Why a gradual rollout complicates a launch
Product teams release some features in stages: to a small share of customers first, then to more, sometimes by plan, by region or by the size of the account, watching for faults at each step before going further. The staged rollout entry in the glossary describes the two jobs a release in stages can do: caution, because a fault reaches a fraction of customers before it is noticed and can be limited by a pause or a rollbackReturning a changed system, such as a tag container, workflow or page template, to its last working version, planned and rehearsed before the change goes live., and evidence, which holds only where chance decides who is in each group and both versions run side by side. Engineering chooses the stages for its own reasons, and this guide does not cover how stage sizes or speeds are set, or any tool that controls them.
The marketing problem starts where the engineering one ends. The launch planning and tiers guide runs its timeline from pre-launch through launch day to post-launch, with briefings, a running order and adoption campaigns all set relative to a single date. A staged release has no single date. It has a first day on which some customers can use the feature and a later day, not always known in advance, on which all of them can. Every asset in the bill of materials has to be placed against one of those days, or against each stage in between.
Launch management is, in the Product Marketing Alliance's account, a core product marketing responsibility, as the product launches hub records, and the launch intake tutorial cites its view that tiering is among the simplest routes to launches that can be repeated. A staged release still needs a tier and a plan; what it adds is a question that intake should ask and the tier review should answer: will this reach every customer at once, and if not, who has it when? Where the answer changes after the release has been tiered, intake's rule on re-tiering when scope or date moves applies here too.
Who has it, when: mapping the rollout stages
Before any message is written, sit down with product management and write the stage map: each stage, which customers it includes, when it is expected to start and what has to be true before the next stage begins. The map is the launch plan's equivalent of a launch date, and every later decision in this guide depends on it.
| Stage field | What to record | Why marketing needs it |
|---|---|---|
| Who is included | Plans, regions, account sizes, platforms or a random share of customers | Decides who can receive which message |
| Expected start | The planned start, and how firm it is | Places briefings and assets against the stage |
| Gate to the next stage | What product management will check before going further | Shows where the rollout could pause |
| What customers see | Whether the feature appears on its own or needs switching on | Decides whether in-product guidance is needed |
| Who is told | Customers in the stage, their account managers, support | Sets the internal briefing for each stage |
| Language and region | Whether every language and market in the stage is ready | Catches stages held back by translation |
Two kinds of stage map need different handling. In one, stages are chosen: a plan, a region or a kind of customer goes first. Marketing can then address each stage directly, because the people in it share something. In the other, the stages are a growing random share of customers, and nothing marks out who has the feature except the switch itself. Marketing can only target those customers through the product, because no list outside it says who they are.
Some stages are not really stages of a rollout at all. Customers chosen to try a feature before it is ready, with terms, feedback and an end date, are in a beta, and the beta programme tutorial treats early access as a beta by another name. Where a stage is defined by plan, the question of which plans get the feature at all is a packaging decision for the guide to placing a new feature in the plans, not a rollout schedule; a plan that will never receive the feature should never be told it is waiting for it. And where a regional stage is late because its translations are not ready, the diagnosis of localisation delaying launches finds the step the work is queuing behind.
Choosing when to announce
There are three main choices, and the tier and the stage map decide between them.
| Choice | How it works | Suits | Watch for |
|---|---|---|---|
| Announce at the first stage | Public announcement on the first day, stating who has it now and who later | A feature the market already expects, or where the first stage covers most of the audience the announcement reaches | Customers searching for a feature they lack; support and sales fielding the questions |
| Announce to each stage as it arrives | In-product message and release note for each group as it receives the feature, no public moment until the end | Minor and feature launches; random stages where nobody outside the product knows who has it | Early customers comparing notes with later ones; the release losing its moment |
| Announce at general availability | Quiet release to stages, with the public announcement held until every customer has it | Major launches with press, email to the whole base or a campaign | Early customers using the feature with no explanation; a public date set before the last stage is certain |
Combining the second and third choices is a natural fit for a release whose tier earns a public moment. Each stage hears inside the product as it receives the feature, and the email, the press and the campaign wait for general availability. That keeps the public announcement in line with what a reader can do on the day they read it, which is the test every choice has to pass.
