How do you run a win/loss analysis with Claude?

Pull the last twelve months of closed-lost deals out of HubSpot with their lost reason, amount, the stage they died at, how long they were open, and the notes and call summaries attached to them. Pull a smaller set of closed-won deals too, so there is something to compare against.

Give that to Claude, either through the HubSpot connector or as an export, and ask it to group the losses into a handful of patterns, with the deals listed as evidence under each one.

Then fix the thing that made this hard in the first place. Turn your closed lost reason into a dropdown built from the patterns you found, and make it required when a deal moves to Closed lost.

It is an afternoon of work, and the second time you run it, next quarter, the data is much better.

Most sales teams have a closed lost reason field. Almost nobody reads it, because what is in it is a mess: "price", "went quiet", "timing", "chose competitor", and a lot of blanks. The useful story is sitting one layer down, in the notes and call summaries, where nobody has time to read four hundred of them. That is the part Claude is good at.

Before you analyse anything, close out the deals that are dead but still open. This two minute video shows how to do it in HubSpot.

Why "price" is almost never the real reason

When a rep closes a deal as lost, they pick the reason that is quickest to type and least awkward to explain. Price is the polite answer. So is timing.

Read the notes on those same deals and a different picture tends to show up. The champion left. Nobody senior was on a call after the first demo. The proposal went out three weeks after the meeting. The buyer asked the same integration question twice and never got a straight answer.

None of that fits in a one-word field. You only see it by reading the deals, which is why this suits Claude.

What to pull out of HubSpot

Take a smaller set of closed-won deals from the same period with the same fields. Losses on their own tell you what went wrong. Losses next to wins tell you what the winners had that the losers did not.

Connector or export?

Either works.

The HubSpot connector for Claude reads deals, companies, contacts and engagements, including notes, calls and meetings, straight from your portal. It needs a paid Claude plan (Pro, Max, Team or Enterprise), and a Super Admin or someone with App Marketplace Access has to connect it. It respects HubSpot permissions, so a rep who connects it only sees the deals they can already see. For a win/loss review across the whole team, the person running it needs access to every pipeline.

If you would rather not connect anything yet, export the deals and notes to a spreadsheet. It is clunkier, and it lets you strip out anything you do not want Claude to see first.

What happens to your data when Claude reads HubSpot is in where your data actually goes when Claude reads your CRM, and how we set the connector up with limits is in we plugged HubSpot into Claude with MCP.

The privacy bit, because deal notes are full of people

Deal notes contain names, phone numbers, the occasional comment about someone's health or their boss. Under the Privacy Act that is personal information, and the obligations follow it into whatever tool you paste it into.

The OAIC's guidance on privacy and the use of commercially available AI products says it plainly: privacy obligations apply to any personal information put into an AI system, and to what comes out. It recommends organisations do not put personal information, and particularly sensitive information, into publicly available AI tools, and it asks you to check whether the product's terms give the developer access to what you input.

In practice, for this job:

• Use a business Claude plan, whose commercial terms you have read, and never a free consumer chatbot.
• Strip what you do not need. Claude does not need a buyer's mobile number to spot that deals stall after the proposal stage. If you are exporting, delete contact names and emails before you upload.
• Check the purpose. APP 6 says you use personal information for the purpose you collected it for, unless an exception applies. Reviewing your own sales deals to sell better is close to home, but it is worth a line in your AI policy so nobody has to guess.

The prompts

Run these in order, in the same conversation. The first one does most of the work.

1. Group the losses

I've given you our closed-lost deals from the last 12 months, with
the lost reason the rep chose, the amount, the last stage, the open
and close dates, and the notes and call summaries.

Read the notes, not just the lost reason. Group the losses into
5 to 8 patterns that explain why we actually lost. For each pattern:
- a one-line name
- how many deals and their total value
- 3 deal IDs that are the clearest examples, with the line from the
  notes that shows it
- whether the rep's chosen lost reason matched what the notes say

If a deal has no notes, put it in a separate "no evidence" group.
Don't guess.

2. Compare against the wins

Now read the closed-won deals from the same period. For each loss
pattern, tell me what the won deals did differently at the same
stage. Only use things you can point to in the notes, with deal IDs.

3. Find where they die

Break the losses down by the last stage before Closed lost and by
days open. Which stage loses the most value? Is there a point, in
days, after which our win rate drops away? Show me the numbers you
used.

4. Draft the new dropdown

Based on the patterns, draft a closed lost reason dropdown with no
more than 8 options. Each option should be something a rep can
choose honestly in 5 seconds. Write a one-sentence definition for
each, so two reps would pick the same option for the same deal.

Read the output like a sceptic

Claude is fast and good at reading, and it can still overreach. Three checks before anything goes in front of the sales team:

1. Open five of the example deals yourself and read the notes. If the quote Claude pulled does not say what Claude says it says, throw that pattern out.
2. Check the counts add up. Every deal should be in exactly one group, including the "no evidence" group.
3. Look at the size of the "no evidence" group. If it is a third of your losses, your biggest finding is that reps are not writing notes. That is worth knowing too.

Fix the field so next quarter is easier

If your closed lost reason is a free-text box, this is where it stops being one. Turn it into a dropdown using the options from prompt 4, and add a separate multi-line field for the detail, so the story still has somewhere to go.

Then make it required. In HubSpot, go to Settings, then Data Management, then Objects, choose Deals, open the Pipelines tab and pick the pipeline. Hover over the Conditional logic rules column and click Add rule, choose your Closed lost stage, add the property and tick Required. HubSpot's pipeline setup guide has the full steps, and you need permission to edit property settings to do it.

From then on nobody can close a deal as lost without picking a reason from a list that means something. If your stages themselves need work while you are in there, this is our take on HubSpot deal stages.

Take it to the sales meeting, then do it again

Pick the two patterns that cost the most money and decide one change for each. If deals die after the proposal goes out, look at how proposals are written and how fast they go (here is how to check a proposal against what the client asked for). If deals die when nobody senior is involved, change who joins the second meeting.

Run the same prompts next quarter. The lost reasons will be clean, and you can see whether the change you made actually worked.

Which pattern do you think will come out on top in your pipeline, and would you bet on it before you run the prompts?

Frequently asked questions

How many closed-lost deals do I need for a win/loss analysis with Claude? Enough that a pattern can show up more than once. If you only have a couple of dozen losses in a year, read them yourself. Claude earns its keep when there are more deals than a person has time to read properly.

Can Claude update the closed lost reason on old deals? Yes, through the HubSpot connector, which can update deal records up to 10 at a time. Have Claude show you the proposed change for each deal first, and only apply it once you have checked a sample.

Is it safe to give Claude our deal notes? Deal notes contain personal information, so the Privacy Act applies. Use a business Claude plan whose terms you have reviewed, strip names and contact details you do not need, and follow the OAIC's guidance on commercially available AI products.

Should I include closed-won deals? Yes. Losses on their own tell you what went wrong. Comparing them with wins from the same period tells you what the winning deals did differently at the same stage, which is the part you can act on.

Sources

• Set up and use the HubSpot connector for Claude, HubSpot Knowledge Base, for what the connector can read and update, the 10 record limit, and plan and permission requirements.
• Guidance on privacy and the use of commercially available AI products, OAIC, October 2024, for how the Privacy Act applies to information put into AI tools.
• Set up and manage object pipelines, HubSpot Knowledge Base, for requiring properties when a deal moves stage.

---

Want a second pair of eyes on what your lost deals are saying? Talk to us and we will run it with you.