In practice
A feature roadmap says: ship CSV export in March. An outcome roadmap says: customers get their data out without asking us, and support tickets about exports fall. Export might be the answer. So might a link in the settings page. The roadmap doesn't say yet.
The idea comes from Josh Seiden's Outcomes Over Output, where an outcome is a measurable change in human behavior that drives business results. Output is what you make. Outcomes are what you get from having made it.
It isn't a roadmap with no dates, and it isn't a list of company goals like "grow revenue". Revenue is too vague to guide a week of work. An outcome is narrower: a specific behavior you can watch for.
How to write one
- Pick two or three outcomes for the quarter. Write each as a behavior: who does what differently.
- Add the evidence: the requests in the customer's words, how many accounts, what it is worth.
- Give each outcome one number to watch and where it comes from.
- List the bets that might move it. Keep them loose. Two or three per outcome is plenty.
- Review on a fixed rhythm. Ask whether the number moved, then drop or change the bets.
Common mistakes
- Writing a feature and calling it an outcome. "Launch the dashboard" is output. "Owners check results without asking us" is an outcome.
- Choosing outcomes nobody can measure, so the review turns into opinion.
- Hiding dates the business needs. Outcomes don't remove the need to say roughly when.
Why it matters for product builders
A list of features is a list of guesses. If one ships and nothing changes, a feature roadmap marks it done anyway. An outcome roadmap keeps it open until the behavior moves.
It also makes the next call easier. When a loud customer asks for something that doesn't serve any outcome on the page, you have a reason to say not now, in plain words.
The cost is that it is harder to write. You need to know what you are trying to change, and for whom, before you pick a feature.
What changes in AI-first products
Agents make output cheap. A feature that took a sprint can take an afternoon, so the shipping count tells you less than it did. That makes the outcome the only number left worth reading.
What didn't change is the hard part: choosing the outcome and checking whether it moved. An agent can build five answers to an outcome and can't tell you which one the customer wanted. That comes from requests in the customer's words and from what you see in use.
See how Annsa does it: Priorities →