Most ideal customer profiles are wish lists. They describe the company the founder imagined selling to, written before there were customers, and never revisited once there were.
A useful ICP is the opposite: a description derived from evidence about who has already succeeded with you, precise enough to tell a rep which accounts to skip.
What an ICP is and isn't
An ICP describes an organisation. A persona describes a person inside it. They answer different questions — which doors to knock on, and what to say when someone answers — and collapsing them produces a document that does neither job.
An ICP is also not your total addressable market. TAM is everyone who could conceivably buy. Your ICP is the slice where you win disproportionately, sell faster, discount less, and keep the customer.
The test of an ICP is what it excludes. If it disqualifies nobody, you've written a market description.
Build it from your own data
Start with customers, not opinions.
Step 1 — pick the right customers to learn from. Not your biggest logos. Your best ones: renewed at least once, expanded or referred, reached value quickly, few support escalations. A large unhappy customer teaches you what to avoid.
Step 2 — find what they share. Look across four dimensions:
| Dimension | What to look for |
|---|---|
| Firmographic | Industry, headcount, revenue, geography, funding stage |
| Technographic | Systems they run that make you a fit or a fight |
| Behavioural | How they bought, who drove it, how long it took |
| Situational | What was happening when they decided to act |
Step 3 — talk to them. Ten conversations teach you more than any amount of CRM analysis. Ask what triggered the search, what they tried first, who pushed internally, and what nearly stopped the purchase. That last answer is the most valuable thing in the exercise and it never appears in your data.
Step 4 — check the negative. Look at churned and lost-to-competitor accounts. What did they share? Anti-patterns are as actionable as patterns and considerably more uncomfortable to write down.
The part most ICPs miss: triggers
Firmographics tell you an account could buy. Trigger events tell you they might buy now, and that distinction is the entire difference between a static list and live pipeline.
Triggers worth tracking:
- A new executive in the function you sell to — new leaders buy, typically within two quarters
- Funding rounds — budget appears and pressure to deploy it follows
- Relevant hiring — job postings are the most honest public statement of a company's priorities
- Expansion into a new market or segment
- Public complaints about a competitor
- Regulatory or compliance deadlines in their industry
An account matching your firmographics with an active trigger is worth ten matching accounts without one.
Write it so a rep can use it
The output should be short enough to hold in your head. If it needs a slide deck, it won't get applied.
Fits when:
- B2B SaaS, 50–500 employees
- Series A through C
- Already runs [specific system]
- Has a dedicated ops or enablement function
- Trigger: new revenue leader, or hiring for the function in the last 90 days
Skip when:
- Under 20 employees (no budget owner, long unpaid sales cycle)
- Heavily regulated without a compliance sponsor
- Deep in a multi-year contract with an incumbent
- No one owns the problem internally
That "skip when" list is where the money is. Every account it removes is research hours, demo time, and pipeline noise you don't spend.
Score it so it ranks
A binary in-or-out ICP wastes the middle. A simple weighted score works better: assign points per criterion, weight by how strongly each predicts a win in your historical data, and rank the list.
The discipline that keeps this honest is refusing to score criteria you can't observe. If you can't reliably tell whether an account has an internal owner for the problem, it doesn't belong in the model — an unverifiable criterion silently becomes a guess with a number attached.
This is also where the ICP stops being a document and becomes infrastructure: scored accounts can be ranked automatically, which is how Twin-Sales decides which companies to surface first rather than handing you an undifferentiated list.
Revisit it quarterly
ICPs decay. The product changes, the market moves, you learn to serve a segment you previously couldn't. The common failure is an ICP written at founding, still hanging in the playbook three years later, sending reps after accounts that fit a product you no longer sell.
Once a quarter, compare closed-won against the ICP. If your best deals keep coming from outside it, the ICP is wrong — not the deals.
Frequently asked questions
What's the difference between an ICP and a buyer persona?
An ICP describes the organisation worth selling to — industry, size, business model, technology, trigger events. A persona describes the individual inside it — their role, what they're measured on, what they fear. You need both: the ICP tells you which doors to knock on, the persona tells you what to say when someone answers.
How do I build an ICP with only a few customers?
Use the ones you have, but weight them honestly. Look at which accounts renewed, expanded, or referred — not just which ones signed. With fewer than ten customers you're forming a hypothesis rather than a conclusion, so write it down explicitly as one and revisit it every quarter as evidence accumulates.
How often should I update my ICP?
Review it quarterly and rewrite it whenever your win-rate data disagrees with it. The most common failure is an ICP written at founding and never revisited while the product moved on — you end up hunting accounts that fit what you used to sell.
Put this playbook to work
Twin-Sales reads your market, finds the accounts worth pursuing, writes outreach in your voice, and runs your pipeline from first touch to closed deal.
Try Twin-Sales free →