How do I build an ICP from public healthcare data?

Build an ICP from public healthcare data by finding the attributes your best customers share, then writing them as thresholds you can apply to the whole market: specialty or facility type and location from NPPES; size, ownership, and care setting from CMS; service lines and financing from the company website; review volume and rating from Google, RealSelf, and Healthgrades; social footprint from Instagram and LinkedIn; growth from job postings; and firmographics from third-party data, all validated against each other. A strong ICP is the foundation of every list, score, and territory that follows.

This guide covers defining the profile. To find accounts that match it, see How do I use NPI data?; to score them, see How do I build a Fit Score?

Where do I start?

Start with your won deals, not your guesses. Your own won-deal data is the ground truth the whole profile derives from: list your best 20 to 50 customers, the ones that closed fast, expanded, and stayed, then find what they share in data you can verify from the outside, so the profile applies to prospects you have never met. No single source describes them fully, so you read several and let them corroborate each other. Provider specialty and location come from NPPES taxonomy codes; size, ownership, and care setting come from CMS files; service lines and financing come from the company website; review volume and rating come from Google, RealSelf, and Healthgrades; social footprint comes from Instagram and LinkedIn; growth comes from job postings; and third-party firmographics fill the rest.

How do I turn the profile into criteria?

Write each attribute as a rule with a threshold, not a vibe. "Multi-location groups with 5 or more providers in these NPPES taxonomies, independent of a health system, with 200-plus Google reviews above 4.5 stars and an active Instagram, in these states" is a criterion a machine can apply to your whole market, and it is exactly what becomes a Fit Score. The structural attributes come from registries, the consumer attributes from web, reviews, and social, and each validates the others.

Illustrative example

Illustrative, based on how SignalRx customers work. A revenue-cycle vendor finds its best customers are independent gastroenterology and orthopedics groups with 4 to 15 providers, not PE-owned, in five states, with strong Google review volume and a multi-location website. That becomes the ICP: those two NPPES taxonomies, that provider-count band, independent ownership (no MSO or PE platform above them), a website showing multiple locations, healthy review counts, and those states. SignalRx applies it across the combined public, web, and review universe and flags every matching account.

Because prospects have no history in your systems yet, SignalRx scores this ICP from external data triangulated across sources, NPPES specialty and location, CMS size and ownership, the company website for service lines and financing, reviews and social for consumer traits, and third-party firmographics, writing a Fit Score and tags onto both the account and its contacts as Revenue Plays and lists native in Salesforce. Where a CRM record is too thin to score, it fills the gaps by triangulating these sources so the account is qualified rather than dropped, and inference decides which to trust when they disagree. Once a matching prospect becomes a customer, SignalRx layers in your own internal signals, revenue, support, and product usage tracked over time, the longitudinal view competitors cannot replicate.

Key takeaways

• Build the ICP from your best won customers, the ground truth the profile derives from.
• Write each attribute as a threshold rule so it can drive scoring.
• Registries are one input, strong for structure like specialty, location, size, and ownership but thin on consumer traits; web, reviews, and social supply those, and the ICP is multi-source attributes validated against each other.

See it in action: Discover High-Fit Prospects with SignalRx.

Related: How do I build a target account list? and How do I find lookalike accounts?