How is an ICP different from a buyer persona?

A persona describes the individual you talk to — their role and motivations. An ICP describes the account worth selling to at all: the firmographics, the trigger, and the problem that make a company a strong fit, drawn from who actually buys and stays. You use the ICP to decide which deals to chase; the persona to decide how to talk once you're in.

Does this lock reps into a rigid scoring framework?

No. The rubric is 5–7 plain-language questions that predict a real deal, aligned to whatever habit you already use (BANT, MEDDIC, or your own), not a 30-field form. And the score is a filter for where to spend time, not a verdict — a rep can always override it when they know something the data doesn't.

How does Claude actually score leads?

You point it at a leads CSV with what you know about each account, and it tiers every row A/B/C against your ICP and rubric with a one-line reason, flags the time-wasters and the hot ones, and tells you what single question would settle the ambiguous rows. It's pattern-matching your own definition of a good deal across the whole list in seconds.

How do I keep the profile from baking in bias?

Define "best fit" on what actually predicts a deal — the problem, the trigger, the buying behavior — never on proxies like geography, name, or demographics. Sanity-check the finished ICP to be sure it isn't quietly encoding something it shouldn't, and remember the score is guidance a human can override, not an automatic filter.

Define your ICP and a qualification rubric Claude can score against

when to reach for this

Reps waste their best hours on deals that were never going to close — wrong size, wrong problem, no budget, no urgency — because "a lead is a lead." The fix is two written things: an ideal-customer profile (ICP) drawn from who actually buys and stays, and a qualification rubric that turns gut-feel into a few honest questions. Once both exist as files, Claude can score a new lead or a live deal against them in seconds, so the team spends its time where fit is real. This is the filter every other Sales playbook runs cleaner with.

gather this first

A list of your best customers and your worst-fit ones — closed-won that renewed and expanded, and closed-lost or churned — as files or pasted notes. The contrast is what defines the profile.
Whatever you know about each: industry, size, the trigger that made them buy, who championed it, how long the cycle ran, and why the bad-fit ones went wrong.
Your existing qualification habit if you have one — BANT, MEDDIC, or just the questions you already ask. Claude adapts your rubric; it doesn't impose a framework.

the workflow

Reverse-engineer the ICP from who actually buys and stays

In Claude Desktop, open the folder with your won/lost notes and ask in the chat — no terminal needed. Don't define the ICP from a wish; infer it from the pattern in real customers. It's far easier to recognize your best-fit buyer than to invent one from a blank page.

you ask

Read my best-customer and worst-fit notes. Reverse-engineer our ideal customer profile: the firmographics they share (industry, size, model), the trigger that makes them buy, the problem they all had, who champions it internally, and the early signals that a deal is a strong fit. Then list the anti-signals — the traits our churned and lost-fit accounts shared. Be specific and testable, not 'enterprises that value efficiency.'

what you get back A concrete ICP plus an anti-profile — "strong fit: 50–500-person services firms onboarding clients manually, triggered by a growth spurt; anti-signal: pre-product-market-fit startups shopping on price" — grounded in your actual deals.

The anti-signals matter as much as the profile. Knowing who to disqualify fast is what gives reps their time back.
Turn the profile into a plain-language qualification rubric

Convert the ICP into a short, scorable rubric — the handful of questions that actually predict a deal, in plain words, mapped to your existing habit so the team will use it.

you ask

Turn that ICP into a qualification rubric of 5–7 questions that predict whether a deal is real — covering fit, the problem's urgency, budget authority, and a compelling reason to act now. Map each to the answer that scores high vs. low, and keep it in plain language a rep can ask on a first call. If we already use [BANT/MEDDIC], align it to that.

what you get back A short, scorable rubric — each criterion with what a strong vs. weak answer looks like — that a rep can run in their head on a discovery call, not a 30-field form nobody fills in.
Score a real lead list against it

Prove the rubric works by pointing Claude at actual leads and having it tier them. This is where the filter earns its keep — and where you catch a rubric that's too loose or too strict.

