Aberrancy Detection & Lead Generation

What You Do Today

Leads reach you from everywhere and in no particular order. A rules-based outlier report lands on a schedule and names providers whose billing sits outside their peer group. A claims processor sends over something that looked wrong. A member calls about an explanation of benefits listing a visit that never happened. A hotline tip arrives with no detail and no name attached. A state fraud bureau or another plan flags a provider you also pay. And pre-payment review sends over what an edit could not resolve, because a claim edit is not an investigation. You sort what is worth an investigator's time, and the sorting is the job. You look at billing patterns over time rather than at single claims: codes that shifted upward as a group, units that do not fit the service, modifiers appearing on nearly everything, a place of service that does not match the work, a rendering provider whose day would not hold the hours billed under it, members who turn up at unrelated providers in the same sequence. The hard part is that being unusual is not being wrong. The provider at the top of the report might have the sickest panel in the county, or be the only one in it doing the procedure at all, or have a coding habit that is incorrect and entirely sincere. And you have more leads than investigators, so the ranking you produce decides what gets looked at this year and what does not.

AI Technologies

How It Works

Peer-group construction builds the comparison from what providers actually bill rather than from the specialty code on their file, which matters because that code is often stale or was never precise, and a comparison group assembled wrong produces outliers that are artifacts of the grouping. Outlier scoring runs each measure separately — coding distribution, units per service, modifier frequency, place of service, services per member over a period — and reports which measure drove the score, so a lead arrives with a reason attached rather than as a rank. Unsupervised detection reads claim and encounter history in time order and surfaces patterns nobody wrote a rule for: a distribution that shifts after a practice changes ownership, a code that appears across a whole panel within a short window, billing that starts the week an authorization requirement was relaxed. Link analysis builds a graph of providers, members, tax identification numbers, addresses, billing agents and referral relationships, and finds structure a per-provider report cannot see — a set of members who appear together at providers with no clinical relationship, several practices resolving to one address, a referral pattern that only runs one direction. Scoring against the unit's own dispositions shows which lead sources have historically produced substantiated cases and which have produced work.