What Makes a Good ICP?

Ask 5 revenue teams for their ICP and you get 5 saved filters: location, industry, headcount, revenue band, and tech stack.

That’s how you get 0 meetings from 800 connections a month.

Most teams blame the copy, tool, or data, but the ICP is where it breaks. Here’s a deep dive on creating an ICP that gets you results faster.

What is an ICP?

An ideal customer profile is a description of the type of company that gains the most value from your product and is likeliest to buy it. It’s built from company-level patterns in the customers you’ve closed rather than from a wish list:

• Firmographics – industry, employee count, revenue, location
• Technographics – software stack
• Situational details – budget cycle, decision-makers, procurement length, current pains and priorities

Situational details are what most ICPs leave out. 

A profile built only on firmographics tells you which companies resemble your customers. A profile that tells the situation tells you which ones are in a position to buy this quarter.

There are 3 distinctions that keep teams out of trouble: 

1. An ICP describes a company. 
2. A buyer persona describes a person inside that company, usually the person who makes the buying decision. 
3. A target audience is the widest group who could plausibly buy from you.

None of them are interchangeable. 

Your ICP decides which accounts get attention. Your persona decides what you say to the finance lead versus the head of operations. Your target audience is a market sizing input.

Most ICPs are database filters with a nicer name

Open a random ICP doc, and you’ll find 3–6 attributes that any contact database can query. 

IBM’s customer profile follows the same shape, framing profiling as the work of collecting information about segments: demographics, psychographic data, purchasing preferences, and pain points. 

That’s a fair description of what most teams produce. But it’s also the ceiling.

While a complete set of attributes describes a population, it doesn’t explain a purchase.

The attributes that are easiest to pull up are the ones every data provider already indexes, which means a competitor can rebuild your list in an afternoon. The attributes that predict a close are the ones nobody has indexed: 

• Workflow that broke last quarter
• New VP with 90 days to show a number
• Compliance deadline with a real date on it

Competitors have already commoditized anything easy enough to query, so it can’t be the thing that differentiates your targeting from everyone else’s.

That’s why you need judgment. 

Judgment answers why 1 company bought while 9 others that looked identical on paper never moved.

AiSDR runs on AiSDR

Every test, written up

We use our own product to book our own meetings. Every experiment gets written up, numbers and all.

Drop your email and the full log opens.

We publish the losing tests too

4 ICP criteria that separate a profile from a filter

Run your current profile through these 4 questions before the next campaign goes out. Each one has a failure mode you’ve probably already lived through.

Does it name a trigger rather than a trait?

Traits describe who could buy at some point in the next 5 years. Triggers describe who has a reason to move now. 

A trigger is the clause that answers why now:

• Funding round
• Leadership change
• Hiring spike on the team that owns your problem
• Migration off a tool you replace
• Renewal date you can see coming

86% of B2B purchases stall somewhere in the process, and fit on its own does nothing to stop that.

The failure mode: Your list is technically correct and mostly dormant. 

Everyone on it could buy eventually, so your team works it at the same intensity all quarter and gets a reply rate that reflects the average rather than the moment. Layering buyer intent data on top of a trait list is how most teams close this gap, and it’s also the cue to revisit the profile when the triggers stop firing.

Would it disqualify a company you want to like?

An ICP that excludes nothing is a wish list with a nicer font. This is the fastest test to run, and the one most teams fail. Writing down who you turn away costs something politically.

Most founders get their first ICP wrong. The correction almost always runs toward something narrower. Every team landed on at least 3 attributes to describe the profile, which is a specificity floor rather than a ceiling.

The failure mode: The enterprise logo everyone wants in the deck. 

It matches on industry and headcount, it takes 11 months to close, it churns in year 2, and nothing in your profile said no. You spend a lot of time and resources only to lose them in the end.

Can your team act on it without asking a follow-up question?

This is the operational test: If a salesperson has to interpret the profile before using it, marketing and sales are working from 2 different lists, and neither knows it.

Buying groups now run from 5–16 people across as many as 4 functions. 

Tailoring content to the buying group lifted consensus by 20%, while tailoring it to an individual buyer pushed consensus down by 59%. A profile that names only the company leaves your team guessing at the functions in the room.

The failure mode: The profile says “mid-market SaaS with a modern stack.” 

The same sentence produces 2 different account lists. The campaign that follows reads as a messaging problem. Asking questions risks taking any conversations back to the starting line.

