Product Requirements as a Decision Contract
Replace a bloated feature specification with a compact decision contract that preserves need, boundaries, assumptions, evidence, and change history.
On this page17 sections
- 01The document before surgery
- 02Write the contract in nine fields
- 031. Decision
- 042. Need and evidence
- 053. Outcome and non-outcome
- 064. Scope boundary
- 075. Assumptions
- 086. Scenarios and rules
- 097. Constraints
- 108. Acceptance evidence
- 119. Open questions, ownership, and change history
- 12The compact rewrite
- 13Use the contract in refinement
- 14What standards can and cannot settle
- 15What to delete from the old document
- 16Sources
- 17Read next
The fictional specification below is fourteen pages long. It names screens, fields, button labels, notification text, API responses, and a launch date.
It does not say why the product should exist.
Nobody can find the evidence behind the decision. An edge case appears in three places with different rules. A late discovery is added as a comment, while the approved document stays unchanged.
A product requirement should act as a decision contract. It records what the team is deciding, which need justifies it, what is inside the boundary, and which evidence would make the result acceptable.
“Contract” here means that the people doing the work can see what they agree on, what is still open, and how a later change will be recorded, so the contract itself can change.
The document before surgery
Consider a fictional B2B reporting product. Customers can schedule reports, but administrators cannot see when a delivery fails.
The proposed feature is an email alert for failed scheduled reports.
The original specification starts like this:
Add a notification centre with tabs for unread and all notifications. Show a red badge in the navigation. Send email and in-app notifications when any scheduled report fails. Include report name, owner, time, error, retry button, settings link, and contact support button.
It continues with page layouts and a field inventory.
The document has made several decisions without naming them:
- notification centre rather than a focused failure view;
- email plus in-product delivery;
- all failures are equally urgent;
- every recipient may retry;
- technical error details are safe and useful;
- report owners are the right recipients;
- a badge is the right interruption pattern.
Those may be sound choices. The specification contains no evidence that lets a reviewer tell.
Write the contract in nine fields
The rewrite begins with a one-page core. Supporting designs, data definitions, and technical notes can remain linked artefacts.
1. Decision
State the choice the team is making now.
Provide account administrators with an actionable signal when a scheduled report fails and requires human intervention.
It names no notification centre or other delivery surface. It commits the team to a recipient, a condition, and an outcome that leads to action.
2. Need and evidence
Name the observed problem and link to its source.
Administrators discover failed deliveries through recipient complaints or manual checks. In six recent support cases, diagnosis began more than one working day after failure. Support records and four administrator interviews are linked.
These numbers belong to the fictional example. They show the level of evidence a real contract should cite, not a benchmark for another product.
The evidence should distinguish observation from interpretation. User interview design helps prevent a requested feature from being mistaken for the underlying need.
3. Outcome and non-outcome
Define the change the team expects and what it is not claiming.
Administrators can recognise an actionable failure and reach the appropriate recovery path before a recipient reports it.
This change does not prevent report failures, guarantee recovery, or replace operational monitoring.
The non-outcome protects the document from absorbing adjacent ambitions.
4. Scope boundary
Use explicit inclusions and exclusions.
Inside
- scheduled email reports;
- failures requiring administrator action;
- account administrators with report access;
- one durable in-product record and one email signal;
- route to retry or support, depending on error class.
Outside
- successful or delayed deliveries;
- recipient-level preferences;
- mobile push;
- a general notification centre;
- automated repair of the report job.
An exclusion keeps the present decision honest. It does not promise the item for a later release.
5. Assumptions
List beliefs that could invalidate the approach.
- administrators are accountable for failed deliveries;
- the system can classify failures into user-actionable and operator-actionable groups;
- email is permitted for the relevant account roles;
- a retry is safe only for selected failure classes;
- technical error text can be translated into safe, useful language.
Attach an owner and a test to material assumptions. Otherwise the list becomes decoration.
6. Scenarios and rules
Scenarios expose decisions better than a catalogue of screens.
Scenario A: expired data connection
The administrator sees which report failed, when it failed, and that the connection needs renewal. The action opens connection settings if they have permission.
Scenario B: transient delivery failure
The record shows that the system will retry automatically. It does not invite a duplicate manual retry.
Scenario C: access without authority
A report viewer can see the report but cannot view account-level diagnostic details or change the connection. They receive no administrator alert.
Scenario D: repeated failure
One ongoing incident should not create an uncontrolled stream of identical emails. The contract must define grouping and update behaviour.
Rules should live once. Screens and messages can reference them.
7. Constraints
Constraints narrow the solution space for a reason.
- notification data follows the report’s retention and regional handling rules;
- recipients must already hold account-level administrative access;
- emails must not include sensitive report content;
- the first release uses existing email infrastructure;
- support needs an auditable event and correlation identifier.
