Personalization Blog | Best marketing strategies to grow your sales with personalization

ICP Refinement Playbook: How to Use Won/Lost Data to Sharpen Your Ideal Customer Profile

Written by Jimit Mehta | Apr 30, 2026 7:18:58 AM

Every B2B SaaS company has an ICP document. Most of them are aspirational rather than empirical: built from the founder’s intuition, adjusted by marketing’s content strategy, and eventually ignored by the sales team who go after any company that will take a meeting.

The problem with a hypothesis-based ICP is that it gets stale. Markets shift. Your product evolves. The customers who were a perfect fit at Series A look different from the customers who drive the best retention metrics at Series C. And yet the ICP doc sits in Notion, unchanged, gradually diverging from reality.

Refining your ICP with closed data changes the relationship. Instead of describing who you think you should sell to, you are describing who you have actually sold to successfully and why. The result is a sharper targeting model, lower customer acquisition costs, and better retention because you are acquiring customers who are genuinely suited to your product.

This playbook walks through the full refinement process.

When to Do an ICP Refinement

ICP refinement is not a once-a-year exercise. The right cadence depends on your stage and the speed at which your market and product are changing.

Do a full ICP refinement when: - You have just hit 100 paying customers and have enough data to build a statistically meaningful view of what your best customers look like - Customer churn is running above your target and you suspect you are closing wrong-fit accounts - Your win rate is declining and you cannot explain it by changes in competition or pricing - The sales team and marketing team are arguing about lead quality (this is almost always an ICP alignment problem) - You have launched a major new product capability that changes who you are most valuable to

Do a quarterly ICP audit when: - You have a defined ICP and want to check whether the data from the most recent quarter is consistent with the existing definition - You have onboarded a meaningful number of new customers and want to see if the cohort looks like your ICP

Step 1: Pull and Segment Your Closed-Won Data

Start with your CRM. Pull all closed-won deals from the last 12 to 18 months. For each deal, capture:

  • Company name and domain
  • Industry and sub-industry
  • Employee count at time of close
  • Annual recurring revenue at time of close (or estimated)
  • ACV of the deal
  • Sales cycle length (days from opportunity created to closed-won)
  • Lead source (how did the account first come to us?)
  • Which product or package they bought
  • The primary use case they bought for
  • Contract length

Now layer in the customer success data: - Current retention status (active, churned, expanded, contracted) - Current ARR (to see if they expanded or contracted from the initial deal) - NPS or CSAT score if available - Time to first value (how long did it take them to activate and get results?)

Create three customer cohorts:

Cohort 1 (your best customers): High ACV, expanding ARR, high NPS/CSAT, fast time to first value, short sales cycle. These are the customers you want more of.

Cohort 2 (average customers): Close to median on all metrics. Retained but not growing, average NPS, average sales cycle.

Cohort 3 (problematic customers): Churned, contracted, low NPS, long time to first value, or involved in significant support escalations.

Step 2: Find the Patterns in Your Best Customers

Analyze Cohort 1 in detail. Look for the patterns that distinguish your best customers from the average.

Firmographic patterns:

  • What industries are most heavily represented?
  • What is the employee count range?
  • What is the revenue range (estimated)?
  • What geographies?
  • What is the funding stage or ownership structure (bootstrap, VC-backed, PE-backed, public)?

Run this analysis in a spreadsheet if your CRM does not support it natively. Create a frequency distribution for each dimension. You are looking for clustering: if 70% of your best customers are Series B to Series C B2B SaaS companies, that is a strong ICP signal. If they are evenly distributed across all stages, that dimension may not be a useful ICP filter.

Technographic patterns:

What technology do your best customers use? Specifically, what CRM, marketing automation platform, analytics stack, and category-adjacent tools do they have?

Pull technographic data from an enrichment provider or from your own product integration data. If your product has integrations, look at which integrations your best customers have enabled. A company that has both Salesforce and Marketo and has enabled your Salesforce integration is probably a better fit than a company running spreadsheets.

Behavioral patterns:

How did your best customers find you? What content did they consume before becoming a customer? How many touchpoints were in the pre-sale journey? How quickly did they move from first contact to close?

These patterns tell you about the buying behavior of your best customers, which informs your demand gen strategy and your ABM target account selection.

Step 3: Analyze Your Closed-Lost Data

Most companies do a thorough analysis of their closed-won data and a superficial analysis of their closed-lost data. That is backwards. Your closed-lost data is where you learn the most.

Pull all closed-lost deals from the last 12 months. For each lost deal, capture: - Reason for loss (as logged in the CRM and as reported by the rep) - The stage at which the deal was lost - The firmographic profile of the account - Who the deal was lost to (competitor, no decision, budget freeze, other)

Cluster the losses by type:

Type 1: Wrong-fit companies that never should have been in the funnel. These are companies that do not match your ICP and were brought in by aggressive top-of-funnel or a rep who was chasing quota. Identify their characteristics so you can exclude them from targeting.

Type 2: Right-fit companies lost to a competitor. These are ICP-appropriate accounts that evaluated you and chose someone else. These losses are valuable: they tell you where you have product or positioning gaps for accounts that should be winnable.

Type 3: Right-fit companies that went to no-decision. Budget freeze, executive change, competing internal priorities. These are not really losses in the same sense. They are deals that paused. A no-decision from a right-fit account is worth putting in a re-engagement nurture sequence.

The critical question from closed-lost analysis: Are the companies you are losing to competition the same companies that look like your best customers? If yes, your ICP is right and you have a positioning or product gap to address. If the companies you are losing look different from your best customers, your sales process may be bringing in wrong-fit accounts.

