At Novo Nordisk one analyst's expertise turned into an AI agent the whole team can rely on to create and deliver faster insights and performance reviews

Novo Nordisk, the Danish healthcare company behind many of the world's leading treatments for diabetes, obesity, and rare disease, delivers medicines to millions of patients globally. Its Rare Disease Insights & Analytics team built an AI assistant that answers business questions from enterprise healthcare data in plain language — turning manual data work into a first-pass answer in under 30 seconds, and reclaiming roughly half an analyst's worth of capacity without adding headcount.

The demand kept growing. The team didn't.

Novo Nordisk's Rare Disease Insights & Analytics team delivers insight for drug launches, market monitoring, KPI reviews, and brand strategy. As the rare disease portfolio expanded and headcount tightened, the gap between the questions stakeholders asked and the capacity to answer them kept widening.

The bottleneck was rarely the thinking. It was everything before it. Analysts spent significant time  generating reports, before they could even interpret anything. Deeper questions took days. The risk came as longer data manipulation times and time consuming reporting which lead to  growing  need for external support at exactly the moment the business needed to move faster.

The person closest to the problem built the answer

A Senior Analyst on the team combined deep knowledge of rare disease market dynamics, healthcare claims and prescription data, and the team's daily workflow to build an AI-powered assistant directly in Dataiku, designed around the team's actual decision cycle — with executive sponsorship from Rare Disease Insights & Analytics leadership and advisory input from a Dataiku consultant.

The result is an agentic analytics assistant: a text-to-SQL agent that lets analysts ask questions in plain language and get answers from enterprise healthcare claims and prescription data, including the Komodo claims database. Underneath, Dataiku connects the full workflow, including ingestion, preparation, GPT-4 integration, and Agent Hub with Visual Agents. It runs against a data layer that Dataiku Scenarios (automated, scheduled pipelines) refresh every day, so the agent always queries current claims and prescription data rather than stale, static dashboards.

From data pullers to strategic partners

The primary users are roughly three analysts across the US and India who once spent much of their time as manual data pullers and dashboard operators. Now the agent handles the low-value technical steps: querying, exploration, first-pass analysis. Meanwhile, the analysts spend their time on interpretation, hypothesis generation, and the strategic storytelling stakeholders actually need. As the team describes it, it's like having an extra analyst on hand, one that takes the time consuming tasks out of our hands so that we can focus on what sharpens the discussion with brand leads on stronger, data-backed footing.

Governance kept close to the expert

The team designed the assistant around one core principle: human-in-the-loop (HITL) validation. Generative outputs serve as  decision support, not final decisions.Before insights are shared with the commercial team, an analyst reviews, digests, and further refines them to ensure they are accurate, contextualized, and actionable. This helps guard against the risk that a language model produces a plausible-sounding response that misreads the data. A focused rollout and reused IT resources kept cost and access under control. 

Capacity reclaimed without added headcount

For Rare Endocrine, the  team estimates a roughly 50% gain in overall analytical efficiency, with reductions as high as 50–70% for exploratory analysis and routine reviews. Questions that once took hours now return a first-pass answer in under a few minutes,  while final interpretation stays with the analyst. That reclaims about half an analyst's worth of high-value capacity, an assistant analyst that offsets part of the resource contraction without new headcount.

Most of all, Dataiku turned one expert's domain knowledge into an operational, governed workflow. Without it, the team would have needed a custom application, separate ETL pipelines, direct LLM integration, and specialized engineering support. Instead, a single practitioner built a repeatable assistant inside the environment the team already used.

What opens up now

This  framework can now be used as the foundation across other  therapeutic areas in Rare Disease.