Win-Loss Analysis Template: Let AI Review Your Lost Deals

Export a quarter of closed deals from your CRM, add the call notes and emails for each, remove names, and have an AI assistant tag every deal against a fixed list of loss themes. Count the themes, interview a sample of lost buyers to test them, and finish with an action list of no more than three changes.

The step small firms skip is the interview. CRM loss reasons are written by the person who lost the deal, and notes are written by the seller. AI can sort and count what is written down quickly, but it cannot know what the buyer was thinking. Treat its table as a set of hypotheses and use conversations to confirm them.

Follow me on Instagram@sagnikteaches

The data you already have, ranked by how far to trust it

Win-loss analysis compares deals you won with deals you lost to see what differed. For a small B2B firm the evidence sits in four places, and they are not equally reliable.

Connect on LinkedInSagnik Bhattacharya

1. CRM fields. Deal value, source, stage history, close date, competitor and loss reason. Reliable for dates and values, weak for reasons. Salespeople often pick the least embarrassing dropdown option.
2. Call and meeting notes. Closer to what was said, but written through the seller's filter. Useful for objections raised and who attended.
3. Email threads and proposals. Show timing, the buyer's questions, how fast you replied and what you offered. Often the most honest record of the decision's last weeks.
4. The buyer's own account. From an interview. The most valuable and the hardest to get. It is also the only one that can contradict the other three.

Start with the first three, because they cost nothing, then use the fourth to check them. If you have used call transcripts, the habits in logging call summaries in your CRM make this much easier next time.

Subscribe on YouTube@sagnikteaches

Building the deal table

An illustrative firm: a six-person marketing agency selling monthly retainers to other businesses. In the last quarter it closed 52 proposals: 19 won and 33 lost. It builds one row per deal with these columns, filled from the CRM export before any AI is involved:

Replace names of people and companies with the deal ID and generic labels such as "the buyer" and "a larger agency". Remove phone numbers, email addresses and anything that would identify a person. If you are not sure your plan or your client agreements allow pasting deal content into an assistant, ask first, and read the data-use settings of the plan you use. That is cheaper than a complaint.

A fixed list of themes, so counts mean something

If you ask an assistant "why did we lose these deals?" you will get a fluent essay that cannot be counted. Give it a closed list instead. A first list for a services firm might be:

• Price or budget. The buyer says it was too expensive or the money was not there.
• Competitor fit. A named alternative matched the need better or had a relevant reference.
• Slow or unclear proposal. We replied late, or the proposal did not answer the question asked.
• No decision. The project was shelved, the sponsor left, or the buyer stayed with the status quo.
• Wrong buyer or poor fit. Not the decision-maker, or outside what we do well.
• Trust and proof. The buyer wanted evidence, references or a pilot we did not offer.
• Other, with a one-line explanation.

Seven themes is plenty. More than ten and the assistant starts splitting hairs, and you will not be able to act on the output.

The tagging prompt

You are helping me review lost sales deals. For each deal below, read the notes and assign: (1) one PRIMARY theme from this list only: Price or budget; Competitor fit; Slow or unclear proposal; No decision; Wrong buyer or poor fit; Trust and proof; Other. (2) one optional SECONDARY theme from the same list. (3) a quote of no more than 20 words from the notes that supports the primary theme. (4) a confidence score: High, Medium or Low. If the notes do not contain enough to decide, write Low and say what is missing. Do not use any information that is not in the notes. Return a table with columns: Deal ID, Primary, Secondary, Evidence quote, Confidence.

[paste 10 to 15 deals at a time]

Three details do the work. The quote requirement makes you able to check each tag in seconds. The confidence score separates deals worth reading from deals the notes cannot support. And "do not use any information that is not in the notes" reduces the plausible-but-invented story that models produce when notes are thin.

An illustrative row of output: "D-0417 | Slow or unclear proposal | Price or budget | 'Proposal sent nine days after the call; the buyer had already met another agency' | Medium." The CRM said "Price". The notes suggest speed. That gap is the finding.

Reading the results: counts, then money, then the gap

Back to the agency. After tagging the 33 lost deals and checking a third of the tags by hand against the quotes, the primary themes come out as follows (illustrative figures):

The shares sum to 99% because of rounding, and the deals to 33. The headline is the last column: the CRM recorded Price 16 times but the notes support it as the primary reason in only 7. Those nine deals are where the sales team believes the problem is price while the notes point elsewhere: six to a slow or unclear proposal and three to a stalled decision. Dropping prices to fix a speed problem would have cost the agency revenue on every deal and fixed nothing.

