Product & Project Delivery
Plan a Private Beta with Clear Participants, Support, and Exit Criteria
Build a bounded private beta participant and support plan, then assess a complete fictional evidence packet. Separate invitations, attempts, submitted requests, independent completion and support costs before deciding whether to expand.
Plan a private beta as a controlled service operation: name who can enter, what they can do, who supports them, what evidence you will collect, and what would make you expand, pause, or stop. Write the exit criteria before inviting participants. Registrations and positive comments do not establish that the whole task works, that support is sustainable, or that a wider audience can use the service.
This LandOffer.ai guide reviews official guidance checked October 27, 2026 and builds a fully fictional workshop-rescheduling beta. The participant plan, operational records, outcomes, and approvals are invented. We checked the authored counts and capacity arithmetic locally; no invitations were sent, customers enrolled, service operated, or rollout changed.

Define the beta's service boundary before its audience size
GOV.UK's beta guidance distinguishes invited private-beta users from wider public-beta access. It emphasizes learning and iteration capacity, support, and the wider service journey. Its assessment process belongs to government delivery; this article does not require every commercial team to pass a GOV.UK assessment or adopt its governance roles.
State which part of an existing service is being tested and what remains authoritative. A beta label does not make conflicting records harmless. Users need to know whether their action changes a real reservation, merely requests a change, or creates a record someone must reconcile with another system.
Our fictional training provider, Linden Workshops, is testing a portal for requesting a replacement date for a nonurgent workshop place. The user submits a preference and later reads a staff decision. The existing staff booking ledger remains authoritative. A portal request does not automatically move a place, guarantee availability, change payment, or issue a refund.
Service owner Sabine decides cohort expansion. Ewen owns daily operational checks, Talia provides direct support and backup, researcher Inez gathers evidence, and engineer Basil owns technical fixes. These roles and capacity acceptances are stipulated inside the example, not real assignments made for this article.
Worked asset: the completed participant and support plan
The planned first window is November 16–20, 2026, five working days. Linden invites no more than twelve named customers who already have an eligible workshop place. Open registration, third-party booking agents, urgent changes, payments, and group bookings are excluded. People can continue using the existing service rather than join the beta.
The recruitment allocation is six frequent customers, four occasional customers, and two customers selected specifically to investigate keyboard or assistive-technology access. These are primary recruitment categories used for the fictional count, not mutually exclusive kinds of people in reality. Twelve is a capacity choice, not a universal sufficient beta sample or representative market estimate.
| Plan area | Filled arrangement and accountable owner |
|---|---|
| Admission | Sabine approves a named twelve-person list; Inez records accepted, declined, and not-started states. No forwarded invitation creates access automatically |
| Participant explanation | State the narrow task, authoritative booking ledger, support hours, optional participation, feedback use, and fallback route before access. Research consent and data handling follow the organization's approved process |
| Support window | Weekdays 13:00–21:00 UTC during the five-day pilot. Talia handles the designated support queue; no overnight or weekend response promise |
| Fixed operational checks | Ewen reserves two 30-minute record-health checks each day: reconcile request/ledger state, inspect stuck work, and review access alerts. These fixed checks exclude direct case handling |
| Direct support reserve | Talia reserves 60 additional minutes across the week for participant cases; planned total support/operations allocation is six hours. Further work requires an explicit capacity decision |
| Backup and escalation | Ewen covers the support queue if Talia is unavailable, displacing a scheduled noncritical check and recording that change. Basil receives technical blockers; Sabine decides suspension or extra capacity |
| Fallback | Existing booking support remains available through the usual approved route. Staff identify the beta request and reconcile it before processing a replacement request |
| Feedback and evidence | Inez combines task records, assistance needed, issue reports, and selected follow-up conversations. Avoid real booking details in broad project channels |
The fixed allocation is five hours: two half-hour checks per day for five days. Adding the one-hour case reserve gives six hours. It is a planning budget, not evidence that participants will need only one hour of help. Coverage availability and hours of active work must also be distinguished in a real staffing plan.
The beta's user-facing introduction says: “Your request is not a confirmed date change. Check the decision recorded for your request; staff will reconcile any accepted change with the existing booking ledger. If the portal fails or you prefer help, use the existing support route and include your request reference. You can leave the beta without losing access to the normal service.” This is authored example wording, not a message actually sent.
Decide what evidence can answer the beta's questions
GOV.UK's research-in-beta guide considers the end-to-end service, including support and offline steps. It recommends including a range of users and examining usability, accessibility, analytics, and support evidence. Its point is broader than whether invited people can open the first screen.
Linden asks four questions: can customers submit and retrieve the decision; can staff reconcile it without contradictory bookings; can users with the selected access needs complete the task; and can the support model operate within its accepted capacity? A satisfaction response alone cannot answer all four.
Choose the unit of each measure. Count people invited separately from people attempting the task. Count requests separately from support contacts: one person can contact support twice about one unresolved request. Record assisted completion separately from independent completion rather than counting a staff workaround as evidence that the interface was usable without help. Here, independent means no help performing the task; admission questions are logged separately.
Inez's task definition ends when the user can retrieve a decision consistent with the authoritative ledger. “Submitted” is an intermediate state. A documented unsubmitted attempt can be accounted for without creating a booking-ledger request. It remains pending support, not completed. These distinctions prevent operational bookkeeping from inflating the success count.
