Ideal Customer
Profile (ICP).
A practitioner-written reference on ICP: the five attributes that matter, three worked B2B SaaS examples, the most common mistakes, and how scoring frameworks work. Built from 12 years of running ICP exercises inside developer-first companies.
By Daria Dovzhikova · Updated May 2026
TL;DR
- An ICP is the narrow definition of which companies your product is built for, tight enough to land on a sample of named accounts where the same message, motion, and pricing all work.
- A complete one combines five attribute types — firmographic, technographic, behavioral, triggering event, and decision-making unit — and gets built from closed-won and closed-lost clustering plus customer-development interviews, not the founder's gut.
- The usual failures are setting it too broad, skipping the triggering event, confusing buyer with user, and never revisiting it; the page walks three worked B2B SaaS examples and an ICP scoring framework.
What an ICP actually is
An ideal customer profile is the narrow definition of the type of company your product is built for. It's the answer to "who do we sell to, specifically?" A well-built ICP should narrow your addressable market to a sample of 1,000-5,000 named accounts where roughly the same messaging, motion, and pricing work across them.
The concept got formal treatment in B2B SaaS via Lincoln Murphy's ICP work and was popularized through HubSpot's inbound methodology and Lenny Rachitsky's practitioner essays. It sits between market segmentation (broader) and buyer persona (narrower). Most B2B SaaS companies need exactly one ICP at a time per product line; trying to operate two parallel ICPs usually dilutes both motions.
ICP is the first artifact of a real go-to-market strategy. Every downstream decision (positioning, pricing, channels, sales enablement) depends on getting this right. Skip it and you build a motion that works against an imaginary customer.
The five attributes that matter
A complete ICP combines five attribute types. Skip any of them and the motion underperforms in predictable ways.
Firmographic
Company size band (employees + ARR), industry vertical, geography, growth stage. The objective floor; gets you to a sample of ~1,000 companies you could plausibly sell to.
Technographic
Existing technology stack, cloud provider, key tools they already pay for, integration points. Critical for DevTools and any product that fits into an existing tool chain.
Behavioral
How they currently solve the problem (manually? competitor? homegrown?), trial signup vs demo request, signal sources (Hacker News presence, OSS commits, GitHub stars).
Triggering event
What changed about their business in the last 90 days that makes them buyable now? Funding round, executive hire, contract renewal, regulatory change, scaling event. ICP without a trigger is a forecast, not a target.
Decision-making unit
Who's the champion (often technical), the economic buyer (often non-technical at enterprise scale), the influencers, the blockers. Different DMUs for the same firmographic ICP mean different sales motions.
ICP vs buyer persona vs TAM/SAM/SOM
Three artifacts that get mixed up constantly. They answer different questions, get used by different teams, and break in different ways. Six axes that surface the differences and the failure modes.
| Axis | ICP | Buyer persona | TAM / SAM / SOM |
|---|---|---|---|
| Question answered | Which type of company is the right buyer? | Inside that company, who decides, champions, and blocks? | How big is the market opportunity in dollars? |
| Level of detail | Firmographic + technographic + behavioral + trigger + DMU shape | Role, goals, day-to-day pains, content habits, objections | Top-down dollar estimate (total / serviceable / obtainable) |
| Scope | A cluster of 1,000-5,000 named accounts where the same motion works | One specific human inside the ICP, usually 2-3 personas per ICP | The entire addressable market, not a specific account or buyer |
| Primary use case | Targeting, account scoring, sales territory design, channel selection | Messaging, content strategy, sales enablement, demo design | Board decks, fundraising, market-entry decisions |
| Refresh cadence | Quarterly + after major product or funding events | Every 6-12 months as the buyer-side org structure shifts | Annually for board materials, otherwise rarely |
| Common misuse | Set too broad (a market segment, not an ICP) | Built from one customer the founder remembers fondly, not from interviews | Used to justify a GTM motion the underlying ICP doesn't support |
Axes synthesized from Lincoln Murphy's ICP work, the OpenView SaaS Benchmarks body of work, and 12 years of running ICP exercises inside developer-first companies.
