Working · 2026-11PrimaryGrowth opsProcess

High-tempo growth testing

Run many small, scored experiments every week from a standing backlog instead of a few big launches per quarter.

Also called: experiment velocity, weekly test cadence

Claim and evidence

The claimThey got it to 10 tests per week.

Twitter · saastr.com · undated

What the numbers show

10/week

testing pace Twitter's growth team reached, per Sean Ellis's talk (anecdotal)

saastr.com
514

run by Fyxer's growth engineers, more than two per work day (as reported to Growth Unhinged)

growthunhinged.com
up to 10/week

run by Popsa (as reported by First Round)

review.firstround.com

Why it works

Diagram, not data
Backlog of ideasScore and rankLaunch and reviewweeklyLog learning backinto backlog
Backlog ofideasScore andrankLaunch andreview weeklyLog learningback intobacklog

Growth is found by trial; more shots on goal per week means faster learning and more compounding wins. A backlog plus a fixed weekly rhythm removes the stalls between ideas, so the team learns from losses as well as wins.

How to run it

  1. Build a backlog of ideas from data, research and team input
  2. Score and rank ideas (ICE, RICE or similar)
  3. Commit to a weekly cadence of launches and reviews
  4. Assign an owner to every test
  5. Log the result and learning, then feed it back into the backlog

Ways to run it

Design options to test, not promised results.

  1. Option AWeekly launch target

    Commit to a set number of launches each week and review them in one standing meeting, as Twitter's growth team did.

  2. Option BAI-assisted analysis

    Let engineers use AI for data analysis so each test needs fewer hand-offs, as Fyxer did.

  3. Option CSmall tests by default

    Favour the smallest test that answers the question so the weekly rhythm holds.

Timeline

  1. 12016 first documented: Sean Ellis, SaaStr Annual talk (year per transcript page text)
  2. 22016 Twitter — Growth team raised its testing pace as described by Sean Ellis in a conference talk
  3. 32026 Fyxer — Growth engineers ran 514 experiments in a year, with AI help for data analysis
  4. 42026-11 status checked: Working

Dated examples

  1. 2026

    Fyxer

    Growth engineers ran 514 experiments in a year, with AI help for data analysis

    “Their secret: running 514 (!) experiments, more than two per work day.”
    ✓ quote checked word-for-word on the page

    growthunhinged.com · 2026-03-04

  2. 2016

    Twitter

    Growth team raised its testing pace as described by Sean Ellis in a conference talk

    “They got it to 10 tests per week.”
    ✓ quote checked word-for-word on the page

    saastr.com · undated

  3. —

    Popsa

    Reported by First Round as running up to 10 growth experiments a week

    “This rapid experimentation continues, with Popsa running up to 10 growth experiments weekly.”
    ✓ quote checked word-for-word on the page

    review.firstround.com · undated

Does it still work?

Workingas of 2026-11

Still the default operating model; AI-assisted teams push the volume higher (Fyxer 2026). Testing speed alone does not fix weak hypotheses. source

Risks

  • Volume without quality hypotheses produces noise and false positives
  • Needs enough traffic; low-traffic products cannot read many tests

Sources

  1. saastr.com · SaaStr · Gretchen DeKnikker · undated
    “The goal of the growth team is to test at a high tempo.”
    ✓ quote checked word-for-word on the page
  2. growthunhinged.com · Growth Unhinged · Kyle Poyar · 2026-03-04
    “Their secret: running 514 (!) experiments, more than two per work day.”
    ✓ quote checked word-for-word on the page
  3. review.firstround.com · First Round Review · undated

Change log

  • 2026-11-20 — harvest v0.1 (ops-ai)