Why Do Industry Filters Miss Most of Your Cold Email TAM?

TL;DR: Industry filters miss most of your cold email TAM because a company's self-reported industry describes who it sells into, not what it sells, and that data decays fast. The fix is to seed from proven-fit companies already in the CRM, harvest the real industries and keywords those companies appear under, then over-pull and qualify hard before paying for contacts. A filter-first approach to a category like peptide manufacturing can miss roughly half the market compared to the seed-and-expand method.

Key Takeaways

• A company's listed industry describes the market it sells into, not the product it sells, which is why industry filters systematically miss relevant companies.
• Industry classifications are self-assigned and rarely revisited, so a filter built months ago already describes a slightly different universe while the export still returns healthy row counts.
• Seeding a search from known-fit companies and harvesting the industries and keywords they actually appear under found roughly double the verified companies compared to trusting a provider's filters alone.
• On one client account, identical cold email copy sent to payment gateways produced one interested reply per 68 people from a lookalike list, against one per 356 from an industry-filtered Sales Navigator list.
• Accounts scored on firmographic fit alone convert to closed opportunities at 8.4%, against 21.3% for accounts flagged by buyer intent signals, a gap of roughly 2.5x, according to The Starr Conspiracy's B2B intent data benchmarks via Datamagnet (2026).
• Campaigns targeting well-defined ICPs achieve 68% higher ROI than broad, industry-only targeting, according to LinkedIn research reported by CXL (2025).

Industry filters miss most of your total addressable market because a company's self-reported industry describes who it sells into, not what it sells. A software company selling to real estate brokerages files itself under real estate, so an industry filter aimed at software quietly removes it. On top of that, the classification itself goes stale, because the industry field is assigned once and almost never revisited. The result is a list that is confidently wrong, and no amount of filter tuning fixes it.

How does the industry mismatch actually happen?

It happens because the industry field is self-reported and describes market, not product. Eric Nowoslawski, founder of Growth Engine X, made this point in a 16 August 2026 video about replacing provider filters with his own build: a company's listed industry is what it sells into, not what it sells.

Once you see that, the failure mode is obvious. You tick an industry box hoping to find companies that make a thing, and you get companies that sell to the people who buy that thing. The two populations overlap, but not where it matters.

The decay makes it worse over time. SIC-to-NAICS mapping is imperfect, with some codes mapping to several others, and industry codes are frequently outdated or misassigned in the first place. A filter built once in March is describing a slightly different universe by October, and nobody notices because the export still returns rows.

What does a real example of a missed TAM look like?

Nowoslawski's example in that video was "peptide manufacturing companies in the US." No such industry category exists in any provider, so a filter-first approach cannot find them at all.

His method ran like this. One seed company expanded to 14 known peptide houses. Probing the provider with those 14 seeds yielded 3 industries, 39 keywords and 42 filters that the real companies actually appeared under. He then pulled everything matching any of those values, bounded only by headcount and geography, and screened 55,000 companies down to 70 verified manufacturers. Trusting the provider's filters alone produced 32, so the seed-and-expand pass found roughly double. He puts the method's coverage at about 85% of TAM, which is his own estimate rather than a measurement.

The shape of that is the lesson. He did not guess better filters. He let known-good companies tell him what the filters should be, then over-pulled and qualified hard. That is the same logic behind how AI and automation generate leads for B2B companies: the machine is useful for screening volume cheaply, not for guessing categories.

What happens when the same email goes to a filtered list and a lookalike list?

I ran that comparison on one of my client accounts, a company that sells to payment gateways. The same four-step email sequence, word for word, went to two lists aimed at the same market.

The first list came from a Sales Navigator search: industry set to Financial Services, company size from 11 to 5,000 staff, and payment keywords such as "payment gateway" and "payment service provider". 1,069 people received the first email.

The second list was a lookalike: it started from companies that clearly fit and expanded to more companies of the same shape. 608 people received the first email.

I counted only replies from a real person saying yes, send it over, or let's talk. Out-of-office notes, unsubscribes and questions with no interest behind them do not count. The filtered list produced 3. The lookalike list produced 9. That is roughly one interested reply per 356 people against one per 68, about five times better on identical copy.

