From Rules to Reasoning: How Agentic AI Is Rewiring MDM in Pharma

In pharma, customer master data is the record of every physician, practice, and health system your commercial and R&D operations depend on. Master data management (MDM) has always been good at one thing: telling you something is wrong. It flags two records that might be the same physician. It queues a data change request from a rep in the field. It hands a steward a confidence score. What it has never done is resolve the problem. That last mile, along with the research, the judgment, and the decision, falls to a person at a desk, one HCP record at a time.

In a typical pharma MDM environment, over 10,000 potential matches and data change requests can sit in the queue at any given moment. Every new data source that gets onboarded pushes the number higher. Backlogs grow faster than headcount, and rules-based logic simply can't keep pace with the volume, variability, and speed of incoming change.

That was the starting point for the latest Axtria Ignite webinar, “From Rules to Reasoning,” where Harneet Kapil, Principal, Data & AI at Axtria, was joined by Somy Silvy, Senior Director, Data Strategy & Products at Johnson & Johnson Innovative Medicine, to look at how agentic AI is changing the economics of stewardship.

The shift: from flagging to reasoning

The core idea is simple. Rules tell you something's wrong. Reasoning agents work through it.

An agent doesn't replace your MDM platform or your stewards. It sits beside the steward, does the research, reasons through the evidence, and brings back a recommendation with sources attached. The human still makes the call. Every merge, every approval, every enrichment stays a human decision with the agent just making sure that decision is already researched and cited, and still defensible six months later when someone asks why.

The webinar demonstrated three agents already doing this work in live client environments.

Match Resolve Agent: For every potential match, the agent pulls profiles from MDM and verifies each one independently against the open-source National Provider Identifier (NPI) registry and the open web. It builds an attribute-by-attribute comparison; what MDM holds versus what authoritative sources say; marks each row match, conflict, or missing; and commits to a recommendation: merge, not a match, or send to review. Each carries a confidence score, a plain-language rationale, and citations the steward can click straight through. The result: 2–3× steward throughput, 50–60% lower cost per record, and 70–80% faster resolution.

(Note: NPI Registry is a skill and guardrail for demo purposes. This can be configured as per individual client needs.)

Smart DCR Agent: Data change requests (DCR) in customer MDM operations keep coming, and manual review means field reps wait weeks for a response. This agent validates the queue in bulk against the National Plan and Provider Enumeration System (NPPES), the open web, and the client's own approved source and value lists, then closes the loop back to both MDM and customer relationship management (CRM) systems. What took ten minutes of manual research per DCR now takes about 35 seconds. One client went from a single steward clearing 50–100 DCRs a day as a rubber-stamp exercise to genuine validation at roughly 60 DCRs a minute, with no added headcount when new sources came online.

(Note: NPPES reference is a skill and guardrail for demo purposes. This can be configured as per individual client needs.)

Enrich HCP Agent: A golden record with gaps isn't golden. This agent finds what's missing, such as a blank NPI, an outdated address, or a sparse specialty, and fills it from named sources using confidence tiers: ≥90% auto-enriches, 60–89% routes to steward review, under 60% goes to manual research. Every field carries its source and a live link. Nothing is a black box.

The three agents share a remit and can operate as a multi-agent ecosystem, coordinating with each other much as human stewards do.

Why it matters beyond throughput

The speed numbers are compelling, but the deeper value showed up in the discussion. Consistent standard operating procedure (SOP) adherence replaces the wide variability of manual review under pressure. Full observability and audit trails make every decision explainable. Stewards aren't cut; they are redeployed to higher-value work. And faster, evidenced responses to field-submitted DCRs rebuild something harder to measure: business trust. When reps see their submissions acted on and rationalized, they engage more, and better engagement improves the data itself.

The practical path

Adoption does not require a re-platform. Each agent layers a reasoning capability onto the MDM platform already in place. Axtria recommends a deliberately phased approach: begin with the queue creating the most friction, validate the throughput gains against live production data, and expand as confidence builds.

The takeaway from the session was clear. Stewardship has seen incremental gains for years with a little more automation here, a faster data pull there, but the research itself stayed manual. Agentic and generative AI is the slingshot that finally moves it.

The next step is a focused pilot: one agent, one live queue, a number you can measure in weeks.

Learn more: Watch the webinar here, or reach out to Axtria’s Data & AI team today.

FAQs