Library · Data Practice OS
Decision inventory
Template · Published · Source reviewed
The Decision inventory lists the decisions that matter, so the practice builds for decisions, not datasets. It is the starting point of the From data to decisions - the decision stack: pick a decision, then build only the layers it is missing. Keep it in a version-controlled table (a dbt seed or a register) with one owner per row, and review it each quarter alongside the readiness re-score (AI readiness assessment).
Status. A proposed template. Not an agreed register. The rows below are illustrative and need confirming with each business owner.
Columns
| Column | What to write |
|---|---|
| Decision | The question, phrased as one: "Can this licence be renewed?" |
| Owner | The accountable business owner (a role, not a person's name) |
| Frequency | Continuous, daily, monthly, annual |
| Volume | Decisions per year, roughly |
| Consequence of error | Low / medium / high, with one line on who is harmed and how |
| Reversibility | Easily reversed / with effort / hard to reverse |
| Authority | The delegation or instrument that authorises it, and the level that holds it |
| About a person? | Y/N. Y means at least medium risk in AI use case intake and risk triage |
| Current support | Manual, legacy system, report, rules engine, AI |
| Data needed | Data products or systems of record for the facts (Data product canvas) |
| Knowledge needed | Policy, legislation, procedure, with the owner of each |
| Rules needed | Rule cards or decision tables, and whether they are approved (Rule authority) |
| Outcome signal | What tells you whether the decision was right (Outcome feedback loops) |
| AI pattern fit | None / read / recommend / approve with review / execute, and which pattern (AI-native patterns) |
| Gaps | Missing stack layers: facts, meaning, knowledge, rules, context, authority, outcomes |
| Retention | Retention schedule for its decision records (Decision records and provenance) |
Illustrative rows
Illustrative only. Volumes, owners and authorities are placeholders, not the organisation figures.
| Decision | Owner | Freq. | Volume | Consequence | Reversible | About a person | Current support | Data | Knowledge | Rules | Outcome signal | AI fit | Gaps |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Can this licence be renewed? | Licensing operations | Continuous | ~10^5 to 10^6 | High: wrongful refusal or unsafe driver | With effort | Y | Legacy system, officer | Licensing record, medical flags, case state | Renewal procedure, legislation | Eligibility, fees (rule cards, draft) | Override, appeal, reversal | Recommend; officer approves | Rule cards not approved; no decision record |
| Which road defects are fixed first? | District maintenance | Fortnightly | ~10^4 defects | Medium to high: crash risk, claims | Yes | N | Spreadsheet and inspector judgement | Defects, assets, traffic, crashes | Maintenance standard | Risk matrix (not effective-dated) | Response time met; re-report within 6 months | Recommend; planner approves | Duplicate reports not resolved; matrix not versioned |
| Does this heavy-vehicle route need a permit, and on what conditions? | Heavy vehicle access | Daily | ~10^4 | High: infrastructure damage, safety | Hard | N (operator) | Officer with maps and assessment tools | Network, structures, restrictions | Access policy, notices | Mass and dimension limits | Permit amended; incident on route | Recommend conditions; officer approves | Network relationships (graph candidate, When a knowledge graph earns its place) |
| Where should this customer enquiry go? | Customer service | Continuous | ~10^6 | Low | Yes | Y | Manual triage | Enquiry text, customer reference | Routing procedure | Routing table | Re-routed; resolved at first contact | Approve with review, then sample | Routing table not owned |
| Which registrations get a compliance check this month? | Compliance | Monthly | ~10^3 checks | Medium: missed non-compliance, burden on compliant customers | Yes | Y | Rules and analyst | Registration, inspection history | Compliance policy | Selection criteria | Hit rate of checks | Recommend with a random sample kept outside (Outcome feedback loops) | No outcome capture |
How to use it
- Fill the first ten rows in a workshop with business owners. Aim for breadth, not precision.
- Score each row: value (volume × consequence) against readiness (how many stack layers are missing). High value with few gaps is a lighthouse candidate (Roadmap and maturity).
- Draw a decision requirements diagram for the top three (Decision modelling with DMN).
- Run the readiness assessment on the chosen decision, not on a dataset (AI readiness assessment, AI readiness scorecard (workbook v2)).
- Set the AI fit conservatively. Start at recommend for decisions about people. Move up only when outcome measures support it and the authority allows it (Human in the loop, AI-permission record).
Not when
- Do not try to list every decision the organisation makes. List the ones with volume, consequence or both.
- Do not treat the AI fit column as approval. It is a hypothesis for AI use case intake and risk triage.