Two caveats, stated plainly. The two campaigns ran about a month apart, and the absolute numbers are small. The direction is still hard to argue with, and it matches the mechanism above: one list was built from a label, the other from examples.

It shows up in raw data too. When I run a name-and-description check over a raw provider pull, 40 to 70% of it turns out not to be in the category it was filed under.

And firmographic fit on its own is a weak predictor anyway. The Starr Conspiracy's B2B intent data benchmarks, via Datamagnet (2026), found that accounts scored on firmographic fit alone converted to closed opportunities at 8.4%, against 21.3% for accounts flagged by buyer intent signals, a gap of roughly 2.5x. Single-dimension filtering is simply too coarse to define an ICP.

In my experience, the list decides more of the outcome than the copy does, and the company list decides more than the contact list. Founders spend weeks rewriting subject lines for a list that was never going to work, when the honest fix is upstream: get the right companies first, because finding a person at a company you already chose correctly is the easy half.

What should you do instead of filtering on industry alone?

Stop choosing filters up front and let proven-fit companies define them for you. Then over-pull deliberately and qualify hard before you pay for anything.

How do you seed and expand the list?

The sequence I build to, which follows Nowoslawski's published method:

1. Seed from the client's own CRM and do-not-contact list. Closed-won, closed-lost, open opportunities. Companies they are already talking to are the only trustworthy definition of the target, because they are proven fits rather than a category guess.
2. Probe the data provider with those seeds to harvest every industry and keyword the seeds actually appear under. Use the harvested values as the pull criteria.
3. Expand the seed set with lookalike searches so the harvested vocabulary is not built on three companies.
4. Pull everything matching any harvested industry or keyword, bounded only by headcount and geography, into one database.
5. Score the whole pull with a cheap model. Per-company AI processing now costs less than paying for precise filters, so over-pulling on purpose and discarding the bad rows is the cheaper path.
6. Audit the live website for the ICP signal as the final gate, before spending anything on contact enrichment.

Two operating details make this affordable rather than theoretical. Qualification runs on my own AI subscription with agents reading real websites, not on a metered per-row API. And already-contacted company domains are suppressed before anything is paid for, across every live campaign, every archived one and the contact database, not just the last pull.

This is slow work the first time. Harvesting filters from seeds, running a screen over tens of thousands of rows and auditing websites takes real hours, and it has to be rebuilt when the ICP shifts. I say that plainly because the alternative, a five-minute export, is exactly how most lists end up broken.

Is there a filter worth keeping?

One. In a 16 July 2026 video, Nowoslawski states his standard first pull as 1% headcount growth in the last 6 months, applied on top of a client's normal filters. His reasoning: any TAM splits into companies trying to grow and companies holding station, and growth means new problems worth buying a solution for. He puts the lift at 30 to 50% on positive reply rate and is explicit that it is used silently as a filter, never mentioned in the copy.

I use a growth filter on a first campaign for the same reason, and I never reference it in an email. Telling someone you noticed they are hiring reads as surveillance, not relevance.

Beyond that, a few practices hold up:

• Cross multiple dimensions. Zeliq's guidance is to combine industry code with size bracket, legal form and geography rather than filtering on industry in isolation.
• Layer signals on top of fit. Unify describes a firmographic gate as telling you who could buy, with intent and behavioral signals telling you who is ready now.
• Write down a negative ICP. ZoomInfo recommends documenting disqualifying criteria as rigorously as qualifying ones, so lists stop burning capacity on accounts that match the filter but never convert.

The payoff for precision is documented. LinkedIn research, reported by CXL (2025), found campaigns targeting well-defined ICPs achieve 68% higher ROI than broad, industry-only targeting. Precision on the list also means you need less volume to hit the same number of meetings, which matters when a warmed mailbox can only safely send 20 to 50 cold emails a day.

How can you tell if your own list was built the wrong way?

Open 20 rows at random and read the actual websites. Count how many you would genuinely sell to. If it is under about 15, the list is a filter artefact, not a market.

