AI use case prioritization matrix: a free template that scores by measured hours
A free AI use case prioritization matrix template. Score by measured hours, ease and risk, with a stated formula, a worked example and decision rules.
TL;DR
Score each AI use case from 1 to 5 on measured hours per week, ease and risk. Priority = (3 × Hours + 2 × Ease + 2 × Risk) ÷ 7, and 5 on risk means low risk. Take hours from observed work, never a survey. Do first what scores 4.0 or more. A separate bucket column names the change: delete, code, agent or human.
On this page
- Scales
- 1 to 5, higher is better
- Formula
- (3 × Hours + 2 × Ease + 2 × Risk) ÷ 7
- Do first
- Priority 4.0 or more
Key takeaways
- Impact and effort matrices ask the room to guess the impact; this one takes hours per week from measured work.
- A bucket column says what kind of change each row needs, so you delete and write plain code before you reach for an agent.
- A risk score of 1 is a rule, not a penalty: a person approves every case.
You have a list of AI ideas and no fair way to rank them. The usual tool is an impact and effort matrix. It works until the impact column is a guess and the loudest person in the room wins.
Below is a sheet you can paste today, a scorer you can try on this page, and the rules for running the session. It scores three things: measured hours per week, ease and risk. A fourth column, the bucket from the four-bucket method, names the kind of change.
Measure the hours. Do not ask for them.
Ask a team how long invoice coding takes and you get a number from memory. Memory drops the exceptions, the rework and the messages asking what a code means.
Example, invented: the team says 10 hours a week. A one-week time sample says 45. At 10 hours the row scores 2 on Hours. At 45 it scores 4. The ranking flips on one number.
Measured hours
Hours per week that everyone on one workflow spends on it, taken from observing the work, a system log or a time sample. Count the workflow, not the person.
The three scales
Score every row from 1 to 5. Higher is always better. That includes risk: 5 means low risk.
| Score | Hours (measured, per week) | Ease | Risk (5 = low risk) |
|---|---|---|---|
| 1 | Under 5 hours | Data is out of reach, no past cases exist, or the steps are disputed. | A wrong result costs money, harms a customer, or cannot be undone. |
| 2 | 5 to 19 hours | Data is reachable with real work. Few past cases. | A wrong result is costly, but a later check would catch it. |
| 3 | 20 to 39 hours | Data and some past cases exist. One or two integrations. | A wrong result costs hours to fix and is usually noticed. |
| 4 | 40 to 79 hours | Past cases exist and the tools are ones you already run. | A wrong result is cheap to fix and usually noticed. |
| 5 | 80 hours or more | A rule or a removal in tools you already use. No model needed. | A wrong result is cheap, obvious and easy to override. |
Priority = (3 × Hours + 2 × Ease + 2 × Risk) ÷ 7. The result runs from 1 to 5. Hours weigh most because you can measure it. Ease and risk are judgments, so they weigh 2 each. If your company cares more about safety, raise the risk weight and divide by the new total.
The bucket column does not change the score. Sort the steps into delete, code, agent or human, then write the bucket that holds most of the measured hours. The free four-bucket sorter does the sorting.
The sheet to copy
AI use case prioritization sheet
Paste this into a doc or sheet. Make one row per workflow. Fill the columns from left to right.
AI USE CASE PRIORITIZATION SHEET
Team: ________ Date: ________ Scored by: ________
ONE ROW PER WORKFLOW. COLUMNS
1. Workflow (name it the way the people who do it name it)
2. Owner (one named person)
3. Measured hours per week (all people on the workflow, added together)
4. How measured (Observed / System log / Time sample / ESTIMATE)
5. Hours score (1-5, from the scale below)
6. Bucket (Delete / Code / Agent / Human: the bucket that holds most of the measured hours)
7. Ease score (1-5)
8. Risk score (1-5, where 5 = low risk)
9. Priority = (3 x Hours + 2 x Ease + 2 x Risk) / 7
10. Estimated cost per task (optional, from the cost per task calculator)
11. Decision and date to re-score
SCALE (higher is always better)
Hours: 1 = under 5 h/week | 2 = 5-19 | 3 = 20-39 | 4 = 40-79 | 5 = 80 or more
Ease: 1 = data out of reach or steps disputed
2 = data reachable with real work, few past cases
3 = data and some past cases, one or two integrations
4 = past cases exist, tools you already run
5 = a rule or a removal in tools you already use, no model
Risk: 1 = a wrong result costs money, harms a customer or cannot be undone
2 = costly, but a later check would catch it
3 = costs hours to fix, usually noticed
4 = cheap to fix, usually noticed
5 = cheap, obvious and easy to override
DECISION RULES (apply in this order)
