Guide

Shipping and specialties

Soft launch and percentage rollout

Make the feature available, learn from real load, and spend the marketing budget only after the path works.

105 of 128 · 2 min read

Between closed beta and a loud GA sits the soft launch: the feature is available to a growing slice of production traffic, or to anyone who finds it, but you are not spending the marketing budget yet. Combine this with deployment strategies — percentage rollouts, ring deployments, geo cuts — so blast radius stays bounded while you learn from real load.

The product rule is simple: do not buy attention until the path you are advertising actually works.

Concept

A beta is a named cohort you talk to. A soft launch is production traffic at growing share. Both reduce risk. Only the second proves load, edge cases and real mix.

Walking through a T2 web feature

A common sequence:

  1. Internal (staff flag) for a few days.
  2. 1% → 5% → 25% → 100% of eligible users over several days, watching error budgets and the activation metric after each step.
  3. Announcement only after 100% is healthy for 24–72 hours, or announce to a waitlist while the rest catch up.

Consumer apps sometimes soft-launch in one country first. B2B apps sometimes enable by cohort (new signups only, or one enterprise tier).

Example

You enable a new checkout step at 5%. Conversion drops 2 points and payment errors spike on one card brand. You roll back to 0% in ten minutes via flag, fix the edge case, then climb again. The blog post never mentioned a broken checkout. That is the whole point: separate availability from narrative.

Enable first, announce later

Decide in advance whether the technical enable and the public announce are the same moment. Enabling first, announcing later is safer. Announcing first, enabling later produces screenshots of 404s.

Write the order into the run-of-show and stick to it.

Common mistakes

  • Announcing at 5%. You are paying for traffic onto a path that is still moving.
  • Climbing percentages without watching the activation metric. Error rate can look fine while nobody finishes setup.
  • Flag forever. Calendar a GA decision date so the soft launch ends.

Check yourself

Checkout conversion drops 2 points at 5%. What do you do? Flag back to 0%, fix the edge case, then climb again. Do not announce yet.

Why enable before you announce? Availability can fail privately. Narrative cannot. Screenshots of 404s are expensive.

Key takeaways

  • A soft launch grows production share without spending the marketing budget.
  • Climb 1% → 5% → 25% → 100% and watch errors and activation at each step.
  • Enable first, announce later. Calendar a date when the flag must become GA.

Further learning

Pair this with a go/no-go list in launch checklists. Map of the whole topic: launching.

Launching products and features

Launch tiers, betas, checklists and measuring whether a launch worked.

Alpha and beta

A bounded group of users who find cliffs before you spend the marketing budget.

Deployment strategies and safe releases

Rolling, blue-green and canary deploys, automated rollback, and zero-downtime schema changes - ship often without betting the fleet on every commit.