Three worked B2B SaaS ICPs
Real ICP write-ups from the kinds of engagements gtm-labs runs. Each shows the company stage + segment, the ICP definition, and the reasoning.
Series B observability platform
100-500 engineer companies on Kubernetes spending $5K+/month on Datadog, where a Senior SRE is the technical champion and VP Engineering owns the budget. Triggering event: Datadog renewal within 60 days. North-American mid-market focus.
Narrow enough that outbound SDRs can identify 500 named accounts inside the segment. Datadog spend signal is detectable via intent data (Bombora, G2). The renewal trigger gives sales a clear urgency hook.
Pre-Series-A AI eval platform
AI/ML teams at 50-300 person Series A-to-C startups deploying LLM-based features to production, where ML Engineering Lead is the champion and CTO is the buyer. Triggering event: first production incident from LLM nondeterminism. US + EU.
The triggering event ('we just shipped an LLM feature and a customer complained about hallucinations') is concrete. ML engineering leads are reachable through dev communities, AI/ML conferences, and Substack newsletters. The ICP is small (maybe 2,000 named accounts) but high-intent.
Seed-stage developer-first compliance tool
20-100 engineer SaaS companies preparing for SOC 2 Type II or ISO 27001 audit, where a Founding Engineer or Director of Engineering owns compliance and the founder writes the check. Triggering event: enterprise prospect blocked deal pending SOC 2.
Triggering event creates urgency (a stalled deal is worth ~$50K to fix). Companies in this band tend to find compliance tools through Hacker News, IndieHackers, and other founder communities. Self-serve trial conversion likely works at this scale, no SDR needed.
How to build an ICP from scratch
The reliable approach combines top-down clustering (from closed-won analysis) and bottom-up interviews (from customer development). Most teams skip one half and end up with an ICP that's either statistically valid but unactionable, or actionable but wrong.
- Closed-won analysis (1 week).Pull the last 20-50 closed-won deals. Tag each by firmographic + technographic attributes. Cluster. Identify the segment with the biggest revenue contribution, the shortest sales cycle, and the lowest churn. That's your ICP starting hypothesis.
- Closed-lost analysis (3 days).Same exercise on closed-lost. The pattern of who you don't close tells you who's outside the ICP, often more useful than the closed-won pattern alone.
- Customer development interviews (2 weeks). 10-20 interviews inside the ICP cluster: champion, buyer, influencer. Ask about the triggering event, the alternatives considered, the buying-committee dynamic. Most ICP refinement happens here.
- Triggering event identification (3 days). Cluster the triggering events from interviews. The most common 2-3 triggers become the targeting signals for outbound and the urgency hooks for sales conversations.
- Validation (ongoing). Run new outbound and inbound through the ICP filter for 90 days. Track conversion rates inside vs outside the ICP. If conversion inside is less than 3× conversion outside, the ICP needs more narrowing.
Total elapsed time for a from-scratch ICP build: 4-5 weeks for a team that has 20+ closed-won deals; 6-8 weeks for pre-product-market-fit teams that need to interview prospective rather than actual customers. Pre-PMF teams should follow Steve Blank's customer-development methodology rather than the closed-won-first approach.
Common mistakes
ICP that's too broad
'B2B SaaS companies between 50 and 5,000 employees' is not an ICP. It's a market segment. A real ICP should narrow to a sample of 1,000-5,000 named accounts where the message you'd send is roughly the same. If your sales team can't list 50 accounts in the ICP off the top of their head, it's too broad.
ICP built from the founder's gut, not customer interviews
The founder usually has a specific company in mind when they describe the ICP and assumes it generalizes. It doesn't. Run 20+ discovery interviews across closed-won and closed-lost deals; let the ICP emerge from clustering the wins, not from intuition.
Confusing buyer with user
Developer-first products often have a technical user (the developer) and a non-technical economic buyer (the VP Engineering or CFO). Both are part of the ICP. Marketing usually optimizes for the user; sales sometimes optimizes for the buyer; the ICP needs both.