Separate a true constraint from a current preference. “Use a red badge” is a design choice unless a tested design system or platform rule makes it mandatory.
8. Acceptance evidence
Acceptance needs evidence that the agreed behaviour and risk controls are present. “Matches the design” is not enough.
For this change:
- scenario tests cover each failure class and permission combination;
- administrators can identify the affected report and correct next action;
- non-administrators do not receive or reveal restricted details;
- retry actions are idempotent where offered;
- grouping prevents duplicate message storms;
- production events connect the failure, notification, action, and outcome;
- accessibility review covers the in-product record and email content.
Some evidence is available before release. Outcome movement requires observation after exposure. Keep those two stages distinct.
9. Open questions, ownership, and change history
The contract should remain credible when knowledge changes.
Open questions include:
- Should delegated administrators receive the same alerts?
- How long should a resolved notification remain visible?
- Which failure classes may be retried without support review?
- Does email delivery need an account-level opt-out?
Record each material change with date, decision, reason, evidence, and affected artefacts.
Name the decision owner and an owner for each unresolved question. Record dependencies as decisions expected from named people, not as a vague list of teams the work “depends on”.
Do not silently rewrite the past. A reviewer should be able to understand why the current boundary differs from the first proposal.
The compact rewrite
The decision contract now fits on one page:
| Field | Contract statement |
|---|---|
| Decision | Alert administrators to scheduled-report failures requiring human action |
| Need | Failures are found late through complaints or manual inspection |
| Outcome | Administrator recognises the failure and reaches a valid recovery route |
| Boundary | Scheduled email reports, selected failure classes, account administrators |
| Assumptions | Role ownership, error classification, email permission, safe retry |
| Scenarios | Expired connection, transient failure, unauthorised viewer, repetition |
| Constraints | Access, retention, sensitive data, existing email infrastructure, audit |
| Acceptance | Behaviour, permissions, idempotency, grouping, observability, accessibility |
| Unknowns and ownership | Delegation, retention, retry policy, opt-out, each with a named owner |
The page does not contain every implementation detail. It tells each discipline where detail belongs and which decision that detail must protect.
Design handoffs can link interaction states and content. System design can hold delivery, idempotency, state, and observability choices.
The contract remains the trace between need, boundary, evidence, and decision.
Use the contract in refinement
A requirement becomes useful when the team challenges it.
In backlog refinement, ask each discipline to attack a different field:
- design tests whether scenarios expose missing states and comprehension risks;
- engineering tests constraints, state transitions, and recovery;
- data tests whether outcome and diagnostic evidence can be observed;
- operations tests ownership, support paths, and failure volume;
- security and privacy test access, disclosure, and retention;
- product tests whether the boundary still answers the need.
Do not ask, “Is the requirement clear?” Ask what decision could still be interpreted in two incompatible ways.
If disagreement reveals a missing product choice, update the contract. If it reveals implementation freedom within a settled boundary, let the responsible discipline decide.
What standards can and cannot settle
ISO/IEC/IEEE 29148 specifies requirements-engineering processes and the information items they produce across the lifecycle. It does not demand that every product decision become a large document.
GOV.UK’s service guidance links user needs, stories, acceptance criteria, complexity, and dependencies. Bill Wake’s INVEST lens also keeps stories negotiable and testable.
The Scrum Guide treats the Product Backlog as emergent. These sources support clarity and adaptation together.
None can prove that the chosen feature will create value. Discovery evidence can be weak, organisational constraints can change, and released behaviour can produce an unexpected response.
So the contract keeps assumptions and outcome evidence visible. It does not present the specification as certain.
What to delete from the old document
Delete repeated rules, speculative future states, and design detail that has no decision attached.
Move interface specifications to the design artefact. Move event schemas and retry mechanics to the technical design. Link both back to the relevant scenario, constraint, or acceptance condition.
Keep history that explains a live boundary. Archive superseded material where the team can retrieve it without forcing every reader through it.
The aim is a shorter path from evidence to decision, from decision to implementation, and from a changed assumption to the work it affects.
Sources
- ISO/IEC/IEEE 29148: requirements engineering
- GOV.UK Service Manual: start by learning user needs
- Bill Wake: INVEST in good stories and SMART tasks
- The Scrum Guide
Read next
Related books
Two books to
read next.
If you want to go further on this topic, these are two good places to start.
01
communication
Made to Stick
by Chip & Dan Heath
Why some ideas survive and others die, revealing the six principles (SUCCESs) that make ideas memorable and shareable.
02
communication
The Pyramid Principle
by Barbara Minto
The foundational framework for structured communication, teaching how to present ideas in a clear, logical hierarchy that makes complex information accessible.
Some outbound links are affiliate links and support independent bookstores.