Use the first record for one workflow; use the second to decide which commitments the organisation can support together. The investment article and portfolio article explain the trade-offs.
Individual decision record
| Field | Required content |
|---|---|
| Decision requested | Bounded assessment, trial, implementation, expansion or closure |
| Maximum commitment | Cash, internal effort, contract term and exit exposure |
| Problem and owner | Process boundary, accountable owner and consequence of doing nothing |
| Baseline | Intake, completed work, withdrawals, backlog, effort, quality and measurement period |
| Alternatives | Process repair, existing product, specialist purchase and custom delivery compared on the same outcome |
| Benefits | Cash reduction, avoided future cost, capacity use and revenue contribution stated separately |
| Costs | One-off cash and internal effort, recurring costs, transition, contingency and closure |
| Sensitivity | Assumption most likely to change the decision and the evidence needed to test it |
| Acceptance | Quality, service, total effort, useful volume and operating constraints |
| Capacity | Review, support and fallback demand at normal and peak load |
| Dependencies | System access, supplier commitments, legal/entity assessment and shared resources |
| Next decision | Evidence due, accountable decision-maker and conditions to continue, revise or stop |
Attach the economics workbook and relevant evaluation/release records. A short memo should point to evidence, not omit it.
Compare the credible options on one basis
Use one row per option and the same task, completion standard and planning period. “Not measured” is a valid entry; an invented estimate is not evidence. Exclude an option that fails a necessary requirement only after documenting the requirement and the evidence of failure.
| Option | Necessary requirements and evidence | Time to usable operation | One-off cash / internal effort | Recurring cash / retained effort | Contract, exit and next evidence |
|---|---|---|---|---|---|
| Process repair | Record tested changes and remaining gap | Record estimate and dependency | Record separately | Record separately | Record reversibility and next test |
| Existing platform configuration | Record capability test and constraints | Record access/configuration path | Record separately | Record separately | Record licence commitment and export/exit limits |
| Specialist purchase | Record acceptance test and supplied service | Record implementation dependency | Record separately | Record separately | Record term, support and exit obligations |
| Custom delivery | Record unmet requirement and design evidence | Record delivery uncertainty | Record separately | Record separately | Record ownership, maintenance and migration obligation |
For an illustrative decision, an existing platform appears cheaper but its ability to preserve required source references has not been tested. The appropriate next step is a bounded capability check, not rejecting it with a generic “insufficient flexibility” label or approving a custom build on an unverified assumption.
Challenge the proposal in the meeting
Ask what would make the existing process or a bought product preferable. Ask which forecast input is observed and which is an estimate. Then ask who will use released capacity, which expenditure can actually disappear, and how service survives when the AI path is unavailable.
If the case depends on a grant, separate the unsubsidised economics, eligibility assumption and approval status. A programme announcement is not an award to the organisation. If it depends on a vendor roadmap, define the consequence of delay before committing to delivery dates.
Approve the next bounded commitment when its learning or operating outcome is worth the cost. Do not treat an evaluation approval as an implicit agreement to fund whatever implementation follows.
Portfolio record
Maintain one row per workflow or shared commitment, with cash by period, internal staff demand by role and period, dependencies, benefit owner, next evidence milestone and decision date. Include live operations, renewals, migration and orderly closure; new pilots are only part of the workload.
Reconcile capacity at the level that constrains delivery. Aggregate engineer-weeks do not show whether the same data specialist is needed by every project in the same month. Reserve the trained review capacity required by the operating systems before promising it to new work.
For example, four hypothetical projects needing six reviewer-weeks each require 24 reviewer-weeks. If only 16 are available in the planning period, the eight-week deficit requires added capacity, less work or deferral beyond that period. Changing the order within the same period cannot eliminate the deficit. Sequencing can resolve a peak conflict only where the total required capacity is available across the chosen horizon.
Record shared costs once. Show how a shared platform or evaluation service supports named workflows, and distinguish its incremental cost from capacity already funded elsewhere. Keep a visible reserve with a decision owner; do not count the reserve as available to every proposal simultaneously.
Finish with the commitments the team can run, the work deferred and the evidence that could change the allocation. Retain closure costs and service obligations for projects being stopped.
Read our editorial principles for evidence, contribution and correction standards.