Running Statsig?

Statsig pulse or layers acting up? We unstick it.

Statsig changed hands twice in two years, and the people who set up your experiment and metric pipelines may well have left. We pick up the pulse metrics that will never converge, the layer conflicts that are quietly double-counting users, and the SDK bootstraps that take long enough to break the user's first impression.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

statsig-events.log

still broken

What it is

Statsig at a glance

Statsig is a feature-flag, experimentation and product-analytics platform founded in 2021 by former Facebook engineers. Its distinctive pitch has been Pulse: a system that attaches every metric movement to the experiment that caused it, so teams can see downstream effects alongside the headline metric.

OpenAI acquired Statsig in October 2025 and announced that Statsig would remain an independent product. In 2026 it was reported that Statsig became part of Amplitude. Through both transitions the core product has kept working, but teams that invested heavily in the platform are watching the roadmap carefully, and some of the integrations behave differently than they did a year ago.

Type
Feature flags, experimentation and product analytics
First released
2021
License
Proprietary SaaS, MAU-based pricing
Maintained by
Statsig, now within Amplitude
Typically used for
Feature gating, A/B tests, layered experiments, pulse metric analysis, product analytics

The honest version

Statsig pros and cons

No technology is all upside. This is where Statsig earns its place and where it tends to cause trouble.

Where Statsig is strong

  • Pulse is a genuinely different ideaAttaching every metric movement in the product to the experiment that probably caused it saves a lot of 'what changed?' meetings.
  • Free tier is generousThe free tier covers a lot of real usage, which is unusual in the experimentation space and makes it easy to adopt.
  • Layers avoid overlap errorsMutually-exclusive experiment groups through Layers keep two concurrent tests from stepping on each other's traffic, when configured correctly.
  • SDK coverage is goodOfficial SDKs for the major server and client languages, with a sensible API surface and reasonable defaults.
  • Dynamic Configs cover remote-tuningBeyond on-off flags, Dynamic Configs let you push JSON-shaped parameters to clients without a deploy.
  • Holdouts and auto-tune reduce manual workHoldout groups give you long-run effect tracking, and the auto-tune features aim at the right variant without a human in the loop.

Where Statsig bites

  • Ownership keeps changingOpenAI in 2025 and Amplitude in 2026. The product survives each move, but roadmaps, pricing and support contact paths do not land the same each time.
  • Pulse reads need interpretationPulse attributes correlations, not causation. A dashboard full of green arrows next to a release can still be noise.
  • Layers trip up newcomersThe rules that govern which tests can coexist are strict, and a misconfiguration silently removes a test from traffic rather than throwing an error.
  • Bootstrap latency hurts first-paintFetching config before the UI renders adds hundreds of milliseconds unless you ship a bootstrap payload server-side.
  • Smaller community than LaunchDarklyFewer third-party integrations, fewer Stack Overflow answers, and more need to read source code when the docs are thin.
  • Billing and MAU boundaries are fuzzyThe MAU count depends on how you call initialize, and overcounting can push a free tier user into a paid conversation with little warning.

Sound familiar?

Common Statsig problems we fix

The Statsig problems that bring people to us, and what you have once each one is fixed.

initialize timeout after 3000ms

SDK initialization is slow or failing

The first paint is delayed because the SDK is fetching config over the network before rendering. A dropped request falls back to defaults, often with the wrong variant for the user.

Bootstrapped config, snappy first paint

metric 'checkout_conversion' insufficient power

Pulse metrics will not converge

A metric with a small effect size and a thin traffic cohort runs for weeks without a signal, and the team starts calling results on noise.

Metric or sample sized honestly

Layer conflict experiment overlaps

Layers are double-counting exposures

Two experiments are running in overlapping segments and the system quietly removes users from one, which skews both reads.

Layers configured, exposures unique

DynamicConfig returned null

Dynamic Config fallback is masking a real issue

The client could not fetch the config and quietly returned the default; the feature that was supposed to launch never did.

Config delivery reliable

Exposure logging dropped 1,284 events

Exposure events are being dropped at the edge

The logging queue filled during a traffic spike, exposures were never recorded, and the resulting metric report undercounts the experiment reach.

Exposure pipeline keeping up

p-value stuck

Experiments plateau without a clear decision

A metric hovers near significance for days. Teams push harder on optics than on the underlying math. We revisit the metric, the guardrails and the planned sample size.

Decision supported by the data

user object missing stableID

User identity is inconsistent across sessions

The stableID is not being provided or not persisted across the auth boundary, so the same person is counted as multiple users in exposure data.

Identity consistent across sessions

What we do

Statsig help, from Pulse to Layers

Call us for a single stuck experiment, or hand us the whole analytics practice.

Rescue
Pulse readings that nobody trusts, Layer conflicts that are hiding real traffic, or a bootstrap pipeline that broke after the Amplitude integration. We untangle the live incident and then address the root cause.
Experiment design
Metric definition, guardrail selection, sample-size planning and minimum run length, so the result that comes back is one you can act on.
SDK integration
Server-side bootstrap payloads, exposure logging that survives traffic spikes, and identity that stays consistent across the auth boundary.
Layer and segment hygiene
Mapping overlapping experiments into mutually-exclusive Layers, reviewing segments for staleness, and documenting what each one targets.
Platform transition support
Helping teams understand what changed with the Amplitude acquisition, which features have moved, and how billing has shifted.
Managed Tech
An ongoing engagement that watches the Statsig console, keeps the experiment hygiene tight, and reviews each major test before it is called.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running Statsig alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

Statsig questions

Something else? Call+1 737 727 7327

What happened to Statsig after the OpenAI and Amplitude moves?

OpenAI acquired Statsig in October 2025 with a commitment to keep it independent, and in 2026 reports surfaced that Statsig was folded into Amplitude. The product is still shipping, but roadmap items and support contacts have moved. We help teams map what they used to rely on to what the current platform actually offers.

Why are our Statsig experiments never reaching significance?

Usually a combination of small traffic, a noisy metric, and a target effect that is smaller than the metric's natural variance. We revisit the hypothesis, pick a less noisy guardrail, compute the required sample size up front, and commit to running the test for a minimum duration before anyone reads the dashboard.

Why does our Statsig SDK slow down first paint?

Because the client is fetching config before it renders. The remedy is to send a bootstrap payload from your server on the initial response so the client starts with a filled cache. We wire that up and measure the before and after so the gain is real.

How do Layers actually work?

Layers are groups of experiments that cannot run on the same user at the same time. Each user gets assigned to one experiment per layer, which prevents two tests from interacting in weird ways. The trap is that misconfigured layers silently remove users from a test rather than warning you, so we audit the layer setup before every launch.

Should we migrate off Statsig while ownership is still settling?

Only if the current product gaps are actively blocking you. Migrations between experimentation platforms are hard, metric histories do not transfer cleanly, and the Amplitude home may well be stable for years. We model the real cost of staying versus leaving and decide on numbers rather than mood.

How much does Statsig help cost?

Estimates come at no charge. Share the experiment or SDK symptom, and we quote a fixed scope before anyone touches the console.

Free estimate

Statsig giving you trouble?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.