Then run the rest of the check:

• Category or examples? If nobody ever looked at the closed-won accounts before building the list, the target was a guess dressed up as a filter.
• Same names every month? Repeating companies across consecutive pulls means the filter has been swept and is returning its own floor.
• What was paid for first? If contact credits were spent before anything verified the company, the spend ran in the wrong order. Company verification is cheap, contact enrichment is not.
• When does suppression run? If it runs after enrichment, part of every bill is for people who were never going to be contacted.
• Does every address pass an external verifier? A finder's own confidence score is a hint, never the gate.

None of this is exotic, and the industry knows it matters. SalesHive reports that in a 2024 survey, 67% of B2B marketers said data compliance and accuracy are their top priority. Knowing it and auditing it are different things.

If the list passes all five checks and replies are still thin, the problem has moved. That is usually the point where the offer matters more than the copy.

When is it not worth fixing your industry filters?

When the qualified pool is too small to support the number of meetings the campaign has to produce. At that point the honest move is to widen the market or change the promise, not to send harder into a thin list.

I check pool size against the promise before a campaign starts. A campaign that has to deliver a given number of meetings needs a few thousand qualified companies behind it, because reply rates and meeting rates are fractions of a fraction. Coverage is a commercial constraint, not a nicety.

Volume matters even with a perfect TAM. Research from the Ehrenberg-Bass Institute, summarized by Unify (2025), puts the in-market slice at roughly 5% of B2B buyers at any point in time. Professor John Dawes's broader point is that only a small share of any industry-defined universe is actually buying right now, which is precisely why filters alone cannot tell you who to prioritize.

Padding a thin list with marginal-fit accounts does not solve it either. Cognism's 2024 buyer research, cited by The Starr Conspiracy (2025), found sales cycles to firms outside a defined ICP run 30% to 40% longer than ICP-fit deals. You buy volume and pay for it in cycle length.

If you are not sure whether your constraint is pool size or conversion, start by checking what share of positive replies become meetings. That ratio tells you which end of the funnel is leaking.

What's next if your TAM is built on the wrong filters?

The company list decides more of your outbound outcome than the copy does, and industry filters alone are a weak way to define it. The field describes who a company sells into, the data decays fast, and a narrow filter eventually hands back the same companies month after month. The fix is to seed from proven fits, harvest the real filters from those seeds, over-pull, and qualify hard before you pay for contacts.

If you want someone to audit how your current list was built and rebuild it properly rather than patch it, book a Growth Mapping Call and we can look at it together. More about how I work is on the About page.

Frequently Asked Questions

Why does filtering by industry alone miss companies that should be in my TAM?

Because the industry field a company lists is self-reported and describes the market it sells into, not the product it makes. A software company selling to real estate brokerages gets filed under real estate, so a filter built for software vendors removes it, even though it is exactly the kind of company you want.

How fast does B2B contact and firmographic data go stale?

Because the industry field is self-assigned and almost never revisited, and SIC-to-NAICS mapping is imperfect on top of that. A filter set built once can describe a noticeably different universe within months, even though exports keep returning rows.

What does seeding and expanding a list actually involve?

It means pulling known-fit companies from a CRM's closed-won and closed-lost records, probing a data provider with those companies to find every industry and keyword they appear under, then pulling everything matching those harvested values, bounded only by headcount and geography, before qualifying with an AI screen.

Is firmographic fit enough to define a good ICP?

No. Accounts scored on firmographic fit alone converted to closed opportunities at 8.4%, against 21.3% for accounts flagged by buyer intent signals, a roughly 2.5x gap, according to The Starr Conspiracy's B2B intent data benchmarks via Datamagnet (2026). Firmographics should be layered with intent signals, not used alone.

How do you know if your filtered list has been swept dry?

Page through the list by a dimension you have not sliced before, such as state. If it returns the same companies you have already contacted, or almost nothing new, the filter recipe has been exhausted and is returning its own floor rather than fresh prospects.

Is there any single filter worth keeping on top of a seeded list?

Yes, a headcount growth filter. Eric Nowoslawski's standard first pull uses 1% headcount growth over a recent period, which he reports lifts positive reply rate by 30 to 50% when used silently, never mentioned in the email copy itself.

When should you not bother fixing your industry filters?

When the qualified pool is simply too small to support the number of meetings a campaign needs. At that point the honest move is to widen the market or change the offer, not keep refining filters on a TAM that cannot support the target volume.