Case Study Banner

CASE STUDY

Governed AI powered self-service analytics for portfolio and R&D decision making

industry-iconCLIENT :Confidential
industry-iconINDUSTRY :Pharmaceutical
industry-iconDURATION :6 months
CLIENT :
Confidential
INDUSTRY :
Pharmaceutical
DURATION :
6 months

Business case

Business teams across Portfolio, Data & Analytics rely on data from multiple systems, reports, and dashboards to make critical portfolio, project, and investment decisions. Core decision inputs, valuation metrics, risk adjustments, R&D metrics, portfolio and project metadata, sat across fragmented sources, each with its own structure, refresh cycle, and, in many cases, its own definition of the same metric.

This fragmentation made timely, trusted insight difficult to get. Business users routinely spent time validating numbers, reconciling discrepancies between reports, and building ad hoc analyses rather than acting on decisions. Any question that fell outside an existing dashboard meant a request to a technical team and a wait for a response, slowing down the pace of portfolio and investment decision making.

As data volume and complexity continued to grow, this reporting model became harder to scale. Business users needed the ability to ask questions in natural language, explore data independently, and get answers backed by consistent, governed business definitions, not just another dashboard, and not an AI tool that traded accuracy and transparency for speed. Existing AI solutions offered promise but raised real concerns about data accuracy, explainability, and trust, which meant any solution had to earn user confidence, not just add automation.

Our Solution

The Portfolio Dev Team built a governed, AI enabled self-service analytics platform, using Dataiku for the underlying data engineering pipeline and Snowflake Cortex to let business users query trusted data in natural language and get consistent, explainable answers back.

The team's approach included:

  • Building the data engineering pipeline in Dataiku to clean, standardize, and validate source data across valuation, risk, R&D, and portfolio/project metadata before it reached the governed layer, addressing a significant part of the model accuracy issues at the data layer, rather than relying on the AI layer alone to compensate for inconsistent inputs.
  • Consolidating Portfolio team’s core decision support data, valuation metrics, risk adjustments, R&D metrics, and portfolio and project metadata, into Snowflake as a single, governed analytical environment.
  • Using Snowflake Cortex to translate natural language business questions into data queries, so users could ask questions directly instead of navigating multiple systems or waiting on technical support.
  • Codifying business definitions and calculation logic for each metric so that answers stayed consistent and explainable regardless of which system the underlying data originated from.
  • Enabling on demand visualizations and interactive dashboards generated directly from natural language questions, reducing reliance on static, prebuilt reports.

Challenges overcome

  • Reconciling inconsistent metric definitions and calculation logic across valuation, risk, R&D, and portfolio/project data before they could be surfaced through a single governed layer.
  • Managing hallucination and accuracy risks in AI generated answers, so that responses were grounded in governed business definitions rather than model inference.
  • Building business user confidence in AI generated answers after prior experience with tools that struggled with data accuracy and transparency.

Benefits

  • Reduced dependency on static dashboards, manual reporting, and technical teams for day-to-day business questions.
  • Faster, more confident portfolio and investment decisions, supported by consistent and explainable answers.
  • Greater trust in AI assisted analytics, driven by governed business definitions rather than ungoverned model output.
  • Business users across Portfolio team can now explore data and generate visualizations independently, in natural language, without specialized technical skills.

Results

Results
SPM Icon
SPM maturity calculator