Use case — ICP definition & market sizing

Turn a fuzzy "ideal publisher" into a precise ICP — then size how many actually exist

Most affiliate teams carry the ideal partner in their head, not on paper. We convert that instinct into a testable signal set, calibrate it against a reviewed slice, and measure the true count across 120M+ classified domains — so you plan against a real number instead of a guess.

tens of thousands–hundreds of thousands

qualifying publishers a full run typically sizes for a single vertical (estimate)

~96%

precision on the reference client's reviewed sample (~4% error)

1,000

domains in a typical calibration slice — the measured basis

Request a Free Pilot See Free Samples

The job to be done

"We know a good publisher when we see one." Now write it down — and count them.

A prose description of your ideal partner cannot be run against the web. It has to become a set of signals a machine can extract and a human can audit. Two questions block every recruitment plan until that happens.

What exactly is "ideal"?

Which signals are non-negotiable, which merely help, and which quietly disqualify a site? Until that is explicit, every reviewer applies a slightly different bar and the list drifts.

How many of them exist?

Is the addressable set 4,000 publishers or 140,000? The answer decides headcount, budget, tooling, and whether the channel is worth building at all — yet most teams have never measured it.

How the pipeline solves it

From an instinct to a calibrated, sized ICP — in five steps

Each step is explicit and inspectable. You see the draft definition, the reviewed slice, the measured error, and the extrapolated size — before a single dollar of budget is committed.

Elicit the signal set

In a working session we translate your "ideal publisher" into required, preferred, and disqualifying signals drawn from a fixed vocabulary — plus any custom fields your program needs. The definition is written down first.

Pull the calibration slice

We draw a representative ~1,000-domain random slice from your vertical inside the 120M+ classified-domain database and screen it with the draft ICP. First-pass output is ready in hours, not weeks.

Review & tune the wording

Your team marks each sampled domain fit / not-fit. Disagreements are the signal: they show which definitions to sharpen. We adjust wording and thresholds and re-screen the slice.

Lock precision, measure error

The loop repeats until measured precision clears your bar — in the reference engagement the reviewed sample landed at ~96% precision, roughly a 4% error rate. The number is measured, not asserted.

Extrapolate the size

The fit rate on the reviewed slice, applied to the full category count, yields the addressable publisher estimate — labeled as an estimate, with the calibration basis shown alongside it.

Hand over a sized ICP

You leave with a documented signal set, a measured precision figure, and a full-run yield projection — the inputs a recruitment plan actually needs.

Worked example

A calibration loop, in numbers

An illustrative loop for a mid-market advertiser recruiting independent content publishers. The slice figures are the measured basis; the full-run figure is an extrapolation, labeled as an estimate.

Draft ICP → calibrated ICPexpressed as a required / preferred / disqualifying signal set

Required — all must hold

Relevant to the vertical Editorial / blog structure Independently run

Preferred — boost the fit score

Affiliate links present Fresh content Newsletter

Disqualifying — excluded on match

Sells own inventory / competitor Thin AI-spun filler

Round 1: draft ICP on 1,000 domains

The first pass over the calibration slice flagged the usual keyword-not-intent errors — competitor storefronts that read as "relevant", and templated affiliate pages with no real editorial. Reviewed precision started well below the bar.

Round 2–3: tune the wording

We tightened the "independently run" and "editorial structure" definitions and hardened the own-inventory disqualifier. Each re-screen moved measured precision up and cut the false-positive rate.

Result: ~96% precision, then size it

Once the reviewed slice held at ~96% precision (~4% error), the fit rate was applied to the full category count to project the addressable publisher universe — reported as an estimate with the slice as its basis.

The measured basis

One reference engagement, end to end

In a large-scale production run on our own classification platform, the same define-calibrate-size loop ran across the travel vertical. The client reviewed a sample of the output against their own judgment.

travel-related domains identified

domains ICP-screened in one run

precision on a reviewed sample

domains in the source dataset

30M

source domains

1.73M

travel-related

1.2M

ICP-screened

~96%

precision confirmed

Reference: a large-scale production run on our own classification platform. Your own sizing figures come from your ICP and your reviewed slice, not from this run.

A note on confidentiality

Any publisher profile we show on a public page is drawn from a real screening run, so we follow standard confidentiality practice: domains are withheld and descriptions generalized. The published profiles are therefore intentionally not traceable to the sites — including through a web search — which protects the publishers without changing the underlying data. Your pilot and client deliverables contain the actual domains with every signal, so everything can be verified directly. Start with a free pilot →

The deliverable

What a sized ICP looks like on delivery

You receive the documented ICP, a per-domain deduplicated CSV, and a sizing sheet. Below is an illustrative slice of the CSV — one row per canonical domain, with the fit verdict and per-field confidence. Values are illustrative; real domains ship in your deliverable.

canonical_domain	is_icp	fit_conf	independently_run	affiliate_links	fresh_content	own_inventory	language
publisher-a.example	true	0.94	0.91	yes	yes	no	en
publisher-b.example	true	0.88	0.79	yes	partial	no	en
publisher-c.example	false	0.72	0.40	no	yes	yes	de
publisher-d.example	true	0.81	0.85	partial	yes	no	fr

Plus a sizing sheet: category count, reviewed-slice fit rate, measured precision, and the extrapolated addressable publisher estimate — every projected figure labeled as an estimate with its calibration basis shown.

Questions

ICP definition & sizing — FAQ

What does "~4% error" actually mean here?

It is the share of domains the reviewed calibration slice marked as misclassified once the ICP was tuned. In the reference engagement the reviewed sample held at roughly 96% precision, so about 4% of the flagged publishers were judged not-fit by the client. It is a measured figure from your own slice, not a marketing promise.

How can you size the market before running the whole thing?

The category count is known from the 120M+ classified-domain database. The reviewed slice gives a measured fit rate. Multiplying the two yields the addressable estimate. Because the slice is random, the extrapolation is defensible — and we always label it as an estimate, with the basis shown.

How big is a typical addressable set?

For a single vertical it usually lands in the tens of thousands to hundreds of thousands of qualifying publishers; broad or multi-country scopes can reach into the millions of domains before screening. The exact number depends on how tight your ICP is — scope is a dial you control.

Who defines the signals — you or us?

Both. We bring a fixed vocabulary of publisher signals and pre-built templates per vertical; you supply the judgment about what "ideal" means for your program. The calibration loop is where the two meet — your fit / not-fit reviews are what sharpen the definitions.

What do I get out of the free pilot for this use case?

The full pipeline configured for your ICP, stopped at the first 20 qualified publishers — free — plus a documented signal set and a full-run yield projection. It is the fastest way to see both a real list and a real market size before committing to anything.