Two hundred and twelve ideas
The week after the offsite, Maya asked for an export of everything that had ever been requested of the product. The list had 212 items. Some were bugs. Some were feature requests from a single customer, logged three years earlier by a salesperson who had since left. Some were genuinely important. Eleven were labelled “top priority.”
The list was not a backlog. It was an archaeological record of every conversation the company had ever had. And it had been driving the roadmap, because the loudest recent request usually won.
Maya did three things in her first month. She archived every item older than a year that nobody could connect to a current customer problem; 130 items disappeared, and in the following year, nobody asked for any of them back. She grouped what remained under the outcomes in the new strategy. And she replaced the old roadmap, a list of features with quarter-end dates, with one organized around problems and outcomes, with dates only where a commitment was genuinely required.
This chapter covers how to choose what to work on, and how to show those choices in a way that builds trust rather than false certainty.
Output versus outcome
The most important shift in modern product management is from output to outcome. Output is what the team ships: features, releases, documents. Outcome is the change in behavior or results that the output was meant to produce: more navigators confirming ownership within 24 hours, fewer missed follow-ups, higher renewal rates.
Output is easy to measure and easy to celebrate. Outcomes are what customers and the business actually care about. A team can ship everything on its roadmap and change nothing that matters. Josh Seiden put it simply: an outcome is a change in human behavior that drives business results.1
Working on outcomes changes how a team behaves:
- It starts each piece of work with a measurable goal, not a feature description.
- It treats the first release as a test, not a finish line.
- It stops or changes direction when the measure does not move, instead of shipping version two of something that did not work.
- It talks about results in reviews, not just about what was delivered.
Outcomes are harder to commit to, because you cannot guarantee that a customer will change their behavior. That discomfort is honest. A roadmap full of committed features is promising something the team does not control: that those features will matter.
The outcome roadmap
A roadmap’s job is to communicate direction and sequencing, and to support decisions. It is not a release plan, and it is not a contract.
The simplest and most useful format I know is Now, Next, Later, a format associated with Janna Bastow and the team at ProdPad.2 Instead of dates on a timeline, items sit in three horizons:
- Now: what the team is working on, with high confidence and detail.
- Next: what comes after, with medium confidence; problems understood, solutions still open.
- Later: what the team expects to tackle, with low confidence and no detail; mostly problems, not solutions.
Three features make this format work.
Every item is a problem or an outcome, not a feature. “Navigators can confirm ownership of discharged patients within 24 hours” leaves room for the team to find the best solution. “Build a discharge dashboard” does not.
Confidence is visible. Moving from Now to Later, the detail thins out on purpose. Stakeholders can see what is committed and what is a direction.
Dates appear only where they are real. Some commitments need dates: a regulatory deadline, a contractual obligation, a conference launch. Put those dates on the roadmap explicitly. Everything else is ordered, not dated.
Different roadmaps for different audiences
One roadmap rarely serves everyone. The board wants strategy and outcomes. The sales team wants to know what they can promise and when. Engineering needs sequencing and dependencies. Customers want to know whether their problem is being addressed.
Keep one source of truth and present it differently for each audience. Never show customers anything you would be uncomfortable being held to, and never give sales a date that engineering has not agreed to. The fastest way to lose a team’s trust is to promise their work to a customer without asking them.
Prioritization frameworks, and when to use them
Frameworks do not make decisions. They make the reasoning behind a decision visible, so it can be challenged and improved. Choose the framework that fits the decision.
RICE scores each option on Reach, Impact, Confidence and Effort: (Reach × Impact × Confidence) ÷ Effort. It was popularized by Intercom and works well for comparing many independent features where reach is measurable.3 Its weakness: it rewards things that touch many users a little over things that matter deeply to a few, and in business software the “few” are often the economic buyers.
Weighted scoring rates each option against criteria you choose, such as strategic fit, customer value, revenue impact, risk reduction and effort, each with a weight. It is flexible and makes strategy explicit. Its weakness: whoever sets the weights sets the answer.
Kano analysis sorts features into basic expectations (customers are angry if they are missing, but not delighted if present), performance features (more is better) and delighters (unexpected value). It is excellent for deciding what must be in a release and what will differentiate it.
Cost of delay asks how much value is lost for each week a feature is not available. Divided by duration, it gives weighted shortest job first, a sequencing rule popular in large organizations.4 It is the right tool when timing matters: a regulatory deadline, a competitor launch, a renewal at risk.
The same backlog can come out in a very different order depending on the framework. Figure 5.2 shows five Brindlepath candidates ranked both ways, using illustrative scores.
The lesson is not that one framework is right. It is that the choice of framework is itself a strategic choice, and should be made openly. If two methods disagree sharply on an item, that item deserves a conversation, not a formula.
The opportunity solution tree
For discovery-heavy work, Teresa Torres’s opportunity solution tree is a useful complement to scoring.5 It starts with the desired outcome at the top, branches into the customer opportunities (needs, pain points, desires) that could drive it, and only then into candidate solutions and the experiments that test them. It prevents the most common prioritization error of all: comparing solutions before agreeing on the problem.
The loudest voice in the room
Every product leader faces the same pressures: the CEO’s idea, the biggest customer’s demand, the deal that will close “if we just add this one thing.” Prioritization frameworks help, but the real skill is handling these conversations well.
