Vic Sol · Decision DeskOne-page sheet · 10 min to use
Do not automate a broken process: the 10-minute diagnosis
For operations managers with a process that is a candidate for automation and an AI proposal on the table.
When to use it
Someone proposes "putting AI on" a process whose documents currently get corrected three times.
You have a demo that promises to save time and you do not know how long the process takes today.
Your slowest process gets stuck in approvals, data or corrections, and you do not know which.
The sheet
One row per step, start to finish. Five to eight rows is enough.
Step
Who
Working time
Wait after the step
Times corrected
Does the approval decide anything?
Exceptions (out of 10)
[Verb + object: "Review quote"]
[Role]
[Real minutes]
[Hours or days until someone picks it up]
[0, 1, 2, 3 or more]
[Yes / No / No approval]
[Cases that leave the normal path]
Illustrative: approve order
Area manager
5 min
2 days
2
No: approves everything
3
Totals: work ____ · wait ____ · corrections ____ · approvals that decide nothing ____ · exceptions ____
Decision:
Automate: few corrections, approvals that decide, low exceptions, and you already have today's measurement.
Fix first: two or more corrections per step, approvals nobody reads, or data that does not add up. AI would repeat that faster, and with a licence.
Leave it alone: the process changes every time, or one person does it in less time than it would take to explain it to a tool.
My decision: Automate / Fix first / Leave it alone · What I fix before: ____________________
How to fill it in 10 minutes
Sit with the person who does the process, not the person who designed it. Ask them to walk through it step by step and write one verb per row.
Fill the wait column first. It is almost always larger than the working time and it is what AI does not fix.
Ask how many times each document gets corrected and why. That is where the rework lives.
For every approval ask: does anyone decide anything with this, or does it exist out of habit? If nobody has rejected anything in six months, it does not decide.
Make the three cuts before deciding: eliminate what exists out of habit, simplify approvals and copies down to a single format, and measure how long it takes and how many times it is corrected.
The most common mistake: automating the wait. A tool that drafts in seconds a document that then waits two days for a signature saves seconds, not days.
The line that is almost always missing
The wait column. AI proposals measure working time, because that is what the tool shortens. But the process lives in the waits, the corrections and the approvals nobody reads. If you automate before measuring that, you will never know whether AI saved you anything, because you have nothing to compare against.
Vic Sol · Decision Desk · vicsol.org/resources/ · Free to use with attribution. This sheet structures a conversation; it is not legal, financial or audit advice.
Open the .docx in Word (or the .xlsx in Excel) or print it before the meeting.
Fill the fields with the team; leave the blanks visible, they are the discussion.
Close with one owner, one date and one next action. Nothing else.
Prompt
Generate this sheet for your case
Paste this prompt into ChatGPT, Claude or Gemini together with your proposal, process or policy. It reproduces the sheet with your data and flags what is missing.
Act as an operations consultant trained in process improvement who has watched broken processes get automated and companies pay a licence to repeat the same problems faster.
Context I am giving you:
- Process (name, where it starts and where it ends): [ ]
- Steps as told by the person who does it, with who does each one: [paste description]
- Working and waiting times per step, if known: [ ]
- How many times each document gets corrected and why: [ ]
- Approvals, and whether anyone has rejected anything in the last six months: [ ]
- Frequent exceptions: [ ]
- Automation proposal, if any: [paste]
What I need:
1. A process table with one row per step: step, who, working time, wait after the step, times corrected, whether the approval decides anything, exceptions.
2. Diagnosis: where the waits, the rework and the approvals that decide nothing are concentrated.
3. The three cuts before automating: what to eliminate (steps that exist out of habit), what to simplify (approvals, copies, formats) and what to measure (how long it takes today and how many times it is corrected).
4. Decision: automate / fix first / leave it alone, with the reason in two lines and, if "fix first", what to fix before putting AI on it.
Output format: one table, three short lists and the decision. Maximum 450 words. No introduction.
Rules:
- Do not invent times. Where data is missing, write [DATA PENDING] and tell me how to measure it.
- Do not recommend automating a step whose wait is longer than its working time without saying that AI does not reduce that wait.
- Label every estimate as an estimate.
- If critical information is missing, ask me at most 3 questions before answering.
- Answer in plain, direct business English, with no sales adjectives.
What to check in the answer:
Does the wait column hold data, or did it leave it empty and still recommend automating?
Did it identify any approval that decides nothing, or accept all of them as necessary?
Does the "fix first" decision come with a concrete list of what to fix, or only with the recommendation?
Vic Sol