Weighing by value matters too. Nine slow-proposal deals at $34,200 a month of potential retainers outweigh seven price deals at $24,500, even though both counts look similar. Now compare with the 19 won deals: if the median wait from first call to proposal was 3 days for wins and 9 for losses, you have a reason to act that no one argued about.

Check the AI's work before believing it

1. Read the evidence quote for every deal you intend to cite. If a quote does not support the tag, retag it by hand and note how often this happens.
2. Re-run the same batch in a new chat. If more than a fifth of primary tags change, your theme definitions are too vague. Rewrite them with an example each.
3. Look at the Low-confidence rows. If they are more than a quarter of the table, your notes are too thin and the main finding is that you need to record more.
4. Test the biggest finding with a number from the CRM that is not AI-generated, such as the days-to-proposal comparison above. A theme with a hard figure behind it is worth acting on.

Interview questions for a lost buyer, and a won one

Ask for 15 minutes. The message is short: "We're reviewing how we work with companies that considered us and chose someone else. Could I ask you three or four questions? It will take 15 minutes and I won't be selling." Send it from the most senior person who knew the buyer, a few weeks after the decision, not the same week.

Questions that work, in this order:

1. What were you trying to achieve when you started looking?
2. Who else did you consider, and how did you narrow it down?
3. What made the final decision easy or hard?
4. Was there a point where we moved ahead or fell behind? What happened?
5. What would we have needed to show or do to change your answer?
6. Is there anything about the way we worked with you before the decision that you'd change?

Avoid "Was it the price?" because it hands the buyer the easy answer. Ask open questions and stay quiet after the answer. For a won deal, use the same list with the wording turned round: "What made you choose us?" and "What nearly stopped you?" The second question is often the most useful of the lot.

Turning interview notes into themes

Take notes in the buyer's own words. After each call, paste the cleaned notes into the assistant with the same theme list and ask it to tag the primary theme, quote the buyer's own words and say whether the interview supports or contradicts the CRM reason. Do the interview first and the tagging second, since you will remember the tone of voice and the AI will not.

Five worked mini-cases of what the notes say versus what the CRM says

• "Price" that was really timing. The CRM says Price. The notes record that the buyer's budget cycle reopened in January and the proposal arrived in October. Primary theme: No decision. Action: ask about budget timing on the first call.
• "Went with a competitor" that was a missing reference. The buyer asked twice for a client in their sector and the replies did not include one. Primary: Trust and proof. Action: build a short reference set by sector.
• "Not responsive" that was a missing decision-maker. Three calls with an operations manager, none with the finance director who signed. Primary: Wrong buyer. Action: ask who signs and invite them to the second call.
• A proposal that answered a different question. The buyer wanted a three-month pilot and received a twelve-month retainer. Primary: Slow or unclear proposal, since the proposal was unclear about what the buyer asked for. Action: start proposals with a restatement of the buyer's request.
• "In-house" as the competitor. The buyer decided to hire a person instead. This is not a loss to a competitor; it is a decision about doing the work internally. Primary: No decision, secondary: Price or budget. Action: write a short comparison of agency cost and a full-time hire's cost for the next such case.

Anonymising notes: before and after

Stripping identity does not take long if you do it consistently. An illustrative note as it sits in the CRM: "Spoke to [first name] at the buyer's head office on the 3rd, who wants a pilot; the CFO is sceptical; they will email me at a work address by Friday; competitor is the bigger agency their old employer used." After cleaning: "Spoke to the main contact on day 3. Wants a pilot. Finance lead sceptical. Contact to reply within 4 days. Competitor is a larger agency the buyer has used before." Everything the theme-tagging needs is still there: the pilot request, the sceptic, the reply, the competitor. Names, dates tied to a person and contact details are gone.

Do this with a find-and-replace on your export, not by hand: replace each company and contact name with the deal ID, and search the notes column for "@", digits that look like phone numbers and any web address. Then read five rows to see it worked. An assistant can help you write the find-and-replace list or a formula from a made-up sample row, but never paste the original notes and ask it to anonymise them, because that sends it the very names you meant to remove.

Comparing won and lost on three numbers

Themes show what buyers said. Comparison shows what you did differently. Add three computed columns to the deal table and compare the medians for won and lost deals:

• Days from first call to proposal. In the agency's data: wins 3, losses 9.
• Contacts involved on the buyer side. Wins 3, losses 1.5. A deal with only one contact on the buyer side has no one to argue for you when that person goes on leave.
• Follow-ups after the proposal. Wins 2 or more, losses 0 or 1.