No triggering event
An ICP without a trigger gives you targeting, not timing. Companies are usually only buyable for a short window after a triggering event (new hire, funding round, contract renewal, incident). Add the trigger and your outbound conversion rates roughly double.
Failing to revisit
ICPs evolve as the product and market change. Quarterly revisits are standard practice. Most companies who 'know their ICP' actually know what their ICP was 18 months ago when they last ran the analysis.
By the numbers
For developer-first products, the ICP's technographic + behavioral attributes have to map to accounts that actually expand. One data point that shows where ICP precision proves out downstream — and a reminder that the broader GTM motion depends on the ICP being narrow enough to operationalize.
Net dollar retention top-decile public SaaS hits per Bessemer's Cloud Index. NDR above 100% is the downstream proof an ICP was defined correctly — you sold to accounts that expand instead of churn. A sharp ICP shows up here months later.
Bessemer Cloud Index · 2026→Each figure links to the primary source. If a number moves in a subsequent annual report, this page gets updated.
FAQ
What is an ideal customer profile (ICP)?
An ideal customer profile is the narrow definition of the type of company your product is built for. It combines firmographic attributes (size, industry, geography), technographic attributes (tech stack), behavioral signals (how they currently solve the problem), a triggering event (what makes them buyable now), and the decision-making unit (champion, economic buyer, influencers). A good ICP should narrow to a sample of 1,000-5,000 named accounts where the same message and motion can win.
What's the difference between an ICP and a buyer persona?
An ICP describes the company; a buyer persona describes a person within that company. ICP comes first: it answers 'which companies do we sell to?' Buyer personas answer 'inside those companies, who decides and who blocks?' Most B2B SaaS companies need one ICP and 2-3 buyer personas (the champion, the economic buyer, the influencer). The ICP should be tight enough that your buyer personas are roughly consistent across accounts.
How do you build an ICP from scratch?
Start with closed-won analysis. Take your last 20-50 closed-won deals and cluster them by firmographic + technographic attributes. Identify the cluster that's biggest, has the shortest sales cycle, and has the lowest churn. That's your starting ICP. Then run 10-20 customer-development interviews inside that cluster to surface the triggering event and the buying-process detail. Total time: 2-3 weeks. Pre-PMF companies without 20 closed-won deals should build the ICP from customer interviews of the buyers they want, not the buyers they have.
How narrow should an ICP be?
Narrow enough that a sales team can identify 1,000-5,000 named accounts inside the segment AND that the same message + motion works across them. If your ICP requires three different messaging approaches for three sub-segments, it's actually three ICPs and you should pick one to start. Common mistake: founders set the ICP too broad ('B2B SaaS 50-5,000 employees') because they're afraid to leave deals on the table. Narrower ICPs convert better, churn less, and produce repeatable motions.
How often should you revisit your ICP?
Quarterly at minimum, plus after every major product launch and every funding round. ICPs drift as the product evolves, as the market changes, and as your customer base grows beyond the original cluster. Quarterly revisits should re-run the closed-won analysis and validate the triggering event is still active. Major launches usually require a full ICP rebuild since the new product targets a different segment.
What's an ICP scoring framework?
ICP scoring is the practice of scoring inbound or outbound accounts against the ICP attributes to prioritize sales effort. Common framework: score each account on 5-7 ICP attributes (firmographic, technographic, behavioral, trigger), weight by importance, and produce a 0-100 score. Accounts above 75 get high-touch sales follow-up; 50-75 get self-serve nurture; below 50 get deprioritized. Tools like Clearbit, ZoomInfo, and 6sense automate the scoring. Don't over-engineer the math; the value is in the discipline of consistent prioritization, not the precision of the score.
Need help nailing yours?
ICP work is in every engagement.
The GTM Diagnostic ($1,500-$2,500) and GTM Clarity ($5,000-8,000) both include a from-scratch ICP build with closed-won analysis and customer-development interviews. See services or take the diagnostic.
Take the diagnosticReady when you are.
Discovery calls are 20 minutes. First one's on me.