Why most AI efforts stall
- No idea where to start. Everyone says “use AI more”, but nobody knows which work to point it at.
- The gains stay local. In a study of 515 startups, firms taught to map AI to their workflows found 44% more AI use cases and made 1.9x the revenue of firms with the same tools and training.
- No way to tell if it's working. Without a baseline, you're guessing whether AI saved time or just moved it around.
Source for the study: INSEAD working paper, Kim, Kim and Koning, 2026 →
A time audit fixes all three. You map the real week (where to start), score and pick (what to automate), then measure before and after (is it working).
The four steps
- See the real week. Real data, not memory.
- Spot the repeats. Find the work that happens every week.
- Score them. Not everything repetitive should be automated.
- Pick 3 and measure. Decide, then check it worked.
Throughout, we follow Sam, the ops lead at a 20-person agency (a made-up example).
See the real week
Start with one normal week of real data. Timesheets filled in from memory miss the small, repeated work, and that's exactly the work you're looking for.
Sam's example
| Client work | Meetings | Admin and repetitive | |
|---|---|---|---|
| Friday timesheet, from memory | 30h | 6h | 4h |
| Tracked automatically | 18h | 11h | 13h |
Memory missed 9 hours of admin and repetitive work in a single week.
How to get one real week
- With Rize: install the desktop app and let it run for one normal week. It tracks automatically, with no timers, and tags time to clients and projects.
- Without a tracker: use your calendar plus the activity history in the tools your team works in (email, Slack, your project manager).
Prompt (ask the Rize agent, or Claude or ChatGPT with the Rize connector):
Summarize my team's tracked time last week by client work, internal work and admin.
Spot the repeats
Sort the week into client delivery, internal work, and admin and repetitive work. Then list the tasks that come back every week. Look across days, people and clients: the same task spread thinly over five people is easy to miss.
Sam's example
780 tracked hours. 55% client delivery, 25% internal, 20% admin and repetitive (156 hours). The repetitive tasks:
| Task | Team hours a week |
|---|---|
| Weekly status reports | 14 |
| Meeting notes and follow-ups | 11 |
| Slack and email triage | 10 |
| Reporting data pulls | 9 |
| Chasing timesheets | 8 |
| Resizing and exporting assets | 7 |
| Invoice prep | 6 |
| Scheduling | 5 |
| Total | 70 |
Prompts:
Which repetitive tasks took the most time last week? Show team hours for each, who did them, and for which clients.
Look at my last two weeks and list the tasks I do every week, with hours spent on each.
Score them
Not everything repetitive should be automated. Score each task on three things, then rank by the score.
Score = Hours a week × Repeatable (1 to 5) × Safe to hand off (1 to 5)
Higher score means automate first.
| Rating | Repeatable: how rule-based is it? | Safe to hand off: what if AI gets it wrong? |
|---|---|---|
| 5 | Same steps every time, clear inputs and outputs | Internal only, easy to check, no harm if wrong |
| 4 | Mostly the same, a few judgment calls | A person reviews it before it goes anywhere |
| 3 | A template with real variation | Client-facing but low stakes |
| 2 | Different every time, needs context | Client-facing and visible mistakes |
| 1 | Pure judgment or creative work | Sensitive data, money, legal or relationships |
Sam's ranked list
| Task | Hrs a week | Repeatable | Safe | Score |
|---|---|---|---|---|
| Reporting data pulls | 9 | 5 | 5 | 225 |
| Weekly status reports | 14 | 4 | 4 | 224 |
| Chasing timesheets | 8 | 5 | 5 | 200 |
| Meeting notes and follow-ups | 11 | 4 | 4 | 176 |
| Resizing and exporting assets | 7 | 5 | 4 | 140 |
| Scheduling | 5 | 4 | 4 | 80 |
| Invoice prep | 6 | 4 | 3 | 72 |
| Slack and email triage | 10 | 2 | 3 | 60 |
Slack triage is one of the biggest time sinks but scores last, because it's hard to hand off safely. Optimise for quality first, and the time savings follow.
For each repetitive task, rate Repeatable (1 to 5, how rule-based it is) and Safe to hand off (1 to 5, 5 = low risk if AI gets it wrong). Score = hours per week x Repeatable x Safe. Show a table ranked highest first, with one line on each rating.
Pick 3 and measure
Pick your top three, make one decision for each, and set a date to check that it worked.