1. Risk score 1: a person approves every case. The risky step never goes to an agent alone.
2. Hours score 1: park the row. Exception: bucket is Delete or Code and Ease is 5.
3. Row marked ESTIMATE in column 4: it ranks below every measured row, whatever its score. Measure it first.
4. Priority 4.0 or more: do now. Delete and Code rows need no model. Agent rows start in shadow mode.
5. Priority 3.0 to 3.9: hold. Collect 20 to 50 past cases or measure more, then score again.
6. Priority below 3.0: park. Score again next quarter.
BEFORE YOU FINISH
- Is every Hours score based on observed work, a log or a time sample?
- Did at least two people who do the work agree on the bucket?
- Did you price the top three rows per task?Five workflows, scored
Try it
Score the five sample workflows, then replace them with yours
Change a dropdown and watch the order move. Rename a row, add your own, then copy the result as a table.
Priority = (3 × Hours + 2 × Ease + 2 × Risk) ÷ 7. A score of 4.0 or more: do this first.
- 4.4
Do this first
- 4.1
Do this first
- 4.0
Do this first
- 3.1
- 2.1
Sample data, not a real company. Green: Priority 4.0 or more. Amber: held back by risk, so the hours alone do not decide it.
| Workflow | Measured h/week | Hours | Bucket | Ease | Risk | Priority | Decision |
|---|---|---|---|---|---|---|---|
| Support ticket routing | 85 | 5 | Agent | 4 | 4 | 4.43 | Do now. Run in shadow mode first. |
| Copying status reports between two tools | 30 | 3 | Delete | 5 | 5 | 4.14 | Do now. Remove the copy step. |
| Weekly revenue review rebuild | 62 | 4 | Code | 4 | 4 | 4.00 | Do now. Plain code, no model. |
| Invoice exception coding | 45 | 4 | Agent | 3 | 2 | 3.14 | Hold. Collect 20 to 50 past cases. A person approves each one. |
| Contract clause review | 28 | 3 | Human | 2 | 1 | 2.14 | Park. A person approves every case (rule 1). |
Status-report copying has fewer hours than invoice exceptions and still ranks higher. Deleting a step is easy and safe, and the formula pays for that.
Ticket routing wins on hours, and its bucket says agent. Start it in shadow mode: the agent works beside the people and sends nothing.
Run the session
- 01Gather rows10 to 15 weekly workflows
- 02Measure hoursObserve, log or time sample
- 03Score aloneEase and Risk, no talking
- 04CompareDiscuss gaps of 2 or more
- 05DecideFormula, rules, one owner
Re-score every quarter
Measure before the meeting. Use the meeting for Ease, Risk and the decision.
Pick 10 to 15 workflows
Take ones that run every week. Get the weekly hours from observed work, a system log or a time sample across the team. Write how you measured in column 4.
Do this todayDo this today: ask each team lead for their three most repeated workflows.
Sort the steps into buckets
Put the people who do the work in the room, plus one person who knows the systems. Write the main bucket in column 6.
Do this todayDo this today: run one workflow through the four-bucket sorter.
Score Ease and Risk alone
Everyone writes both scores without talking. Reveal them at the same time, or the most senior person sets the number.
Do this todayDo this today: use sticky notes or a hidden column.
Discuss only gaps of two or more
Ask the lowest and the highest scorer why. One of them often knows an exception the others do not. Agree one score and write the reason down.
Do this todayDo this today: cap each discussion at five minutes.
Apply the formula and the rules
Run the six decision rules in the order given. Measured rows rank above ESTIMATE rows. Name one owner for every row that says do now.
Do this todayDo this today: put the formula in the sheet so the score fills itself in.
Price the top three rows per task
For each agent row, estimate cost per task before you commit. Drop any row where the run cost is close to the value of the hours.
Do this todayDo this today: enter one row’s volume and review time in the calculator.
Put the next scoring date in the calendar
Do it before the meeting ends. OpenAI’s guide advises re-evaluating each quarter, because high-effort ideas can become low-effort. Measure the hours again too.
