Plan Around the Next Decision
How to choose product bets, allocate capacity and plan around evidence when building gets cheaper.

On this page
A team can create several working prototypes quickly, then find it has neither enough customers to test them nor enough specialists to review them. Even a successful prototype adds work if someone has to operate it.
Planning starts with choosing which problem deserves attention, what could invalidate the bet and what the team will put aside to make room.
Make the direction useful
A strategy should give the team enough context to make local decisions. Describe the customer and problem, the outcome the organisation needs, the advantage it intends to build and the boundaries it will not cross. Be specific about what is out of scope when that changes a choice.
For the hypothetical support workflow in The Product Builder's Learning Loop, a team might prioritise accurate responses to routine policy questions. That does not automatically include autonomous account changes. The customer outcome connects the work; the authority boundary limits the first bet.
Find the constraint
Building is one part of the system. A product bet may also depend on access to customers, representative data, expert review, integration, support or a route to market. Identify the constraint that would prevent the team learning or delivering value, and plan for it.
A prototype can be ready in a day while customer testing takes weeks. A pilot can pass an eval but create more review work than the team can absorb. A useful product can fail to reach the people who need it. The relevant limit depends on the work; it is not always engineering capacity.
Reserve capacity for existing obligations, quality, operational maintenance and learning. When an urgent bet is added, make explicit what moves or stops. Avoid filling the roadmap merely because making an artefact is cheaper.
Choose between real constraints
Imagine a team responsible for a scheduling product. It has capacity for one substantial initiative this quarter. Customers report missed appointments when reminders arrive too late; sales wants an AI assistant that proposes appointment times; operations is dealing with a backlog of failed integrations.
The team looks at the evidence. It can see repeated reminder failures and has access to affected customers. The scheduling assistant has enthusiastic interview responses, but no evidence yet that it improves booking completion. Failed integrations are fewer, but some prevent customers from receiving updates at all. The integration owner has limited capacity because they also handle incidents.
The team chooses to address the integration failures first, including the failures affecting reminders. It protects time for incident work and runs a small research test on appointment planning rather than promising the assistant for the quarter. That choice delays a potentially valuable feature. It is justified here by the known customer impact and the shared engineering constraint, not by a rule that reliability always wins.
The team records what could change the choice: if the integration fix proves smaller than expected, or research shows that appointment planning is a larger source of missed bookings, it can reconsider. The scenario is hypothetical; the decision in a real portfolio depends on the evidence, commitments and cost of delay.
Test the uncertainty that matters

For each active bet, name the assumption most likely to change the decision. It might be customer urgency, usability, model quality, data access, review effort or willingness to pay. Choose a test that can challenge that assumption.
A customer conversation can clarify whether a problem is urgent; a prototype can test an interaction. An eval shows how the system performs on selected cases, while a bounded pilot can expose adoption and operating costs. Discovery explains how to choose a test for the question at hand.
For a consequential or hard-to-reverse decision, look for a smaller, safer test and plan the recovery path. An experiment should have an owner, a next decision and a condition that would lead the team to stop or change direction.
Plan for models that may change
Keep a list of ideas that depend on a capability the current system cannot yet provide. Record the task, the limitation and a test that would make the idea worth revisiting. Retest when there is credible reason to expect a change.
A future improvement is not a substitute for today's release evidence. Keep the system boundary clear enough to compare alternatives, but avoid building a complex routing layer before it solves a measured problem. The harness chapter covers those choices.
Use the roadmap to expose choices
Distinguish work the team is committed to from opportunities it is still testing. For each material item, show the intended outcome, evidence, unresolved uncertainty, owner and the next decision. A date may be appropriate for a real commitment; it should not disguise an untested assumption as a promise.
Review the portfolio when evidence arrives or a decision is due. Ask what has changed, where work is waiting and which bet no longer deserves capacity. Keep a record of decisions so the team can understand the reasoning later.
Try it: make a choice with limited capacity
Choose two opportunities your team is considering. For each, write down the customer problem, the evidence, the next decision and the work required. Include an existing obligation that cannot simply disappear, such as incident response or a committed delivery. If you do not have a live portfolio, use the scheduling example above.
Assume the team can make substantial progress on only one opportunity. Decide which you would pursue, what you would defer and how you would protect the existing obligation. Write the strongest argument against your choice and one piece of evidence that would change it. Then ask a teammate with a different responsibility to challenge the constraint and the trade-off.
An assistant can help expose an overlooked dependency or compare the options, but it cannot know your team's capacity or commitments unless you provide them. Check those with the people doing the work before presenting the choice as a plan.
v3.1 · Updated Oct 2026