Sales & CRM

How to run a win/loss analysis with AI

Sales & CRM3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

A win/loss analysis is a systematic review of why deals closed or fell apart — interviews with buyers, patterns across lost opportunities, themes from won accounts. Done well, it tells you whether you're losing on price, product gaps, sales execution, or timing. Most operators know they should run one. Few do it consistently, because it requires pulling data from multiple sources, talking to real people, and synthesizing qualitative signal into something the team can actually act on.

AI feels like a natural fit here because the hardest part isn't collecting the data — it's making sense of it. You have call transcripts, CRM notes, email threads, and post-mortem conversations that are all over the place. An LLM can read across all of that, spot recurring themes, and draft a structured summary faster than any analyst. The analysis is fundamentally a pattern-recognition problem on messy text, which is exactly what these models are good at.

ChatGPT, Claude, and Gemini can genuinely help with win/loss analysis today. You can paste in CRM notes, deal stage histories, and interview transcripts and get coherent theme extraction, competitor mention tallies, and draft reports. Claude handles longer documents well. ChatGPT's Custom GPTs let you set a consistent analysis template. Gemini integrates with Google Workspace if your notes live in Docs. None of them connect to your CRM or call tool automatically — but if you're willing to do the data prep manually, the analysis itself is tractable.

Skip the manual work → See the Starch version

Sales & CRM3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ClaudeChatGPTGemini

Step-by-step

1 Export your lost and closed-won deals from your CRM (HubSpot, Salesforce, whatever you use) for the past 90 days. Pull the deal name, stage history, close reason, deal size, and any notes your reps logged. Export to CSV or copy-paste the relevant fields into a doc.

2 Pull call transcripts or meeting notes for the deals you want to analyze. Gong, Chorus, and Otter all have export options. If transcripts don't exist, write a short structured interview template (5-6 questions) and send it to the buyer contacts you still have access to.

3 Open Claude and paste in your first batch of deal notes and transcripts — start with 10-15 deals to stay within context limits. Ask it to identify the top 3-5 reasons deals were lost or won, flag recurring competitor mentions, and note any objections that appeared across multiple deals.

4 Ask the model to organize its output into a structured template: win themes, loss themes, competitor mentions, objection patterns, and deal size correlations. Copy this output into a doc or spreadsheet you'll use as your working analysis.

5 Repeat the paste-and-analyze process for the next batch of deals. Because LLMs don't retain memory between sessions, you'll need to manually merge the outputs from each batch — look for themes that appeared across all batches, not just within one.

6 Use ChatGPT or Claude to draft the final write-up: an executive summary of the top findings, recommended actions for the sales team, and product gaps flagged by lost buyers. Paste your merged theme notes in and ask for a structured report.

7 Distribute the report manually — copy it into a Notion page, email it to your team, or drop it in Slack. Set a calendar reminder to run the same process again next quarter.

Prompts you can copy

Here are CRM notes and call transcripts for 12 lost deals from Q1. Identify the top 4 loss reasons, note any competitor that came up more than twice, and flag the most common objection by deal stage.

Analyze these 10 closed-won deal notes. What patterns explain why these deals closed? Focus on timing, buyer role, deal size, and any product capabilities the buyer specifically cited.

I'm building a win/loss report for my sales team. Here are my merged findings across 30 deals. Write a 500-word executive summary with three recommended actions — one for product, one for sales, one for pricing.

Compare these two sets of deal notes — one group of lost deals and one group of won deals, all in the same ICP segment. What's different about how the conversations went, what objections came up, and how quickly deals moved through stages?

Here are five buyer interview responses from deals we lost to [Competitor X]. Extract the specific gaps or concerns they mentioned and categorize them as: pricing, missing feature, trust/credibility, timing, or other.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your CRM — every run starts with a manual export, and if your deal notes are incomplete, the analysis reflects that gap.

Context window limits mean you can't analyze 90 days of deals in one pass; you're batching and manually merging outputs, which introduces inconsistency in how themes get labeled.

Nothing persists between sessions — the structured template you carefully built last quarter isn't what you get next time unless you re-paste it as a system prompt every single run.

Transcripts and call recordings live in separate tools (Gong, Otter, Zoom) with no direct connection to the LLM; gathering them for each analysis cycle is a recurring manual task.

Output structure drifts — the model may categorize loss reasons differently across batches, making it hard to track whether 'pricing objection' counts went up or down quarter over quarter.

The finished report lives wherever you paste it; there's no version history, no automated distribution, and no trigger to run the analysis again — it's a one-off until you remember to repeat it.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — it builds and runs the win/loss analysis workflow as a persistent app connected to your live CRM data, call notes, and email threads, so you're not re-assembling the same inputs manually every quarter.

Connect HubSpot or Apollo.io once through Starch's integration catalog — the agent queries your live deal data, stage history, close reasons, and contact records every time the analysis runs, with no manual export.

Start from the Sales Agent CRM app, which already pulls HubSpot and Apollo data into a structured surface. Describe the win/loss view you want on top — by rep, segment, deal size, or competitor — and the agent builds it.

Starch syncs your Gmail or Outlook threads on a schedule, so email context from the deal — including late-stage objections buyers raised in writing — is available to the analysis without you hunting through your inbox.

Tell Starch: 'Every Monday, pull closed-won and lost deals from the past 30 days, extract win/loss themes, and post a summary to Slack.' That automation runs on its own — no calendar reminder, no re-prompting.

Because the app persists, win/loss patterns accumulate over time. You can ask 'did pricing objections increase this quarter versus last?' and get an answer grounded in your actual deal history, not a one-session snapshot.

Describe a custom analysis surface in plain English — 'show me a dashboard of loss reasons broken down by ICP segment and deal size, updated weekly' — and Starch builds and refreshes it without you writing a line of code.

Get closed-beta access →

Toolkit