- Look at the chain, not just the task. Tasks often feed each other. Sam's top two are one chain: pull the data, build the deck, send the status update. Automate the chain end to end, or the time you save on one step gets stuck at the next.
Make one decision per task
| Decision | When to use it | Sam's tasks |
|---|---|---|
| Automate | High score, rule-based, safe to hand off | The reporting chain (data pulls and status reports), meeting notes |
| Delete | The work only exists because of a broken process | Chasing timesheets (automatic tracking ends it) |
| Delegate | Repetitive but needs a person, or a better tool already exists | Invoice prep, scheduling |
| Keep human | Needs judgment, context or relationships | Slack and email triage |
- Set a baseline and check in 30 days. This week's numbers are your baseline. In 30 days, track the same tasks again and compare. If the hours didn't go down, find out where the time moved. The hardest part of AI is usually getting people to change how they work, not the tech.
The one-slide plan for leadership
- Automate the reporting chain and meeting notes: 34 hours a week, about 136 hours a month
- Delete timesheet chasing: 8 hours a week
- Check again on Dec 17
Turn my ranked list into a one-slide plan: the top 3 decisions (automate, delete, delegate or keep human), hours at stake per month (hours per week x 4), and a check date 30 days out.
The Time Audit report
This one prompt runs the whole audit: totals, repetitive tasks, scores, chains, decisions and the one-slide plan. Run it once for your baseline, then schedule it monthly so the 30-day check happens on its own.
Run a time audit of the team's last full week to find automation opportunities. 1. Totals: tracked hours, split into client delivery, internal work, and admin and repetitive work (hours and %). 2. Repetitive tasks: find work that recurs every week (the same kind of task across days, people or clients), using labels, time entries and window titles. For each: task, team hours per week, people involved, main clients. 3. Score each task: - Repeatable, 1 to 5: how rule-based and consistent it is. - Safe to hand off, 1 to 5: 5 means low risk if AI gets it wrong. Client-facing judgment and sensitive data score lower. - Score = hours per week x Repeatable x Safe. Show a table ranked highest first. 4. Chains: flag tasks that feed each other (for example data pulls that feed status reports) and recommend automating them together. 5. Decision for each task: Automate, Delete, Delegate or Keep human, with one line on why. 6. One-slide plan: the top 3 decisions, hours at stake per month (hours per week x 4), and a check date 30 days out. Keep it plain. Use only numbers from the data.
Run it in Rize
- Go to Reports and create a new report from scratch.
- Name it “Time Audit: Automation Opportunities”, set it to your team, and choose the last 7 days.
- Paste the prompt into the AI analysis and run it.
- Turn on a monthly schedule so it re-runs and you can compare against your baseline.
- Run it in Claude or ChatGPT: connect the Rize connector, paste the prompt into a new chat, and ask it to use last week's data. To repeat it monthly, use a scheduled task in Claude or ChatGPT.
Time audit template
Copy these columns into a spreadsheet, one row per repetitive task. Fill in hours from your tracked week, rate the two 1 to 5 columns with the rubric in Step 3, and sort by score.
| Task | Hours a week | Repeatable (1 to 5) | Safe to hand off (1 to 5) | Score | Feeds into | Decision | Owner | Check date |
|---|---|---|---|---|---|---|---|---|
| Reporting data pulls | 9 | 5 | 5 | 225 | Weekly status reports | Automate | Sam | Dec 17 |
Score = Hours a week × Repeatable × Safe to hand off. Hours at stake per month = hours a week × 4.
Want it filled in for you? Build your time audit template →
Prompt cheat sheet
| Step | Prompt |
|---|---|
| See the real week | Summarize my team's tracked time last week by client work, internal work and admin. |
| Spot the repeats | Which repetitive tasks took the most time last week? Show team hours for each, who did them, and for which clients. |
| Spot the repeats | Look at my last two weeks and list the tasks I do every week, with hours spent on each. |
| Score them | For each repetitive task, rate Repeatable and Safe to hand off from 1 to 5, multiply by hours per week, and rank them. |
| Pick 3 | Which of these tasks feed each other? Group them into chains. |
| Measure | Compare this week to my baseline week for the same tasks. Where did the time go down, and where did it move? |
Next steps
| If you want to | Go here |
|---|---|
| Watch the workshop recording | The Rize workshops page |
| Try it yourself | Start a free Rize trial |
| Roll it out to a team | Book a demo |
| Automate project time tracking | The AI Time Tracking Playbook |
| See which clients make money | The Client Profitability Playbook |