HCP Audience Generation: Lookalikes, Modeled Lists, and When to Avoid Them

Modeled audiences come up in almost every HCP media conversation, usually when a campaign is underdelivering. The partner offers to "open up" the audience with lookalikes. The line starts spending. Reach looks better. Months later, someone asks why so few of the impressions went to Tier 1 prescribers.

What is HCP audience generation?

There are three common types:

• Lookalike expansion. The partner takes your NPI list as a seed, finds the shared traits (specialty, region, content behavior), and adds providers or devices with similar traits.
• Predicted HCP segments. The partner models which devices likely belong to healthcare professionals in general, or to a specialty, without reference to a specific list.
• Behavioral or contextual segments. Devices that read clinical content about a disease area are grouped into an "interested HCP" segment.

All three differ from list-based targeting in one important way: the people reached are, by definition, not on your target list, so you cannot report reach against your list for them. If you are new to these distinctions, the HCP targeting explainer covers the basic methods.

When do modeled HCP audiences help?

• Before launch. Prescribing data for the new category may not exist yet, so the target list is a guess. Modeled audiences around disease-area content can find relevant providers the list missed.
• Very small lists that will not spend. A small, adjacent audience can keep frequency on the core list from climbing too high while still putting impressions near relevant clinicians.
• Exploring a new segment. If you suspect a specialty you have not targeted is writing in your category, a modeled test can show whether engagement is there before you invest in a list.
• Awareness among the broader care team. Nurses, pharmacists, and office staff may not be on an NPI list but can matter for access.

When should you avoid them?

• Your list is well matched and the campaign is spending. Expansion only dilutes.
• The specialty is narrow and the brand's opportunity is concentrated in a few hundred or few thousand providers.
• The partner cannot separate modeled delivery from list delivery in reporting.
• Expansion is being proposed to fix underdelivery without anyone diagnosing why the list underdelivers. See what to do when a campaign underdelivers first.
• The model's inputs include data your contracts or privacy commitments do not allow for this use.

What does drift look like?

Drift is the gradual movement of spend away from the providers you care about toward the ones that are cheapest to reach. Optimization algorithms reward clicks and low CPMs. Providers who browse a lot, on cheap inventory, at all hours, will win more impressions. Those are rarely the busy high-volume specialists at the top of your list.

A hypothetical example: a campaign starts with 100% of impressions on list NPIs. Expansion is turned on in month two. By month four, the illustrative split might look like this.

If Tier 1 keeps 40% of list impressions in each month, the Tier 1 share of all impressions is 40% times the list share: 30% in month two, 22% in month three, and 18% in month four. Total reach probably went up the whole time. The audience that matters most got less than half the attention it started with. These numbers are illustrative.

How to validate a modeled HCP audience

1. Ask what the model was trained on. The seed list, the features, and the source data. If training used your list, the model learns your list's biases too.
2. Use a truth set. Endemic logins, verified email lists, or a known set of NPIs can be compared against the modeled segment. What share of modeled devices that also appear in the truth set are actually HCPs in the right specialty? The site's piece on audience data accuracy validation covers how to design this.
3. Run a split test. Run list-only and list-plus-modeled cells with a holdout and compare outcomes per dollar, not reach.
4. Check time and placement patterns. HCP traffic tends to cluster in working hours and on clinical or professional content. A modeled segment that delivers mostly at night on gaming apps is not an HCP audience.
5. Review the model's audience size over time. If it grows quickly, the model may be loosening its criteria to meet spend.

How to cap and report modeled expansion

• Set a maximum share of impressions or budget for modeled audiences in the insertion order.
• Require separate line items or separate reporting rows for list and modeled delivery.
• Never include modeled reach in "target list reach." Report it as its own number.
• Review the cap monthly against Tier 1 reach. If Tier 1 reach drops, reduce the cap.

If the root problem is a list that is too small for the budget, revisit list design with how to build an NPI target list before reaching for expansion. And if a partner pitches fully synthetic HCP audiences, read synthetic audiences in healthcare media first.

Practical takeaway

Write a rule into the next IO: modeled or lookalike HCP audiences may take no more than a set share of impressions (pick a number with your team, such as 20%), must be reported on a separate line, and must be switched off if Tier 1 reach declines two months in a row.

Editorial note. Analysis and frameworks are the author's own and do not represent Acxiom or any current or former employer, client, or named platform. Examples labeled hypothetical or illustrative are not results from real campaigns. Nothing here is legal, regulatory, or medical advice.