Turn your last 10 wins into your next 40 targets.
Builds a win profile from your closed-won deals, weighing behavioral signals over firmographics, and keeps only the traits that separate wins from losses. Then scores lookalike accounts against it and writes a shared-situation first-touch angle for each.
In plain English: “Who looks like my best customers?”
Inside: The win-profile extraction, the wins-vs-losses contrast test, the scored lookalike list, and the per-account angle.
Install it in one line, or paste it in.
~/.claude/skills/ and runs automatically when it is relevant.Connect your context. Set it to your motion.
it reads wins and losses automatically and pulls the fields you forgot to export.
sources the lookalike candidates and fills the behavioral signals, hiring, stack, recent changes.
adds pre-sale usage as a profile signal, which is often the strongest one you have.
This was built for a B2B SaaS org sourcing outbound from a win profile. Set these to your stack:
The loss set is not optional. Without it you cannot tell a win signal from a coincidence, and the profile becomes a description of your market, not your winners.
| Set this | What it is | Default / Example |
|---|---|---|
| N_WINS | how many wins to profile | last 15 (10 to 20 works) |
| LOSS_SET | the losses to contrast against | closed-lostsame windowsame segment |
| LOOKALIKE_SOURCE | where candidate accounts come from | Clay tableCommon Rooma list you paste |
| SCORE_THRESHOLD | minimum score to make the list | 70 of 100 (re-tune to your volume) |
Everything the skill does, in full.
Takes your last N closed-won deals and turns them into a win profile, then into a scored list of net-new accounts that match it, each with a first-touch angle you can send this week. Where closed-won-analysis is the retro that tells you why you won, this skill consumes that read and produces the forward artifact: the scored lookalike list and the per-account opener. The profile is not just firmographics. It is what was true at the account before the deal started: what they were hiring for, what tool they were replacing, what changed. Those are the signals a rep can actually act on.
- 1Win-profile extraction (behavior first)
Read every win and pull two layers. The firmographic shape: segment, size, industry, region. Then the layer that matters more: the behavioral and situational signals present before the deal started. What roles were they hiring for. What tool were they replacing. What changed at the account, a new leader, a funding round, a stack migration. When we backtested our scoring model against real won and lost deals, the strongest single predictor of a win was a behavioral signal at 3.56x lift; the firmographic fit score everyone trusted came in at 1.13x. The profile weighs behavior over firmographics because the data votes that way.
- 2Signal vs coincidence (wins against losses)
A trait counts only if it appears in wins meaningfully more than in losses. "Most of our wins are mid-market" is noise if most of your losses are too. Run every candidate trait through the contrast: share of wins that show it, share of losses that show it, and the ratio. Keep the traits that separate. Drop the rest, even the ones the team believes in.
- 3Lookalike list build (source and score)
Source candidate accounts that match the profile from LOOKALIKE_SOURCE, excluding current customers and open pipeline. Score each one against the profile, weighting the traits by how strongly they separated wins from losses. Every score shows its inputs: which traits the account hit, which it missed, and where each read came from. Accounts above SCORE_THRESHOLD make the list.
- 4First-touch angle (the shared situation)
For each account on the list, write one opener built on the shared situation, not the shared logo. The shape is "companies that looked exactly like you chose us when X", where X is the situational signal the account matches: the hire, the replaced tool, the change. Never name a customer without written permission. The angle works because the situation is recognizable, not because the name is impressive.
- A profile trait counts only if it separates wins from losses, not wins from thin air. Show the contrast for every trait that makes the profile.
- Every lookalike score shows its inputs: the traits hit, the traits missed, the source of each read. A bare number is not a score.
- Never name a customer in outreach without written permission. The default angle is the anonymous shared situation.
- No angle drafted from a signal the skill cannot source. A guessed situation is worse than a generic opener.
WIN PROFILE · built from 15 wins vs 18 losses (sample data)
Trait In wins In losses Keep?
Hiring RevOps or SalesOps 67% 22% YES (3.0x)
Replacing spreadsheet/DIY 53% 17% YES (3.1x)
New sales leader < 6 months 40% 11% YES (3.6x)
Mid-market, 100-500 staff 73% 61% NO (1.2x, noise)
LOOKALIKE LIST · 3 of 41 candidates above threshold 70
Account Score Traits hit First-touch angle
Northwind 88 RevOps hire, DIY stack, new VP New sales leader + spreadsheet ops
Vertex 79 RevOps hire, DIY stack Hiring RevOps onto a manual stack
Acme 74 New VP, DIY stack New leader inheriting DIY tooling
Next move:
1. Northwind first. It hits all three separating traits.
2. The angle for all three: "companies that looked exactly like
you chose us when the new leader hit the spreadsheet wall."
Where an operator takes this next.
The list is step one. Here is where an operator takes it once the manual version proves out.
The profile decays as your product and market move. Rebuild it every quarter or it becomes last year's playbook.
Point a scheduled Claude task at Salesforce and your enrichment tool weekly, rebuild the profile as new wins close, and rescore the candidate pool so the list never goes stale.
Feed the kept traits into Clay, or run Ocean.io-style lookalike sourcing through your enrichment tool, so candidates arrive pre-matched to the profile instead of hand-picked.
Connect Common Room so the behavioral signals include who is showing up in your community and content before the deal exists, often the earliest separating trait you have.
Push each account's score and traits-hit into a CRM field so reps see the why in the record, not in a doc they never open.
One skill is the on-ramp.
A single skill does one job. Chained into a playbook, or run as a full build, it becomes a system. Here is where this one plugs in.