you ask

Here's a CSV of new leads with what we know about each. Score every row against our ICP and rubric: tier them A/B/C, give a one-line reason, and flag the 3 that look like a waste of time and the 3 worth a same-day call. Where a lead is missing the data to judge it, say what one question would settle it.

what you get back A tiered list with reasons — "A: fits ICP, growth trigger, named budget owner; C: pre-PMF, price-shopping" — plus the disqualifiers and the hot ones, so the team works the list top-down instead of left-to-right.
Assemble the icp.md and a deal-scoring checklist

Pull it into two reusable artifacts: the profile the team aligns on, and a checklist a rep applies to any deal. Short enough to actually use mid-pipeline.

you ask

Assemble two files: an icp.md (the profile, the anti-profile, the buy triggers) and a qualify.md (the rubric as a scoring checklist with the high/low answers). Add a one-line 'when to disqualify and move on' rule at the top of qualify.md. Keep each to one page.

what you get back A paste-ready icp.md and qualify.md — the shared definition of a good deal and the checklist to score one — that the outreach engine targets from and the pipeline review judges against.

Save both. icp.md sharpens who the outreach engine targets; qualify.md gives the pipeline review an honest yardstick for 'is this deal real.'

make it your own

Sharper outreach targeting: feed icp.md into the outreach engine so it scores and prioritizes the leads CSV before you write a single opener — you reach the A-tier first.
Honest pipeline hygiene: the pipeline review gets a real test for 'is this deal real' when you score stuck deals against qualify.md — happy-ears rarely survives the rubric.
Refresh quarterly from won/lost: re-run the ICP step as new deals close and churn, so the profile tracks who's actually buying now, not who bought two years ago.
Make it a scorer (Power Track): advanced teams can save the rubric as a /qualify command or a scheduled agent that tiers each week's new leads off a fresh export (see the Features tab).

watch out for

A rubric is a filter, not a verdict. It tells a rep where to spend time, not who to ignore forever — a B-tier lead with a sudden trigger can outrank an A. Use the score to prioritize, and let a human override it when they know something the data doesn't.
Don't bake bias into the profile. "Best fit" must mean the problem, the trigger, and the buying behavior — not proxies that quietly screen out by geography, name, or demographics. Define the ICP on what predicts a deal, and sanity-check that it isn't encoding something it shouldn't.
Lead and customer data is private — scoring a real leads CSV means it carries names and context. Keep it in your approved workspace, and keep icp.md/qualify.md themselves free of any individual prospect's private details; they're a team reference, not a CRM.
Garbage in, garbage out: if your won/lost notes are thin or only capture the wins, the ICP will be lopsided. Feed in the losses and the churn honestly — the anti-profile is built from exactly the deals you'd rather forget.

you'll end up with A shared `icp.md` and a scorable `qualify.md` — the team's honest definition of a good deal — plus a tiered read on any lead list, so reps chase fit instead of noise and the pipeline tells the truth.

Questions people ask

How is an ICP different from a buyer persona?: A persona describes the individual you talk to — their role and motivations. An ICP describes the *account* worth selling to at all: the firmographics, the trigger, and the problem that make a company a strong fit, drawn from who actually buys and stays. You use the ICP to decide which deals to chase; the persona to decide how to talk once you're in.
Does this lock reps into a rigid scoring framework?: No. The rubric is 5–7 plain-language questions that predict a real deal, aligned to whatever habit you already use (BANT, MEDDIC, or your own), not a 30-field form. And the score is a filter for where to spend time, not a verdict — a rep can always override it when they know something the data doesn't.
How does Claude actually score leads?: You point it at a leads CSV with what you know about each account, and it tiers every row A/B/C against your ICP and rubric with a one-line reason, flags the time-wasters and the hot ones, and tells you what single question would settle the ambiguous rows. It's pattern-matching your own definition of a good deal across the whole list in seconds.
How do I keep the profile from baking in bias?: Define "best fit" on what actually predicts a deal — the problem, the trigger, the buying behavior — never on proxies like geography, name, or demographics. Sanity-check the finished ICP to be sure it isn't quietly encoding something it shouldn't, and remember the score is guidance a human can override, not an automatic filter.