When should you use it?
- Your CRM forces a lost reason from a short dropdown and the same two or three options always win
- Reps write useful notes that nobody has time to read in bulk
- You are planning sales training or process changes and want evidence rather than opinions
Not for: Teams whose lost leads have no free-text notes to read; Exports you cannot strip of personal details before sharing with an AI tool; Judging individual reps, since the notes reflect what was written down, not everything that happened
What do you need, and what do you get?
You need
- A CSV or sheet of lost leads with the selected lost reason and the notes, one lead per row
- Names, phone numbers and email addresses removed or replaced before upload
- A short description of what you sell, so the model can tell product fit from timing
You get
- A rewritten reason for each lead, based only on its notes
- Up to eight reason groups with counts
- The share of leads where the dropdown and the notes disagree
- The top causes you could have controlled, each backed by a few quoted notes
How does it work, step by step?
- Export one periodPull one quarter of lost leads with the dropdown reason and the notes. A single period keeps the comparison fair.
- Strip personal detailsDelete or replace names, phone numbers and email addresses before anything leaves your CRM.
- Run the comparisonPaste or upload the rows with the prompt, and ask it to mark rows where the notes are too thin to judge.
- Check the evidenceRead the quoted notes under each top reason. If a group rests on two vague notes, treat it as a lead, not a finding.
- Change one thingPick the biggest controllable reason, change one step in the process, and rerun the same comparison next quarter.
The prompt
Our own prompt, inspired by Hemad Mirchi and written to work in ChatGPT, Claude, Gemini and others. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
You are reviewing lost sales leads for {business_description}. Each row below has the lost reason chosen in our CRM and the rep's notes. Personal details have been removed. For each row, write in one line the reason the notes actually show. If the notes are too thin to tell, write "unclear". Then group those reasons into no more than {max_groups} categories, give a count for each, and report the percentage of rows where your reason differs from the CRM reason. Name the three most common reasons we could have influenced (for example response time, product fit, follow-up or language), and under each give three short quotes from the notes as evidence. Rows: {rows}
Inspired by Hemad Mirchi: LinkedIn post: the CRM lost-reason field versus the notes. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Pasting raw CRM exports puts client details into a third-party tool. Remove them first, as the author also warns.
- If notes are short or missing, the model will fill gaps with plausible stories. Ask it to label thin notes as unclear rather than guess.
- Counts from a few dozen leads are noisy. Look at the quoted evidence before you change training or targets.
- Reps may write defensive notes once they know these are analyzed. Present the findings as process problems, not personal scores.
Our take
Lost-reason fields are a classic case of data that is easy to collect and close to useless, because the person choosing the option is also the person who would look bad if the real answer were slow follow-up. Hemad Mirchi's point is that the honest version is usually already sitting in the notes, and reading notes in bulk is exactly the kind of tedious job a language model handles well. We like that his version asks for a disagreement rate between the dropdown and the notes, which turns a vague hunch into a number you can track each quarter. We also like that he raises data handling directly, since a CRM export is full of client details that should never be pasted into a chat tool as is. Our version adds two guards: label thin notes as unclear instead of guessing, and show quoted evidence under every group so a manager can check the reasoning. In our dry run with forty fictional lost leads from a home builder, 22 of the dropdown reasons disagreed with the notes, three notes were marked unclear, and slow callbacks came out on top.
Where the time goes back: A manager gets a grounded read on why deals are lost in minutes, instead of skimming thousands of notes or relying on a dropdown that rarely says anything new.
FAQ
Which CRM does this work with?
Is this only for real estate?
How many leads do I need?
Source and credit
Inspired by Hemad Mirchi. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- LinkedIn post: the CRM lost-reason field versus the notes
- Creator
- Hemad Mirchi
- Published on
- Original date
- Nov 5, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- ChatGPT
- AI dry run
- Nov 9, 2026: passed. Dry-run tested by our AI test agent on sample input. Reasons rewritten from notes, thin notes marked unclear, 22/40 disagreement reported, top three controllable causes quoted.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Nov 9, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.