ICP & market mapping
Decide exactly who to sell to, and turn that decision into filters a machine can run.
Roadmap: GTM foundations & technical baseline, Data, enrichment & signals, Portfolio & interview readiness
Why it matters
Every system a GTM engineer builds sits downstream of one decision: who are we trying to reach? A vague ICP ("mid-market SaaS") produces huge, expensive lists, generic copy and low reply rates, and no amount of enrichment or AI fixes it. The engineering part is translating a sales team's instinct into explicit, testable criteria: firmographics you can filter on, signals you can detect, and exclusions you can enforce. Hiring managers notice quickly whether a candidate starts from the customer or from the tool.
What it's made of
- Mining closed-won and closed-lost deals for patterns
- Separating firmographic fit from timing (intent) and from persona
- Writing an ICP as filterable criteria, not adjectives
- Sizing the market (TAM) bottom-up with real counts
- Tiering accounts (A/B/C) and matching effort to tier
- Mapping buying committees: champion, economic buyer, blocker
- Writing exclusions: customers, open deals, competitors, bad-fit segments
- Turning pains per persona into messaging hypotheses
Start from deals, not opinions
Export the last 12–24 months of closed-won and closed-lost opportunities. For each, note industry, headcount, region, tech stack, the title of the person who started the conversation, sales cycle length and deal size. Patterns show up fast: maybe 70% of won revenue comes from companies with 200–1,000 employees using a specific tool, while sub-50-person companies churn within a year.
If the company is too young to have deals, use the best 10–20 customers or design partners and interview the people who sold to them. Label the result as a hypothesis and plan to test it.
Three separate questions
Mixing these up is the most common ICP mistake. Keep them as separate columns in your data:
- Fit: could this company ever buy? Industry, size, geography, tech stack, business model. Mostly stable.
- Timing: is now a good moment? Hiring, funding, new leader, tool change, expansion. Changes weekly.
- Persona: who inside feels the pain and who signs? Titles, seniority, department.
Write criteria a machine can run
Rewrite every criterion until it maps to a field a provider or your CRM actually has. "Fast-growing" becomes "headcount up 20%+ in 12 months". "Serious about sales" becomes "5+ people with SDR or BDR titles". "Uses modern tools" becomes "website shows tag X or job posts mention tool Y".
Then size it bottom-up: run the filters and count. 4,200 accounts × an average of 3 relevant contacts is about 12,600 people. At 50 new contacts a day per sender you know how many senders and months that market represents, and whether you need to widen or narrow.
Tier, then match effort to value
Not every account deserves the same treatment. A common split:
- Tier A (top 50–200): researched by hand or by an agent plus human review, multi-threaded, sometimes ads and direct mail.
- Tier B (next 1,000–3,000): signal-triggered sequences with AI-assisted personalisation on one or two lines.
- Tier C (long tail): light-touch, segment-level messaging or nurture, or skip entirely.
- Always-excluded: customers, open opportunities, partners, competitors, do-not-contact lists.
Mid-level looks like
Builds a sensible ICP from what sales says, with industry, headcount and titles, and can pull a list that matches it.
Senior looks like
Derives the ICP from won/lost data, sizes each segment with real counts, tiers accounts by expected value, and revisits the definition when results disagree with it.
Common pitfalls
- Defining the ICP by what the product could serve rather than who actually buys, renews and expands.
- Using criteria no data provider can filter on ("innovative", "growth-minded").
- Treating every account in the TAM the same, so your best accounts get the same thin effort as the long tail.
- Forgetting exclusions, then emailing current customers or accounts in an active deal cycle.
Practice drills
Drill 1
You're given 120 closed-won and 200 closed-lost deals as a CSV. Produce a one-page ICP with fit criteria, personas and exclusions, each one tied to evidence in the data.
A strong answer: Criteria mapped to filterable fields, win rates or revenue share per segment, a stated sample-size caveat, and explicit exclusions.
Drill 2
A founder says the ICP is "B2B companies that care about data". Turn that into filters, size the resulting market and propose three tiers.
A strong answer: Concrete filters, a bottom-up count with assumptions labelled, tier sizes with different effort levels, and questions back to the founder about what they've actually sold.
Drill 3
Map the buying committee for a product sold to finance teams at 500-person companies. For each role, write their main pain and the one sentence that would get their attention.
A strong answer: Distinct pains per role (not the same pitch reworded), a clear champion vs signer, and a likely blocker with a plan for them.
Best resources for this skill
Repos, courses, articles and podcasts, ranked by how much they teach.
Read next
Fanatical Prospecting
Jeb Blount
A blunt guide to the daily prospecting work your tooling is meant to amplify. Read it to understand what reps actually need from lists, sequences and call blocks.
SPIN Selling
Neil Rackham
Research-backed questioning for large, complex sales. It helps you design discovery fields, call notes and AI summaries around the questions that matter.
The Challenger Sale
Matthew Dixon, Brent Adamson
Argues that teaching buyers something new beats relationship building in complex deals. Useful context for writing outbound copy and insight-led sequences.
Gap Selling
Keenan
Frames every deal as the gap between a buyer's current and future state. A practical lens for deciding what research and enrichment fields a rep needs before the first call.