Step 4: Run Win/Loss Interviews

Quantitative data tells you what happened. Win/loss interviews tell you why.

Target 15 to 20 interviews per quarter: split roughly evenly between recent won accounts, recent lost accounts, and churned customers.

Interview structure for won accounts:

  • “What was the trigger that started the evaluation?” (What changed internally or externally that made you look for a solution?)
  • “What other solutions did you consider, and what was the final deciding factor?”
  • “What did you expect to be true six months after implementing, and how does reality compare?”
  • “What would you tell a peer at a similar company about us?”

Interview structure for lost accounts:

  • “What were you trying to solve for, and why did our solution not fit?”
  • “What would have made the decision go differently?”
  • “What solution did you choose, and what made it the right fit?”

Interview structure for churned customers:

  • “When did the product stop delivering value, and what changed?”
  • “What would have kept you as a customer?”
  • “What are you using now and how is it different?”

Record and transcribe these interviews. Look for recurring themes. Three win interviews where customers mention the same deciding factor is a signal worth acting on. Three churn interviews where customers cite the same unmet expectation is a product or onboarding gap that affects your ICP definition.

Step 5: Update the ICP Document With Evidence

At this point you have quantitative firmographic and behavioral data from your deal history, qualitative insight from win/loss interviews, and a clear view of who churns versus who expands. Use all of it to update the ICP definition.

The ICP document structure that actually gets used:

A usable ICP document is not a long essay. It is a set of specific, matchable criteria organized by priority:

Primary criteria (must-haves): The dimensions that are present in nearly all of your best customers and absent in most churned accounts. If every best customer you have is a B2B SaaS company with 50 to 500 employees using Salesforce, those are primary criteria.

Secondary criteria (strong signals): Dimensions that are present in many best customers but not universally. Company growth rate over 20% year over year, specific industry verticals, funding stage.

Negative criteria (disqualifiers): Attributes that consistently appeared in churned or lost accounts. Companies below a certain revenue threshold, companies in industries you do not serve well, companies without a relevant technology foundation.

Add customer quotes. The most effective ICP documents include verbatim quotes from your best customers describing why they bought and what value they get. Quotes make the ICP feel real and grounded rather than theoretical.

Step 6: Operationalize the Updated ICP

An ICP refinement is only valuable if it changes what the company does. Build the updated ICP into your operating systems.

Update the CRM scoring model. If you have an account scoring model, update the ICP fit dimensions and weights to reflect the refined definition. Rescore the active account database.

Update the outbound targeting criteria. Work with marketing and the SDR team to update the criteria for which accounts enter the ABM target list and the outbound sequence pools.

Update the qualification criteria. Work with sales to update the MEDDIC or qualification framework fields that reflect the ICP. If the new ICP includes a tech stack requirement, add it as a required qualification field in the discovery call.

Update the marketing messaging. The ICP refinement will often reveal that your best customers are buying for a reason that your current messaging does not fully reflect. Update your homepage messaging, core email templates, and sales deck to reflect what your best customers actually say about why they buy.

Step 7: Build a Quarterly ICP Review Cadence

ICP refinement is not a one-time exercise. Build a quarterly cadence for checking whether your ICP definition is still accurate.

The quarterly ICP check runs through four questions:

  1. Did the deals we closed this quarter look like our defined ICP? (Run the firmographic comparison)
  2. Did we win or lose accounts that do not match the pattern? (New signals to investigate)
  3. Did any customers in the last quarter churn, and what did they have in common?
  4. Are there any market signals (new competitor positioning, new use cases emerging in customer calls) that suggest the ICP should evolve?

This check should take 60 to 90 minutes and involve input from the sales and customer success teams. Document the conclusions and any changes to the ICP definition.

A living ICP document that gets updated quarterly based on evidence will diverge significantly from a static document over 12 to 18 months. The divergence is a good thing: it means the ICP is evolving with the market rather than reflecting a hypothesis from two years ago.

Frequently Asked Questions

How many closed-won deals do you need before an ICP analysis is meaningful? With fewer than 20 closed-won deals, an ICP analysis will identify patterns but you will not have enough data to distinguish signal from coincidence. The patterns you see should be treated as strong hypotheses, not conclusions. With 50 or more closed-won deals, the analysis becomes statistically meaningful enough to act on. With 100 or more, you have enough data to run cohort analysis and isolate the dimensions with the strongest predictive relationship to deal quality. What if your best customers are all in the same industry by chance? It may not be chance. Early customers are often acquired through founder networks and early partnerships, which can produce industry clustering that reflects business development history rather than true product-market fit. When you do the ICP analysis, look at whether the industry concentration is accompanied by other shared attributes (tech stack, company stage, use case) that would explain the fit beyond industry alone. If the industry is the only clustering dimension, it may be worth testing accounts in adjacent industries explicitly to see if the fit holds. How do you handle ICP disagreement between sales and marketing? Surface the disagreement with data. Bring both teams to a review of the closed-won and closed-lost analysis and let the data drive the conversation. If sales is pursuing accounts that are not converting or are churning after close, that is a data point that changes the conversation. If marketing is generating leads from segments that sales cannot close, that is similarly a data point. The goal is not to declare a winner. The goal is to build a shared understanding of what the data says about where the best customers actually come from. Should the ICP be different for different product lines or packages? Yes, potentially. If you have a self-serve or SMB product and an enterprise product, the ICP is likely different for each. Define ICPs at the product or package level rather than at the company level. The ICP for your starter plan may be a 10 to 50 employee company using HubSpot. The ICP for your enterprise plan may be a 500-plus employee company using Salesforce with a dedicated RevOps function. Separate ICPs for separate segments produces more actionable targeting criteria than a single ICP that tries to cover too much ground.