business differentiation.

The cost of API calls and other resource usage is always relevant. Integration involves cobbling together software components each of which may charge a cost for a set of services, with rate of API calls being one typical measure. Want a higher rate, the cost goes up. Often these costs aren't linear with respect to pricing, but instead are tiered use and lumped in with other services. This section mentions performance, though measures like reasonable throughput caps and acceptable data latency should also be considered early in the design process.

re:mart needs to source information from retailer websites in order to operate, syncing the prices of the hundreds of thousands of products sold in your area. This is extremely brittle and slow, and so we do it it after hours. Since price data doesn't change all that often, we can trust that the price at 2am will be the price at 2pm. And about this being brittle, if ever a store changes their website, which they do often, our code can break. So we need to have monitoring, measurement, and retry strategies for every piece of data. My advice is not all data has the same value, and you may not need to know it at the times you think you do, so delay and prioritize computation based on how vital each external API is to your company.

Many other examples of tools exist, these are ok commercial examples. Tools are important, some from a development side, then for functional testing- most of that should be automated during development. Then tracking how the system operates is more dev ops, but critical. Then managing how the API is impacting business can be more lagging or need additional data, but should be most important thing to measure.

Many integrations can be visualized as one of many worker nodes connected to a central system of record. A poorly implemented integration can consume system of record resources to the point that other integrations are blocked from performing far more critical tasks. Be sure to alert the admins whenever one integration consumes a) noticeably more resources than the baseline, as well as b) more than its fair share of resources.

Some of these tests should be done and provided by the API supplying platform and available. Having processes to take action on poor API performance or outages is also key point of customer service. It’s not if API fails, but when.