Layer Firmographic Filters

Outcome: the account universe cut to addressable accounts, with each filter’s removal rate recorded and every disqualified row retained rather than deleted.

Surface: App and MCP server
Level: Intermediate
Uses: enrich_organization · formula columns
Credits: ~0.5 per company for enrichment
Prerequisite: Lesson 03's sourced universe

Enrich once, filter many times

Take the full firmographic bundle in one enrich_organization call rather than as separate columns. Same data, one lookup, one charge — and every filter after it is a free formula.

That is the shape of this lesson: one paid step, then a series of free cuts.

Reliability, and what it means for filtering

Field	Reliability	How to filter on it
Country	High	Hard filter
Founded date	High	Hard filter, with a rule for blanks
Headcount	Good, but a floor	Band filter, never an exact threshold
Industry	Medium — heuristic classification	Soft: use to score, not to exclude
Revenue	Low for private companies	Score input at best. Never a hard filter.
Funding	High when present; absence is ambiguous	Positive filter only

Two rules follow, and they matter more than any individual value:

Hard-filter only on reliable fields. Excluding on a modelled revenue estimate removes real accounts on the strength of a guess.
Absence is not a value. “No funding recorded” means unfunded or unreported, and treating it as “unfunded” quietly deletes bootstrapped companies that may be your best segment.

A hard filter on headcount thresholds is the most common quiet error here. Headcount is usually derived from professional-network presence, so it undercounts in markets and industries where that presence is thin — a filter of “over 50 employees” can remove a 200-person manufacturer entirely. Band, do not threshold.

Filter order

Cheapest and most decisive first.

Free exclusions first

Suppression list, competitors, partners, existing customers, unsupported regions. Costs nothing, and removes rows that should never have been in the universe.

Then geography

Usually the largest single cut, and the most reliable field you have.

Then size band

The second largest cut. Band, with a documented rule for missing values.

Then everything else

Founded date, funding, industry — as soft filters or score inputs rather than hard cuts.

Record the removal rate at each step. The sequence of numbers is your record of what the definition actually does, and it is what you check when the market size changes next quarter.

Missing values need an explicit rule

For every filter, decide what happens when the field is blank, and write it in the filter itself.

Field blank	Sensible default	Why
Country	Keep, flag for review	Often a data gap, not a foreign company
Headcount	Keep, flag	Common in smaller and non-English markets
Founded date	Keep	Rarely load-bearing
Industry	Keep	Classification gaps are frequent
Domain	Drop	Cannot be enriched or deduped — genuinely unusable

The pattern: keep and flag, do not silently drop. A blank field is a data problem, and dropping on it systematically removes exactly the segments where coverage is thinnest.

Keep the disqualified rows

Filtering is a status column, not a delete.


qualified        = passes all hard filters
disqualified_for = the first filter it failed

Three reasons this matters:

Re-qualification. A company that grows into your size band next quarter is already in the table, already enriched — it just crosses the line.
Diagnosis. When the addressable count drops, disqualified_for tells you which filter changed.
Cost. Re-sourcing a deleted company means paying for it again.

Do this now

Enrich the universe once

The full bundle, on every sourced row.

Apply the free exclusions

Suppression, customers, competitors. Record the removal rate.

Apply geography, then size band

Recording removal at each step.

Write the missing-value rule per filter

Into the filter itself, not into a note.

Build the status columns

qualified and disqualified_for.

Compare the survivor count to lesson 02’s addressable estimate

A large gap means an assumption was wrong. Find which filter.

Read 20 disqualified rows

If any obviously should have qualified, the filter is wrong, not the data.

Check your work

One enrichment call, many free filters
No hard filter runs on modelled revenue
Every filter has a written missing-value rule
Disqualified rows are retained with a reason
Survivor count is close to your addressable estimate

Where this breaks

Deleting disqualified rows makes every future run more expensive and destroys your ability to explain a change in market size. Six months later the addressable count has moved, nobody can say why, and re-creating the comparison means re-sourcing and re-enriching the whole universe. Filter with a status column. Never with a delete.

Further automation

The disqualified_for column becomes a monitoring surface: counts per reason, tracked run over run. A rising “too small” count means your sourcing drifted toward SMB; a rising “no domain” count means a data-quality problem at intake. Both are worth catching early.

Next lesson

05 — Tech-stack and traffic fit, the qualification layer that a directory filter cannot express.

Reference for this lesson: enrich_organization, Company, Company total employees, Company revenue.