Does it explain why your best customer closed rather than what they look like?

This is the causal test, and it takes years and several wins under your belt to answer. 

The answer is usually a sentence rather than a firmographic: 

• “The just lost the person who did this.”
• “The board wants a number they can’t hit.”
• “The tool they were using stopped supporting their work.”

Most profiles describe what the best customers look like. Very few say why they bought. Realistically, that’s the only part that transfers to a company you haven’t met yet.

The failure mode: You can list your top 10 accounts and describe none of their reasons. 

Getting to that sentence takes pattern recognition across a few hundred deals, which is why this test separates teams with tenure from teams without it.

A FILTER ISN’T A PROFILE

Ami reads the situation, not just the headcount. Funding, hiring, tech changes, who’s moving.

SEE AMI CHOOSE →

How to build an ICP from data you already have

Most guides on how to create an ICP start with firmographics. 

Try starting at the other end instead. The inputs that carry the best signal are the ones you already own.

There’s a floor under all of this. Below a certain number of closed deals, you’re pattern-matching on noise. 3 wins in the same vertical is a coincidence wearing a trend’s clothes.

That’s the actual reason judgment takes years to build. A salesperson who’s watched a few hundred deals close has crossed that floor without ever naming it. A team 6 months into selling hasn’t, no matter how sharp the read looks in a slide.

The threshold moves with your deal count and cycle length, so an early profile is a hypothesis you’re testing rather than a decision you’ve made. Rerun the exercise every quarter until the pattern stops changing.

Keep the output in a fixed format so every team works from the same document.

Ideal customer profile examples that pass the 4 tests

Most ideal customer profile examples stop at the trait layer. These 5 carry a trigger and a disqualifier as well, which is what makes them usable on a Tuesday morning.

Now here’s the same SaaS row written the way a profile should read.

Trait layer: B2B SaaS companies with 50–200 employees and $5M–$20M in revenue, running HubSpot or Salesforce, selling into North America and Western Europe.

Trigger: They posted a sales hire in the last 90 days. That job description is the company saying out loud that pipeline is now somebody’s problem, and it comes with a start date you can plan around.

Disqualifier: Founder-led sales with no CRM. These companies buy quickly and churn quickly, because there’s no system for the tool to plug into and no one whose job depends on it working.

Why they closed: The first sales hire arrived with a number and no list. Every deal in this segment closed in the window between that person starting and their first quarterly review, and you build the profile around defending that window.

Why they closed is something a database can’t produce. It comes from reading the deals.

Where the judgment comes from when you haven’t had the years

Everything above assumes you’ve watched enough deals close to see the pattern. But many teams, especially new startups, haven’t. And years aren’t something you can buy or shortcut.

AiSDR trained Ami on what those years teach: 

• What a good ICP looks like
• What messaging lands and what gets deleted
• When a campaign is working 
• When it needs a rewrite

It’s learned all this from over 14,000 campaigns AiSDR has run, built alongside the operators who ran them.

Give it your business, and it decides who’s worth chasing, writes the outreach, launches it, reads what comes back, and rewrites what isn’t working, unprompted. You set the destination and clear the runway. Something that already knows the route does the flying.

FAQ

What makes a good customer profile?

A good customer profile names a trigger, excludes companies you’d otherwise chase, reads clearly enough that a salesperson can act on it without asking a follow-up question, and explains why your best customers bought. Attribute lists cover the third requirement only. The other 3 are what turn a description into a targeting decision.

What is an ICP example?

A usable example reads like this: B2B SaaS companies with 50–200 employees running HubSpot, who posted their first sales hire in the last 90 days, excluding founder-led teams with no CRM. The trait layer sizes the market, the trigger says why now, and the disqualifier keeps the list honest.

How many ICPs should you have?

Start with 1. Teams at scale can support 2 or 3, but each one needs its own trigger, its own messaging, and someone accountable for it. If you can’t staff them separately, what you have is a segmentation exercise inside a single ICP.

What is the difference between an ICP and a buyer persona?

An ICP describes the company you want to sell to, using firmographics, technographics, and situational context. A buyer persona describes an individual inside that company, including their role, priorities, and what blocks them. You use the ICP to choose accounts and the persona to choose what to say once you’re in.

Your outbound has a memory now

Meet Ami, that remembers what we tried, what worked, and uses it to decide what deserves another shot

put ami to work →