Amplitude deep dives
Findry talks to Amplitude in both directions. Amplitude is the analytics source you measure outcomes against, and Amplitude is the analytics canvas you stamp Findry decisions onto. These four guides cover the full surface.
Contents
What you can do
One Amplitude connection per workspace unlocks every Amplitude-related surface in Findry. After connecting:
- Read events → outcome verdicts. Bind any Findry metric to an Amplitude custom event, saved chart, inline funnel, behavioral cohort, or baked-in template. The outcome-sweep cron pulls the metric, computes baseline + significance, and assigns a verdict.
- Push Findry events → Amplitude charts. Bet promoted, bet shipped, and outcome verdict each drop a labeled vertical line on every Amplitude chart in the project.
- Pick an Amplitude cohort in the bet's predicted-impact form to scope variance to that cohort.
- Link an Amplitude Experiment to a Findry Test for nightly state sync (no double entry between the rollout tool and the decision record).
- Use sum / avg / percentile aggregations for revenue, latency, and distribution-shaped metrics — not just count + unique-user.
Note: unlike PostHog, Amplitude doesn't ship a first-party survey product, so there's no auto-import of qualitative responses as Signals. If your team uses a survey tool alongside Amplitude (Sprig, Typeform, Delighted), Signals can still be created manually or by webhook — see the Signals guide.
The four guides
Read in order if you're new to the integration. Skip around once you're past Connect.
The closed loop
A single workflow that exercises every Amplitude surface, in order:
- A PM in Amplitude has a saved chart tracking weekly complete_checkout conversions for the mobile cohort.
- In Findry, the PM creates a metric "Mobile checkout completion" and maps it to that Amplitude saved chart by chart ID.
- The PM writes a hypothesis, picks the same Amplitude cohort in the predicted-impact editor, and promotes it to a Bet. Amplitude gets a "▲ Bet promoted" annotation immediately.
- An Amplitude Experiment is linked to the Bet's Test. The experiment goes live; nightly cron syncs its state into Findry.
- The Bet ships, the PM sets the ship date. Amplitude gets a "▲ Bet shipped" annotation at the ship date.
- 30 days later, the outcome-sweep cron pulls the chart from Amplitude, computes a +9.4pp lift with p=0.0008, and assigns verdict hit. Amplitude gets a "● Outcome verdict" annotation at the measurement timestamp.
- The PM scrolls the Amplitude complete_checkout chart. The spike lines up perfectly with the ▲ shipped annotation. The chain is closed both directions.
What is not yet
Four deliberate deferrals (vs PostHog parity):
- No first-party survey import. Amplitude doesn't ship a survey product; PostHog Surveys → Findry Signals doesn't translate. Use a separate survey tool + manual Signal entry.
- No multi-project routing on a single connection. Amplitude is one project per API-key-pair, so a Findry workspace that needs multiple Amplitude projects connects them separately.
- Cohort-segmented outcome measurement. Cohorts filter the bet's predicted impact today, not the post-ship measurement. Tier B, deferred until a customer asks.
- Releases mapping. Amplitude Releases are a structured deploy-marker concept; cleaner than annotations for shipping events but a different API surface. Annotations cover the use case for now.
Self-hosted Amplitude isn't supported either — only US Cloud (amplitude.com) and EU Cloud (analytics.eu.amplitude.com). Add the host as a free-text field if a self-hosted customer asks.