Put a readiness gateA go-live check that confirms the work is built, such as tracking, pages, stock, support and legal review, before a launch, market entry or peak goes ahead. in front of whichever public moment the plan includes. The product launch playbook names an owner for every gate who reports its status, and when a gate fails it allows three honest courses: postponing, launching less, or going ahead with the risk written down and a mitigation in place. For a staged release the product readiness gate has a sharper meaning: the feature works for everyone the announcement will reach on the day it goes out, not only for the stages that already have it.
Decisions between stages follow the same discipline. The phase-gate reviews framework writes down, before each phase starts, what it must show and what happens if it does not, and limits the decision to going on, going with changes, holding or stopping. Product management owns the gates between stages; product marketing asks for a seat at them, because each decision moves the dates that every message depends on.
Messages for customers who are still waiting
Customers in later stages will hear about the feature before they can use it: from colleagues in other accounts, from a public announcement, from the changelog or from a competitor's seller. What they need is short and honest.
Say what is true about availability. Every release noteNotes published with a product release telling the people who use it what changed, who has it and from when, written for users, not as ticket titles. states who has a change and from when, and the release notes and changelog tutorial names "rolling out gradually" as one of the states a note should declare, with the entry updated as availability changes. Write the note once with the rollout state, then add a dated line at each stage and at general availability rather than publishing a new note each time.
Tell nobody to look for something they cannot find. An in-app messageA message shown inside an app during a session the person started, such as a banner, modal or tooltip, needing no operating-system permission. goes only to customers who have the feature, placed where it lives and pointing to it rather than to an announcement page, as the in-app message design guide recommends for any announcement to existing users. A launch email is the same: the launch email tutorial sends the announcement to everyone eligible for it, and a customer who cannot open the feature is not yet eligible.
Promise nothing the stage map does not support. If a waiting customer asks when they will get it, the answer comes from the stage map and the rules on future capability in the guide to talking about the roadmap: a period only where product management has committed to one, no date otherwise, and no request for a customer to be moved into an earlier stage promised by anyone who cannot grant it. Tell waiting customers that the feature will arrive without any action from them and how they will know it has, such as a message inside the product when it appears.
Sales and support while availability is mixed
A staged rollout gives sellers and support agents a question they cannot answer from memory: does this customer have it? Answer that before the first stage starts.
Give sales a single line on availability that matches the stage map, updated at every stage: who has it now, who will have it next and what to say to a prospect. A prospect who signs during the rollout may land in any stage, so the line says plainly whether new customers receive the feature at once or join the rollout like everyone else. Demos run on an account that has the feature, and sellers say that it is being released in stages if the prospect asks when they could use it.
Give support a saved replyA pre-written answer a support agent inserts and then edits for the customer in front of them, keeping facts and tone consistent across the team. for the question "where is the new feature?", written as the guide to building a library of saved replies advises: facts consistent, tone agreed, and room for the agent to address the customer's own situation. The reply needs an owner who updates it at each stage, because a saved reply that holds out-of-date availability repeats the old answer with the authority of an official one. Give support a way to check whether a particular account has the feature, so that the reply is accurate for the customer in front of them, not only for the rollout in general.
Brief account managers on their own accounts' stages. A large customer that hears about the feature from a smaller one first, or sees a competitor's seller mention it, should already have heard from its account manager when it will receive it, within what product management has approved.
Measuring adoption among the customers who have it
The measuring launch success guide reads adoption as a sequence, discovery, then first use, then activation, then habitual use, each as a share of the eligible customers. During a staged rollout, eligible means customers who have had the feature switched on, not the whole base. Divide by everyone and adoption looks weak until the last stage, whatever the customers who could use it did. The conversion rate calculator gives the share at each step when it is fed the right denominator: customers with access, then customers who tried it, then customers who reached activation.
Read each stage as its own cohortA group of customers who share a starting event and period, tracked together over time., from the date it received access, so that a stage that has had the feature for longer is not compared directly with one that received it last week. Agree with product management, before the first stage, how the activation rateThe share of new users or sign-ups who complete the activation event, the action marking first real value, within a set window. for this feature is defined and which events will be recorded, so the reading does not start with a gap in the tracking.
