Never flying blind: analytics, replay and long-term logs

We couldn't see what a stuck customer was doing, and our server logs had already vanished by the time we looked. This lays out what to add so that never happens again.

2026-09-15 · proposal, not a build ticket · related: Microsoft connect flow · Western Slope go-live tracker — mailbox RCA

On this page

  1. What happened
  2. What Vercel offers today
  3. The three layers and the tools
  4. Best practices we'll follow
  5. Options and recommendation
  6. Decisions for Fede
  7. First things to instrument
  8. Sources

What happened

On September 15, a Western Slope leasing agent got stuck on a screen in our Microsoft connect flow (see the linked page for the flow itself). We had no way to see what she'd actually clicked — no recording of her screen, no record of which button she pressed. We also had no product analytics telling us how many people hit that same screen before her. By the time we went looking, the server logs covering that window had already expired, hours later. Meanwhile our own admin page told us the connection was "granted" when it wasn't actually usable. Fede: "we can't be flying blind again."

What Vercel offers today

We host on Vercel. Here's what it gives us out of the box, and what it doesn't.

ToolWhat it doesLimit / cost at our planHow far back it goes
Web AnalyticsPage views and custom events from the browserPro plan: 25,000 events/month included, then $3 per 100,000~30 days
Speed InsightsPage load speed scoresIncluded on Pro~30 days
Runtime logsServer request/error logsHobby: 1 hour. Pro: 1 day. Add-on "Observability Plus": 30 daysSee limit column
Log DrainsStreams every log line to an outside address we controlFree on Pro planForever — wherever we send it
Session replay (watching what a user actually did)Vercel doesn't offer this at all

Bottom line: Vercel can hold on to logs a little longer if we pay for the add-on, and it can hand us every log line for free via a drain — but it will never show us a recording of what a customer's screen looked like. That has to come from somewhere else.

The three layers and the tools

Seeing what happened requires three separate layers. No single vendor covers all three.

LayerWhat it answersOptionsCost at our sizeHow long it's kept
Product analytics"How many people reached this screen, and how many got stuck?"PostHog, Mixpanel, AmplitudePostHog: free up to 1 million events/month. Mixpanel: roughly $450/month at our size. Amplitude: roughly $600+/monthPostHog keeps 3–7 years (or we self-host it and keep it forever)
Session replay"What did her screen actually show, click by click?"Sentry Replay (we already use Sentry), PostHog Replay, Microsoft Clarity, LogRocketSentry: 500 free, then roughly $30–50/month. PostHog: 5,000/month free. Clarity: free. LogRocket: $99+/monthSentry & Clarity: 30–90 days. PostHog: 3 years
Long-term logs"What did our server actually do, and can we still see it a year later?"Vercel's log drain pointed at our own storage (AWS S3 + CloudWatch)Roughly $10/month~400 days

A receiver for the log drain is already being built (PR in flight) — that's the piece that lets us point Vercel's free log drain at our own AWS storage instead of paying for Vercel's short-lived add-on.

Best practices we'll follow

Options and recommendation

OptionToolsCost nowWhat it misses
A — Lean (recommended)PostHog for analytics, PostHog or Sentry for replay, log drain to our own AWS storage~$0–40/monthNothing major at our size
B — Freemium mixMicrosoft Clarity for replay, PostHog for analytics, Sentry, log drain to AWS~$0–80/monthClarity can't build funnels and only keeps 90 days
C — Sentry-onlySentry for replay, Vercel Web Analytics, Vercel's Observability Plus add-onSimilar cost, fewer vendors to manageOnly 30 days of logs; weak at answering "how many people dropped off here"

Recommendation: Option A. One tool (PostHog) can cover both the "how many people got stuck" question and the "what did her screen show" question, plus our own long-term log storage — for close to nothing at our current volume.

Decisions for Fede

  1. Which option? A, B, or C. Recommendation: A.
  2. Which tool records the screen? PostHog or Sentry. Recommendation: PostHog if we go with A, so one tool covers both drop-off counting and screen recording. Sentry is fine too, if you'd rather not add a new vendor.
  3. OK to create the PostHog account and production key? This is a new third-party sign-up, so it needs your go. Recommendation: yes.
  4. OK to point Vercel's log drain at our own AWS storage once that receiver's PR merges? This is a production settings change on Vercel, so it needs your go. Recommendation: yes.

First things to instrument

Once a tool is picked, these are the first places we'd add tracking — each one tagged with which user and which company it belongs to:

Sources

PropFlow Docs