Set expansion, pause, and stopping rules in advance
The team writes the following case-specific gates before the first invitation. The thresholds are invented planning choices, not published standards or statistical guarantees:
| Gate | Evidence required at the expansion review | Decision if unmet |
|---|---|---|
| End-to-end task | Every task attempt is accounted for; submitted requests link to the booking ledger. Completion and pending support are reported separately | Hold expansion and resolve unexplained state mismatches |
| Access and usability | At least one completed task from each declared recruitment category, with no unresolved task-blocking access issue | Keep expansion on hold; provide the existing support route and investigate the blocked journey |
| Support capacity | Fixed checks plus direct handling fit the six-hour allocation, or extra capacity is explicitly accepted before expansion | Revise support scope, staffing or cohort size; do not rely on unapproved extra work |
| Immediate stop | Any credible unauthorized disclosure or conflicting booking state with possible customer impact triggers the response plan | Pause the affected beta route, preserve necessary evidence, involve the responsible owners and reconcile through the existing service |
| Expansion authority | Sabine reviews Inez's evidence, operational owner acceptance and unresolved issues | No automatic move to wider access because the window has ended |
Not observing a disclosure in a small cohort does not prove security. The stop rule describes what the team will do if the signal appears; it is not a security test or an excuse to skip required pre-beta checks. Those actual approvals and checks must exist before real customer access.
Likewise, an access need remains part of the scope even if the invited person does not start. Do not remove a recruitment category at review to make the results look complete. Record the missing evidence and plan appropriate follow-up rather than redefining success after seeing the data.

Worked asset: a mixed evidence packet and the exit decision
The following is a completed fictional review packet for November 20. It is a teaching simulation, not a report of a beta we ran. All counts and issue descriptions were authored for this article.
| Primary recruitment category | Invited | Accepted | Attempted | Authored task disposition |
|---|---|---|---|---|
| Frequent customers | 6 | 5 | 5 | Five completed independently; one invitation declined |
| Occasional customers | 4 | 3 | 2 | One independent completion, one assisted completion; one accepted participant did not start; one declined |
| Keyboard/access investigation | 2 | 2 | 1 | One task blocked by inaccessible focus movement; the second accepted participant did not start |
| Total | 12 | 10 | 8 | Six independent completions, one assisted completion, one unresolved task |
The local arithmetic gives 6/8, or 75%, independent completion, and 7/8, or 87.5%, completion including assistance. These are descriptive ratios for eight invented attempts. They are not population estimates, an experiment result, or proof of improvement over the previous service. The two accepted people who did not start remain visible outside the attempt denominator.
All eight task attempts are accounted for. Seven submitted requests have recorded decisions consistent with the booking ledger. The eighth is an unsubmitted draft linked to a pending support attempt; it has created no service request or booking change. The bookkeeping gate therefore passes, while task completion remains seven rather than eight. The pending support owner and next action remain visible.
The access issue remains open as B-07. The affected participant cannot reach the final request-submission control using keyboard navigation in this fictional case. Talia helps them through the existing support route, but that does not establish that the beta interface's barrier is fixed. The stated access gate fails, and the second participant's unstarted task leaves an additional evidence gap.
The support ledger contains nine direct contacts, each taking an authored ten minutes, for 90 minutes. Together with 300 minutes of fixed checks, that is 390 minutes, or 6.5 hours. It exceeds the six-hour allocation by 30 minutes. Reporting that overrun exposes the staffing decision needed before a larger cohort. These are non-overlapping work categories; direct handling is explicitly excluded from the fixed checks, so the calculation does not count the same work twice.
Talia's extra half-hour is an overrun in the example, not approved capacity merely because the work occurred. Nine contacts also do not mean nine different customers. The log preserves repeat contacts about the same problem so Inez can examine the cause without inflating the audience.
Sabine's completed decision is do not expand. Suspend additional beta invitations in the plan, keep the normal booking-support route available, retain the affected attempt as pending support, and prioritize B-07's repair and verification. The review retains useful findings and directs remediation while withholding wider access.
Close the loop with users and the operating owners
The next record assigns Basil the B-07 fix proposal, Inez the repeat task/access review, and Ewen the resolution of the pending support attempt through the approved existing service. Talia and Sabine must agree a realistic support allocation before another cohort. No repair, repeat study, staffing change, or invitation is marked complete at this checkpoint. Sabine revisits expansion after Inez brings verification evidence; an elapsed review date does not close the gates.
For the affected participant, the planned communication states the known limitation, the existing help route, and how the team will provide the request's current status. It does not promise a fix date the engineering owner has not accepted. Participants whose work completed receive the appropriate service outcome, without turning a research update into marketing permission.
GOV.UK's assisted-digital guidance discusses designing support, measuring it, and considering its resources and end-to-end routes. Its government assessment duties are context-specific. The transferable question is whether your actual support arrangement can serve the people included, including those who struggle to use the digital path.
If a beta must stop, identify every outstanding request, its authoritative record and owner, the approved fallback, and the communication needed. Ending the experiment does not remove responsibility for unfinished customer work. Keep access, data retention and service closure actions under their respective owners rather than assuming a closed signup form completes the exit.
Before the next cohort, bring the filled participant plan, support ledger, task dispositions, unresolved issue record and explicit expansion decision to review. More workplace guides are available from LandOffer.ai. Expand only after the named evidence and operating owners support the change; an invitation count is a starting condition, not the outcome.
Sources and Further Reading
- GOV.UK: How the beta phase works — updated February 19, 2021; current body checked October 27, 2026.
- GOV.UK: User research in beta — published November 18, 2016; current body checked October 27, 2026.
- GOV.UK: How assisted digital support is assessed — updated May 23, 2016; current body checked October 27, 2026.