For the executive’s idea: treat it as a hypothesis, like any other. Ask what problem it solves and what evidence would show it is working, and put it through the same discovery as everything else. Executives usually respect a product leader who takes their idea seriously enough to test it.
For the biggest customer: separate the need from the request, as in Chapter 1. Then ask whether the need is shared by other customers in your target segment. If it is, it may belong on the roadmap. If it is unique, consider whether it is better solved by configuration, a services engagement or a partner, rather than core product.
For the deal that needs one more feature: ask how many deals the feature has really decided in the past year. Sales teams are sincere, but every lost deal has many reasons, and the missing feature is the easiest one to name. Track requests by deal and outcome, and the pattern usually becomes clear.
When you say no, say it with reasons and an alternative: “We’re not building that this half, because our focus is post-discharge ownership. Here’s what we can do instead, and here’s when we’ll revisit it.” A clear, reasoned no builds more trust than a vague maybe.
Making room for what keeps the lights on
A roadmap full of new features eventually collapses under its own weight. Technical debt, security work, performance, and maintenance of what already exists all need capacity, and they rarely win a head-to-head scoring contest against a shiny feature.
The fix is to allocate capacity explicitly before prioritizing within each bucket. Figure 5.3 shows the split Maya agreed with Dev Raman for Brindlepath’s next two quarters.
The exact split depends on the product’s stage and health. A young product may put most of its capacity into new outcomes; a mature product with aging infrastructure may need far more for reliability. What matters is that the split is a decision made in daylight, reviewed each quarter, rather than the result of a hundred small trade-offs nobody saw.
Brindlepath in practice: one hard no
Six months after the new roadmap, Brindlepath’s largest prospect, a health system that would have doubled the company’s revenue in its region, asked for a patient-facing mobile app as a condition of purchase. The request came with a deadline: the app had to be live within six months.
The sales team argued for it, reasonably: the deal was large, and a patient app was on the “Later” list anyway. Maya ran the request through the same process as everything else. She asked what problem the app would solve for the prospect, and learned that the real need was a way to confirm patients’ contact details and preferred call times after discharge. She checked the evidence: none of Brindlepath’s existing customers had asked for an app, and two had said patients rarely downloaded them. She estimated the cost: at least two-thirds of a team for six months, displacing the standard connector that every existing customer needed.
The answer was no to the app, and yes to the need. Brindlepath offered a short text-message confirmation flow, built on the recently acquired messaging tool, which could be live in eight weeks and served the same purpose. Maya joined the sales call herself to explain the reasoning: the prospect’s patients would get what they needed sooner, and the prospect would be buying a product whose roadmap was not hostage to any single customer, including them.
The prospect signed, three months later, without the app. Not every hard no ends that well. But a reasoned no with an alternative ends well far more often than sales teams fear, and a roadmap that bends to every large request does not stay a roadmap for long.
Common mistakes
Treating the roadmap as a promise. Dates on features become commitments customers hold you to, whether or not the features turn out to matter. Commit to outcomes and to the few dates that are genuinely required.
Letting the backlog become a museum. A backlog of hundreds of items is not a plan. Archive anything no current customer problem supports; the important items will come back.
Scoring precisely, deciding badly. Frameworks produce numbers with decimal places that hide guesses. Use scores to structure the conversation, not to end it.
Saying maybe instead of no. “We’ll look at it” sounds kind and creates false expectations. A clear no, with reasons and an alternative, builds more trust.
Starving the foundations. Without an explicit capacity allocation, reliability, security and debt lose every head-to-head contest with new features, until an outage or breach forces the issue.
One roadmap for every audience. The board, sales and engineering need different views. A single document trying to serve them all usually serves none of them well.
Try this
Export your current roadmap and backlog. For each item, write the problem it solves and the measure it should move. Any item where you cannot do both goes into a “needs a problem” list for review. Then rebuild the roadmap in the Now, Next, Later format, with only problem-framed items, and mark the commitments that genuinely need a date. Finally, agree a capacity split with your engineering lead for next quarter, and write it at the top of the roadmap where everyone can see it.
How AI changes this
AI can do the bookkeeping. An assistant can de-duplicate a sprawling backlog, link requests to the customers and deals that raised them, draft RICE or weighted scores from available data, and summarize the evidence behind each item. Maya’s 212-item cleanup would take an afternoon today rather than a month. Treat the scores as a first draft. The model does not know your strategy unless you tell it, and it does not know which customer is about to churn.
AI creates roadmap pressure. Since 2023, almost every product team has faced demands to “add AI.” Some of those demands are right; many are a solution looking for a problem. Run AI features through the same discipline as any other: what outcome will this change, for whom, and how will we know? An AI feature that does not move a measure is an expensive demo.
AI features need different confidence. A traditional feature either works or it does not. An AI feature works some of the time, at a quality that varies with the input. Put AI items on the roadmap with explicit quality targets and evaluation plans, not just a ship date. Chapter 17 covers how.
AI can help you say no. Ask an assistant to write the case against your top three priorities, or to find the items on your roadmap that do not connect to your stated strategy. It is easier to hear an uncomfortable argument from a machine than from a colleague, and easier still to act on it before the planning meeting.