Plan a beta that tells you whether you are ready to launch
Six steps to run a beta that ends in a defensible go/no-go: launch questions, exit criteria, a real mix of testers, weekly triage and a written decision.

A beta that ends with “people seemed to like it” tells you very little about whether to launch. Friendly users are polite, comments arrive in scattered pieces, and the go/no-go becomes a gut call in a meeting. This guide shows how to plan a beta so that it ends with evidence and a decision you can defend.
Who it is for
Product managers at B2B or B2C software companies who are about to run a beta or early access programme before general launch.
What you will have at the end
A one-page beta plan: launch questions, exit criteria, a participant mix, a schedule, a triage routine and a written decision record.

Decide what the beta must prove
Do this before you recruit anyone. Pick two or three launch questions. Each one should be answerable with a yes or a no from data, not an opinion.
- Can users reach the core value on their own, without a call or a walkthrough?
- Does it hold up at real data volumes, not demo data?
- Can support handle the questions this feature creates?
“Do people like it?” is not a launch question. Liking is easy to get and says little about readiness.
Set exit criteria in numbers
Turn each launch question into a measure and a threshold. Typical ones are an activation rate, a crash-free sessions threshold and the number of open severity-1 bugs. Add a sign-off from support if one of your questions is about support load.
Agree the numbers with engineering and support, not for them. Write them down with the date. Mark each as a must-meet or a should-meet, so that you know in advance which misses block a launch and which ones you can launch with.
Write the finish line before the race starts. Moving it afterwards is how a beta turns into a story.
Recruit a deliberate mix
Do not take whoever says yes first. Build the group on purpose:
- Most of the group from your target segment, the people you are launching for.
- A few edge-case power users, who push on volume and unusual set-ups.
- Some skeptics, who will tell you what the fans will not.
Cap the size. The right number is the one where you can talk to every participant at least twice. A hundred silent testers give you less than twenty you actually speak to.

Instrument and schedule
Set up three things before day one:
- Event tracking for every step that feeds an exit criterion. If you cannot measure it, it cannot be a criterion.
- One in-product feedback prompt, in a fixed place, so that comments arrive in one channel with context.
- Check-in calls in weeks 1 and 3. Week 1 shows first-use problems. Week 3 shows what holds up once the novelty has gone.
Give the beta an end date and keep it. A beta with no end never produces a decision. Four to six weeks is a common length, but the real rule is that the length must be enough to collect the data your criteria need.
Triage weekly and publish it
Once a week, sort everything into three buckets: fix before launch, fix after launch, and won’t fix. Name one owner for the triage, usually you. A shared job with no owner is a job nobody does.
Then publish the result to participants. A short list is enough. When people see that their report landed in a bucket, they keep reporting. Also keep feature requests apart from readiness. A request tells you what people want next. It does not tell you whether the product is ready.
Run the go/no-go review
At the end, hold one review with engineering and support. Put the results next to the criteria you wrote in step 2. The decision is then one of three: go, go with risks, or no-go.
Write it down. Record the decision, who agreed, the known risks, who owns each risk, and what you will do about it after launch. This record is what you will open in three months when someone asks why you shipped.
Worked example: an invoicing feature
A team runs a six-week beta of a new invoicing feature with 40 small-business accounts. The numbers below are made up to show the shape of the plan. Use your own.
Launch questions. Can a user send a first invoice within 10 minutes? Does syncing work for accounts with more than 500 customers? A third, for support: can the team answer the new questions from its help articles?
| Criterion | Type | Target | Result at week 6 |
|---|---|---|---|
| Accounts that send a first invoice within 10 minutes, unaided | Should | 70% | 68% |
| Accounts with over 500 customers that sync correctly | Must | All of them | All of them |
| Crash-free sessions | Must | 99.5% | 99.6% |
| Open severity-1 bugs | Must | 0 | 0 |
| Top support questions covered by help articles | Must | Support lead signs off | Signed off |
The group is built on purpose: most accounts are typical small businesses, a handful have customer lists above 500, and a few are doubtful about moving their invoicing at all. Calls happen in weeks 1 and 3.
Triage at week 3
Fix before launch
Importing a large customer list times out for one big account. It threatens a must-meet criterion, so it goes first.
Fix after launch
The invoice template allows only one logo position. Annoying, but nobody is blocked.
Won’t fix
Several requests for multi-currency invoices. Out of scope for this release, and logged as a request, not a readiness issue.

Decision at week 6: go, with risks
All must-meet criteria are met. The one miss is a should-meet: activation is 68% against a target of 70%. The review accepts it because the team had agreed in advance that a small miss on a should-meet criterion does not block launch, as long as the cause is understood and owned.
- Risk: new users who import a customer list at setup are slower to a first invoice. Owner: the PM.
- Action: simplify the import step, first item after launch, with activation re-measured in the first weeks.
- Support has a short note on the import step ready for day one.
Mistakes to avoid
- Recruiting only fans. They forgive what customers will not. Add skeptics and edge cases.
- A beta so long it never ends. Set the end date at the start and hold it.
- Counting feature requests as readiness signal. Requests show demand. Readiness comes from your criteria.
- No named owner for triage. Without one person, the weekly sort quietly stops.
- Changing the exit criteria after seeing results. If a number was wrong, say so before you look at the new data, and record who agreed.
Before you recruit, check that you have: two or three launch questions, numeric criteria agreed with engineering and support, a participant mix with a cap, tracking and two call dates, a triage owner, and a date for the review.


