You track CAC, burn multiple, and NRR by the quarter. Every one of those numbers is downstream of a question that almost never gets asked, is the company selling to the right accounts in the first place?

Sit through enough board meetings and you'll notice the pattern. When growth slows, the conversation goes straight to execution - improving pipeline coverage, rep ramp time and conversion rates. Rarely does anyone ask what's underneath all of it. 

A founder defines their ideal customer in year one, usually by looking at who's already bought, or their gut feeling as to who they’ll serve best and that definition calcifies into a slide is rarely revisited. Eighteen months and two pivots later, the sales team is still qualifying against a target list that was really just a description of an early hunch. Not to mention they’re then further diluted by when pushed into the limited available filters in traditional data platforms. 

A misaligned target list can have a knock on impact on every efficiency metric you already watch. Longer cycles, lower win rates and expansion revenue that should be there and isn't. None of those say "wrong ICP" on their own, but all of them can be caused by it. They’re also likely to get diagnosed as execution problems, then handed to a VP of Sales to fix with more reps or better enablement.

This isn’t an isolated issue only impacting a few organisations, either. Across the markets GoodFit has mapped, companies on average are missing 60% of their ICP and the accounts sitting in a typical CRM run roughly 70% outside the company's actual qualified market. 

That means most of go-to-market motions are pointed at the wrong list, generating real CAC against accounts that were never going to convert at the rate the model assumes.

Here's an audit worth sharing with portfolio CEOs to stress test ICP definitions.

Most ICP definitions are built entirely from who's already bought, which isn't the same thing as who's actually qualified. Almost nobody checks whether the same filter would also let in a list of accounts they'd never sell to.

Take the stated criteria - industry, headcount band, region - and apply it to two lists: accounts you'd sell to (qualified) and accounts you wouldn't, for whatever reason (disqualified). 

Neither list needs to be limited to companies that have actually transacted, a qualified account is one that fits, whether or not they've bought yet. A disqualified one is a genuine non-fit, whether or not they've ever been in a deal. If a similar share of both lists would pass the exact same filter, that filter isn't predictive - it's describing who you've engaged with, not separating fit from non-fit.

Not every loss belongs in this control group, though. A deal lost to timing, a change in stakeholder, or a rep who mishandled it isn't evidence the account was a bad fit, it's evidence the deal didn't close. 

Filter the closed-lost list down to losses attributed to fit (no budget, no real need, wrong operating model) before running it against the criteria. Mixing in execution losses just adds noise to the test and can end up disqualifying accounts that were never actually wrong.

One competitive-intelligence software company took this seriously enough to enrich both a qualified and a disqualified list with a deep set of company-level attributes before touching their outbound motion. Rather than defining their ICP from instinct or from whatever happened to close, they used the disqualified list as a control group, then operationalised what actually survived the test into their market mapping and prioritisation.

Firmographic filters are popular because they're the easiest attributes to pull from any data provider, but they're rarely the most predictive.

It can be useless to assess ‘if we dropped this one criterion entirely, how much would the win rate actually move?’ If the honest answer is "not much," the team is qualifying on correlation, not causation, and correlation degrades. 

A data infrastructure company selling into engineering teams found their firmographic filters told them almost nothing. The signal that actually predicted fit was structural, whether a prospect had meaningful web-scraping activity, a genuine data-engineering function, and active hiring in that area, together. None of that shows up on a standard firmographic profile. 

In their own words, even layered on top of a data provider, they struggled to reliably find or triangulate which companies in the market actually had that combination. They knew what predicted a good account but finding more of them at scale was the part that broke down.

That gap - between knowing the signal and being able to act on it across a whole market - is the one most revenue teams hit, and it's the one most relevant to how you should be reading the next two sections.

A company can be a perfect fit on paper and still be the wrong account, for reasons that have nothing to do with demand.

Does the deal clear the cost of acquiring it? A logo that takes six months to close and lands a small contract isn't a qualified account, even if it genuinely wanted the product. An MSP-tooling vendor learned this from their own data. Their real minimum viable customer was an MSP servicing 20+ end customers, translating to roughly £12k in annual spend. Below that, even a willing buyer was, in their words, "often not profitable to pursue." The threshold had existed informally inside the team for a while but it had never been written into the ICP itself, which meant sales kept chasing accounts the unit economics couldn't support.

Can the company actually deliver to this account well? Regulatory limits, language gaps, and geographic coverage tend to get treated as delivery problems rather than qualification problems, but an account that can't be served properly is churn risk. A fintech company found geography wasn't a GTM preference but a hard wall - no serving customers outside the EEA, UK, or US, full stop, because of KYC and financial-services constraints. A prospect can want the product and still fail compliance.