These are illustrative, but the exercise is real: three numbers a spreadsheet gives you in minutes, with no AI involved, often explain more than the narrative notes do. Use the assistant for text and the spreadsheet for arithmetic. If you ask a model for medians over a long table it may get them wrong, so use the sheet's MEDIAN function.

When the notes are too thin to tag

If a third of your rows come back Low confidence, do not push the model to guess. The honest finding is that your deal records cannot answer the question. Fix that going forward: add a three-line closing note to every deal, with the buyer's stated reason, the alternative they chose and one thing you would do differently. Meanwhile, interview more lost buyers, because interviews do not depend on what the seller wrote down. A quarter of thin notes and eight interviews will still produce a sound first review.

The action list: three changes, each with an owner and a date

Findings that end as a slide stay slides. Convert each into an action that someone owns, with a measure you can look at next quarter.

Limit the list to three. Anything beyond that does not get done. If the change affects your CRM fields, cleaning up a messy CRM with AI deals with the same records, and the follow-up to buyers who said "not now" can use re-engagement emails for cold deals.

A small sample that points the wrong way

An illustrative mistake: a three-person recruitment agency tagged only eight lost deals, saw four marked "Competitor fit", and decided to rebuild its service around a rival's model. A second look at the quotes showed all four came from one prospect type, a company with an in-house recruiter that had simply used its own contact list. With eight deals, four is a coincidence waiting to happen. The review should have said: not enough data, so gather another quarter and ask the four buyers directly. Set a rule before you start. Under about 12 deals, you collect hypotheses; from 20 to 30, you act on the biggest theme and verify it.

Keeping the habit cheap

The quarterly review takes a day the first time and about half a day afterwards if the data is tidy. Two habits make it lighter. First, at the moment a deal closes, either way, write one line: who decided, what the alternative was, the reason in the buyer's words. Second, keep the theme list and the tagging prompt in a saved document, changed only on purpose, so one quarter's counts can be compared with the last. If your CRM has an AI feature for summarising deals, test whether it follows your theme list before relying on it; features that sound useful in a demo may not apply a fixed list. For a look at what such features do, see which AI CRM features save time.

If you would like to work through your own deal data with someone, that is the kind of problem my AI implementation consultation covers: you can start from the contact page. Otherwise the sequence is straightforward: export, anonymise, tag against a closed list, check a third by hand, interview a few buyers, then write three actions with owners.

Questions about reviewing lost deals with AI

How many lost deals do I need for a useful win-loss review?

Around 20 to 30 deals from one product line or segment gives you themes worth acting on. With fewer than about a dozen, treat findings as hunches to test in the next calls, not as patterns. Mix in wins, since the comparison is what shows what worked.

Is the 'lost reason' field in my CRM good enough?

Rarely. It is usually a dropdown filled in by the salesperson in a hurry, and 'price' is picked far more than it is true. Use it as a starting hypothesis, then compare it with call notes, emails and what the buyer says in an interview before deciding what really lost the deal.

Can I paste client call notes into ChatGPT or Claude?

Only if your plan and your client agreements allow it. ChatGPT Business and Claude's business plans do not train on your content by default (feedback you choose to send can be an exception), but check the settings and your contracts, and remove names, contact details and figures that identify a buyer before you paste. A tagged reference number for each deal is enough.

Will lost buyers really talk to me?

Some will, and how many depends on the relationship you had during the sale. Your odds improve when the request is short, comes from someone senior and offers a fixed 15 minutes. Ask within a few weeks of the decision. Promise no sales pitch and keep that promise, because it affects whether they would ever talk to you again.

How often should I run this review?

Quarterly suits most small firms: enough deals to see patterns, recent enough that buyers remember. Between reviews, record one line on every closed deal, won or lost, covering who decided, what the alternative was and the main reason as the buyer gave it, so the next review starts with usable data.

Further reads

• What It Costs to Set Up an AI CRM for a Small Team — What a CRM with AI features costs a small team to set up.
• Is It Worth Switching CRM Just for Better AI Features? — Whether changing CRM is worth it for better deal records.
• AI Supplier Management: Track Prices, Lead Times, and Risk — Apply the same review habit to your own suppliers.
• How to Evaluate an AI Implementation Proposal or Quote — Judge a proposal the way a buyer would have judged yours.
• How to Compare Supplier Quotes Side by Side With AI — A comparable method from the other side of the table.
• AI Personalisation in B2B: Tailor Outreach to Each Trade Account — Personalise B2B trade outreach and quotes account by account using structured notes sheets and human-reviewed AI prompts to increase reply rates.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the reader's own CRM exports, call notes and buyer interviews; vendor privacy documentation for the assistants named.