Define your ICP from deals you won
Compares closed-won and closed-lost deals criterion by criterion and returns an ICP with the deal counts behind every line.
icp-from-won-deals/SKILL.mdDownload- name
icp-from-won-deals- description
- Use this skill when you want an ideal customer profile built from evidence instead of opinion. It reads your closed-won and closed-lost deals from the CRM, or from a CSV export when there is no CRM access, compares them criterion by criterion and returns an ICP with the deal counts behind every line, plus the exclusions.
- license
- CC-BY-4.0
Define your ICP from deals you won
Most ideal customer profiles are written from memory, so they describe the deals people remember. This skill builds one from the record: it compares the deals you won with the deals you lost on industry, size, location, buyer, trigger and channel, keeps only the criteria that actually separate them, and shows the count behind each one. Use it before you build a list, set targeting rules or brief a new seller.
Inputs
- The deals: closed-won and closed-lost from the last 12 to 24 months. With CRM access, read them directly. Without it, ask for a CSV export (links in Sources).
- The fields that exist: account name, industry, employee count or revenue, country and state, deal amount, created and close dates, source or channel, lost reason, and the role of the main contact.
- What counts as a real loss: ask whether deals closed as lost for housekeeping (duplicates, tests, never qualified) should be excluded.
- Anything you already believe about your ICP, written down, so the result can confirm or challenge it.
Steps
- Pull and clean. Remove duplicates, test deals and deals with no account. Keep open deals out of the analysis. Report how many won and lost deals remain. With fewer than 10 won deals, say clearly that the result is a hypothesis to test, not a profile.
- Fill missing firmographics only with permission. If industry or size is blank, ask before looking anything up. Use the account's own website, its LinkedIn company page (employee range) or, in Brazil, the Receita Federal's open CNPJ data (CNAE activity code, share capital). Mark every filled value as "added" with its source.
- Band the values. Group employees and revenue into bands (for example 1 to 49, 50 to 199, 200 to 999, 1,000 and up), industries into five to eight groups, and locations into regions you actually sell to.
- Compare won and lost per criterion. For each value, count won and lost deals and compute the win rate: won divided by won plus lost. Add average deal size and median days to close per value.
- Keep what separates. A criterion enters the ICP only when its win rate is clearly above your overall win rate and it rests on at least five deals. Everything else is noise, however familiar it sounds.
- Read the lost reasons. Group them: no fit, price, timing, no decision, lost to a competitor. A value that keeps producing "no fit" losses becomes an exclusion.
- Find the trigger. In the won deals, look at notes and the first activity for the event that started the conversation: a new contract, a hire, a funding round, a regulation, a failed tool. Count how often each appears.
- Write the ICP and test it. Score five current open deals against it and say which ones fit and which do not.
Output
# ICP from [n] won and [n] lost deals, [date range]
Overall win rate: [x]% ([won] of [won + lost])
## Must have
| Criterion | Value | Won | Lost | Win rate | Evidence |
|---|---|---|---|---|---|
| Size | 50 to 199 employees | 7 | 2 | 78% | 7 of 9 deals in this band were won |
## Nice to have
| Criterion | Value | Won | Lost | Win rate |
|---|---|---|---|---|
## Exclusions
- [value]: [n] of [n] losses were "no fit"
## Trigger events
- [event]: started [n] of [n] won deals
## Buyer
- Main contact role in won deals: [role] ([n] of [n])
## Open deals against the ICP
| Deal | Fits | Misses |
|---|---|---|
## Limits
- Sample size, missing fields, values added from outside the CRMGuardrails
- Reads: deals, companies and contacts in the CRM, read-only, or the CSV the user shares.
- Writes: nothing in the CRM. The ICP stays in the conversation or in a file the user asks for.
- Asks before: any web lookup that sends account names to an outside tool, and any CRM change, such as adding a fit score field.
- MUST show the won and lost counts behind every criterion, never a percentage alone.
- MUST state the sample size and the date range at the top.
- MUST keep "the data says" separate from "the team believes" when the two disagree.
- NEVER fill a blank field with a guess or treat an open deal as won or lost.
- NEVER put contact names, emails or phone numbers in the output; roles are enough.
- NEVER write to the CRM without a person's yes.
Sources
- Export your records, HubSpot Knowledge Base.
- Exporting data from Pipedrive, Pipedrive Knowledge Base.
- Export a report, Salesforce Help.
- CNPJ open data layout, Receita Federal: CNAE code and share capital for Brazilian companies.
Made by ALTARA, the AI sales agent your team texts. More skills: https://www.altaratech.com/toolkit
Use it with
-
Runs in ALTARAPlaybook Research a marketMap a new market in a day with ALTARASegments, accounts that fit, who decides and the competitors already there, each number with its source. Approved accounts go into your CRM. One working dayRun it →
-
Connection Keep the CRM currentHubSpotALTARA reads one HubSpot pipeline, read-only, and sends each change with a link to the deal. Your own AI reads the same records through HubSpot's MCP. Reads (rolling out)See how →
-
Connection Keep the CRM currentSalesforceThe ALTARA team connects Salesforce on request, read access first. Your own AI reads the same records through Salesforce's hosted MCP servers. Connected on requestSee how →
-
Connection Keep the CRM currentPipedriveALTARA keeps your Pipedrive pipeline current from the chat, after your yes. Your own AI reads the same deals through Pipedrive's MCP server. Reads and writesSee how →
-
Connection Keep the CRM currentZoho CRMThe ALTARA team connects Zoho CRM on request, read access first. Your own AI reads the same records through Zoho's pre-built MCP servers. Connected on requestSee how →