Be careful what the comparison between stages can show. The guide to improving conversion without A/B tests counts a staged rollout as a fair comparison only where chance assigns the groups and both run over the same weeks. A rollout that goes to small accounts first and large ones later compares different kinds of customer at different times, and a difference in adoption between them says as much about the customers as about the feature. Early stages are also small, so apply the habits of the small numbers guide, with the count beside every rate and a range rather than a single figure.
When a rollout pauses or reverses
Rollouts pause when a stage shows a fault, a load problem or a result product management did not expect. Plan the messages for a pause before the first stage, so that nobody writes them under pressure.
| Situation | Customers who have it | Customers still waiting | Public messages |
|---|---|---|---|
| Pause, feature stays on for those who have it | Nothing new unless the fault affects them | Nothing new; availability line unchanged | Hold any dated announcement; update the release note if it gave a date |
| Pause with a fault customers have seen | A short note saying what is wrong and what to do meanwhile | Nothing new | Correct anything public that described the feature as working |
| Rollback for some or all who have it | A direct notice before it disappears, with the reason and what happens next | Nothing new | Correct the release note with a dated line |
| Rollout stopped for good | Treated as a retirement for anyone who relies on it | Told, if they were told it was coming | Removed from every claim |
A pause before any public date has been given costs marketing little: the stage map moves, the availability line for sales and support moves with it, and the public announcement waits. A pause after a public date has been given is a slipped date, and the roadmap guide's rule applies: the customers who were told hear first, from the people who manage their accounts, before they discover it.
A reversal is harder. Customers who have started to rely on a feature and then lose it are experiencing a removal, and the guide to retiring a feature regards removal as a one-way door for the customers who depend on it, with messages by impact and a way forward. Where the fault is live and customers are affected, the marketing incident management playbook runs the report, containment, correction and telling the people who need to know; the launch plan waits until the incident is closed.
A school catering software announces in stages (illustrative)
Vosquerid holds its announcement for the last stage
Vosquerid, an invented menu-planning service for school kitchens, built automatic allergen labels and planned to release them in 4 stages: single-site kitchens first (about 1,100 of its 2,600 customers), then small groups of schools, then councils' kitchens, then the 140 large catering contracts. The first plan was a full launch on the day of stage 1, with an email to all 2,600 customers and a press release. That would have sent about 1,500 kitchens looking for labels they did not have.
The product marketer mapped the stages with product management, with each stage's gate and expected start. Each stage received an in-app card on the menu screen as its labels switched on, and the release note said "rolling out gradually" with a dated line added at each stage. Sales received a one-line availability answer and support a saved reply, with a way to check any kitchen's stage. The email and the press release were held for general availability, behind a readiness gate.
Adoption was read only among kitchens with access: in stage 1, 640 of 1,100 kitchens printed labels within 4 weeks, about 58%. In stage 2, 209 of 410 kitchens did so in the same period, about 51%, which the team did not read as a verdict on the labels, because small groups of schools differ from single-site kitchens. When a fault in recipes with substitutions paused stage 3, the 1,510 kitchens with access kept their labels, the waiting kitchens heard nothing new, and the public date was moved by 3 weeks before any email or press release had gone out. Stage 4 followed, and the announcement went to all 2,600 kitchens on the day every one of them could use it.
Common mistakes
- Announcing to everyone on the first day. A public launch that many readers cannot act on sends them to support and sellers to improvise.
- No stage map. Without a shared record of who has it when, every message and briefing is a guess.
- Prompts and emails to customers without access. Announcements go only to people who can open the feature.
- A release note that never changes. The note states the rollout state and gains a dated line at every stage.
- Promising a waiting customer a date. Only periods product management has committed to are shared, through the approved route.
- Measuring adoption across the whole base. The denominator is the customers who have the feature, read stage by stage.
- Calling a chosen rollout a test. Stages picked by plan, region or size compare different customers at different times; only random, concurrent groups give a fair comparison.
- No plan for a pause. Messages for a pause, a fault and a rollback are written before the first stage, not during one.