Six ways the sheet goes wrong
- Hours from a chat thread. Someone writes “about ten hours a week” and it goes in the sheet. Mark it ESTIMATE. It ranks below every measured row until someone counts.
- Scoring the wiki page. The page lists six steps. The work has eleven, and three of them are asking the person who knows. Measure the work.
- Timing your fastest person and multiplying. One person tells one story. Observe at least two, and add up the whole workflow.
- One bucket for a ten-step workflow. Sort the steps first. Ten steps often hold a delete, a code step and one agent step.
- Letting hours outvote risk. A 5 on Hours does not buy back a 1 on Risk. Rule 1 holds whatever the total says.
- Skipping the run cost. 90 hours a week looks good until each task costs more in model fees and review than the clerk it replaces.
What the existing frameworks do, and where they stop
We read five public ones on 2026-10-14. All are fine starting points. In the pages we read, none tells you to measure hours per week from observation.
| Framework | What it does | Where it stops |
|---|---|---|
| OpenAI, “Identifying and scaling AI use cases” | A quadrant of value against effort with four zones. No numeric scale. | Does not say how to measure value. It does advise re-scoring each quarter. |
| GSA IT Modernization Centers of Excellence, “Identifying AI use cases in your organization” | Three factors, impact, effort and fit, as checklists for U.S. federal agencies. | No score and no way to size the problem. |
| Clearwork, “AI Use Case Prioritization Framework + Free Scoring Sheet” | Four pass or fail gates, then seven criteria scored 1 to 5 with set weights. Scores are discounted for Estimated or Assumed evidence. | The closest to this one. Business value comes from a dollars-per-year band, and it scores whole use cases, not steps. |
| CIGen, “AI use case prioritization template” | An Excel template with a dashboard, scoring business impact, data readiness, compliance, effort and risk. | The page calls it a planning tool that does not replace deeper discovery. I did not see the formula on the page. |
| Prometheus, “AI Use Case Prioritization Matrix” | Impact and effort, 1 to 10, for eight use cases picked from a library of more than 50. | Inputs are answers typed into a form. No observed hours. |
Price the agent rows
A high score can still lose money. For each agent row, price one task all in: model cost for every attempt, review time, and time spent fixing mistakes. The cost per task guide has the method and three worked examples. The free AI cost per task calculator compares two models on the same task.
Then check one rule. Weekly run cost (cost per task × tasks per week) should sit well below the value of the hours the agent takes over. Count only the hours in the agent bucket.
Where Nolli fits
Where Nolli fits
Nolli maps each workflow from how the work really runs, sorts the steps into the four buckets, and shows what each change is worth. That gives you evidence for the Hours and bucket columns without a survey. Nolli does not replace the scoring session: your people still set Ease and Risk, and your company decides what Nolli may learn from. You can fill in this sheet by hand with a time sample instead. No public price. Request a demo.
Request a demoQuestions people ask
What is an AI use case prioritization matrix?
A table that scores each AI idea on the same scales so you can rank them. Most score impact and effort. This one scores measured hours, ease and risk, and adds a bucket column.
How do I measure the hours for a workflow?
Observe the work, read a system log, or take a time sample across the team for a week. Add up everyone on the workflow. Do not use a survey, and do not time only one person.
Why does Hours get a weight of 3?
Hours is the one input you can measure instead of judge, so it counts most. Ease and risk are judgments and count 2 each. You can change the weights. Write them on the sheet and use them for every row.
What does the bucket column do?
It names the kind of change a row needs: delete the step, run it as plain code, give it to an agent, or keep a person on it. It stops you building an agent for work that a rule or a deletion would fix.
How often should we re-score?
Every quarter, and whenever a workflow changes. OpenAI’s guide advises re-scoring each quarter because high-effort ideas can become low-effort as AI improves.
Sources
- “Identifying and scaling AI use cases.” OpenAI, accessed October 14, 2026.
- “Identifying AI use cases in your organization (AI Guide for Government).” U.S. General Services Administration, IT Modernization Centers of Excellence, accessed October 14, 2026.
- Avery Brooks. “AI Use Case Prioritization Framework: Gates, Scores and an Evidence Check.” Clearwork, October 9, 2026.
- “AI use case prioritization template for risk-averse adoption.” CIGen, accessed October 14, 2026.
- “AI Use Case Prioritization Matrix.” Prometheus Agency, accessed October 14, 2026.
Plain-text copy: /blog/ai-use-case-prioritization-matrix-template.md