Both of these are CAC and NRR problems. If a portfolio company's CAC is creeping or its NRR is softer than product quality should produce, this step is one of the first places to look before assuming the problem is pricing, support, or churn-prevention motion.

Boards tend to focus diligence on whether a target list is too narrow. Lists that have grown too wide are just as common and far less discussed, because accounts get added on optimism and almost never get removed once they prove out badly.

A marketing-analytics company treated accounts tagged non-ICP as not worth sales time - until a closer look at the pipeline showed 61% of current deals were sitting in accounts marked non-ICP. The tagging had calcified rather than being corrected by results. Nothing in their process ever triggered anyone to go back and re-examine a designation once it was set, in either direction, which meant the company's own sense of its market was actively drifting from where its revenue was coming from, with no mechanism to catch it.

This is worth a fixed slot in the board cadence rather than a one-off conversation.

If a portfolio company sells into more than one region, or to more than one buyer persona, a single ICP is almost certainly masking two different problems. What predicts a win in the US frequently doesn't transfer to Europe due to different procurement norms, buying committees, commercial thresholds. A model built to cover both tends to be mediocre in both.

A marketing technology company found their first ICP model, which lumped North America and Europe/Rest-of-World into one definition, was giving them an inaccurate picture in both regions. Splitting the model in two - separate definitions per region - improved accuracy materially in each.

If a portfolio company is expanding into a new region or pursuing a second persona, this is the moment this matters most, and the moment it's cheapest to fix. 

The diagnosis usually isn't the hard part. Most revenue leaders, pressed on it, will admit their ICP is more of an assumption than a tested model.

The harder problem is operational. Standard data vendors hand you firmographics and nothing about how a company actually operates internally. Knowing that a genuine data-engineering function, a spend threshold, or a regulatory status predicts fit is one thing. Finding which of the thousands of companies in a market actually has that attribute is a different problem entirely. 

This is also where the fix becomes a recurring cost rather than a one-off project. Roughly 7% of a market's qualified accounts change every month, as companies grow into or out of fit, so a definition that was accurate at diligence is already stale by the next board meeting unless someone is re-checking it. 

For a portfolio company without the resourcing to do that continuously, this tends to show up later as the exact metrics you're already tracking: CAC drifting up as reps chase accounts that no longer fit, NRR softening because expansion accounts were never flagged as they came into range.

The accounts that fit aren't missing from the market, they're sitting in the same databases everyone already has access to, just not visible through the filters those databases were built to support.

That gap between knowing what predicts a good account and being able to find more of them at scale, is the specific problem GoodFit is built to close. The process runs in four steps: Define, Clean, Map, Grade. 

Define means testing a company's stated ICP against a disqualified list as well as a qualified one - layering in commercial, serviceability, and structural constraints, auditing existing CRM drift, and checking fit against GTM economics and segments, until what's left is a definition that's causal rather than correlational, and complete rather than partial.

Clean means starting from a market that isn't full of duplicates, dead accounts, and stale records, so the definition is being tested against reality rather than CRM debris. 

Map means sourcing every company in the market that matches the validated definition, including the ones a company's existing tools would never have surfaced because they don't show up under firmographic filters. 

Grade means ranking that full, qualified market by expected value - typically by contrasting accounts a company has won against accounts it's lost, since that's the clearest signal of who converts fastest and at what value. 

Not every loss belongs in that contrast, though. A deal lost to timing, a change in stakeholder, or a rep who mishandled it isn't evidence the account was low-value, it's evidence the deal didn't close. Filtering out execution losses before ranking keeps the grade honest.

The part that's actually hard, and the part most teams quietly give up on, is Map. A firmographic provider can tell you a company's headcount, region and industry. It can't tell you whether that company has a genuine data-engineering function with active hiring, whether it owns its fleet versus outsourcing it, or whether it's crossed the operational threshold that makes a problem worth paying to solve. Those signals live in job postings, team composition, and how a company is structured internally, not in a standard data field. 

GoodFit applies human-level reasoning at scale to read the signals that actually matter, classifying every account against them rather than relying on whatever a standard provider's database already happens to capture.

GoodFit helps revenue leaders define what a qualified account actually looks like, source every company in their market that matches it, and grade those accounts by expected value, so go-to-market shifts from a tactical guessing game into something a team can actually plan and allocate resources against with conviction.

GoodFit examines the same underlying challenge in Capturing your invisible market, a practical guide to identifying the accounts your company should be selling to, testing whether your ICP is accurate and uncovering qualified buyers missing from your CRM.