Add p50 & p95 latencies to docs

[vivekjain1397]

@tjulien @gianiek what do we think about exposing the latencies per endpoint in our docs? @tjulien I rounded down across the board based on the metrics you had shared. Few of the p95's that stick out:

• Autocomplete (shared: 486ms, docs 300ms) - Took a look at a few of our accounts that've been live in production (Michaels, VSCO, Bevmo) etc and noticed ~250-400 so I think 300ms is a valid avg to share.
• Forward (shared 728ms, docs 500) - This one was pretty variable across our accounts that are live, 500 seemed like a fair # to share.
• Validate (shared: 752ms, docs 700) - This one just seems like a high number to share but haven't heard customers bring that up as most other services have a fairly high latency as well.

Pros of sharing latency per endpoint:

• SE's get this question all the time, esp about geocoding & autocomplete. As we ramp-up more A/B tests, it would be good to point to our docs and set expectations.
• F/U q's that SE's usually get are related to optimization techniques to decrease latency, setting expectations can help with this.

Cons:

• We've heard that for geocoding/autocomplete, our latencies can be a bit higher than what customers are used to with Google/Loqate etc so might come off as a negative.

What?

Why?

How?

Screenshots (optional)

Anything Else? (optional)

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
documentation	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jan 28, 2025 4:04pm