Session

Closing the Insight-To-Action Gap With Databricks Genie and Apps

Overview

Experience	In Person
Track	Analytics & BI
Industry	Enterprise Technology, Consulting & Services, Financial Services
Technologies	Genie
Skill Level	Intermediate
DOWNLOAD SESSION SLIDES

Finance teams often identify insights but lack streamlined ways to operationalize them. In this demo-driven session, KPMG demonstrates how Engagement Management Coordinators (EMCs) within the Finance organization close the insight-to-action gap using Databricks Genie and Databricks Apps. EMCs use conversational analytics to evaluate engagement performance and financial risk across scenarios, while Databricks Apps embed those insights directly into workflows such as forecasting revisions and exception handling. Genie provides a consistent business-facing entry point supported by Unity Catalog governance. The session shares adoption strategies, governance considerations, and lessons learned from enabling finance teams to move from analysis to action through guided demonstrations.

Session Speakers

GIdget Santos

/Senior Director
KPMG LLP

Mohana Gudepu

/Director, Strategy and Innovation
KPMG

Full Summary

How KPMG closed the gap between finance insight and action

Many finance teams are not short on data or dashboards, they are short on decisions.

KPMG's Client Zero build, showcased at Data + AI Summit, tackles the gap between analysis and execution by unifying discovery, conversational analytics, and decision capture on the Databricks Platform. The architecture is rooted in the medallion pattern for data organization, which separates raw, conformed, and curated layers to give finance users a trustworthy, governed foundation for every query and decision.

The result is a governed, three-surface experience that treats workflow, semantics, and governance as a single product. Unity Catalog underpins every interaction, so the same permission model, lineage, and audit apply from the first question to the final approval.

Why insight-to-action often breaks in finance

Organizations generate abundant analysis, yet decisions stall for predictable reasons: context is lost across siloed tools at handoff, self-service assumes SQL fluency that many business users lack, and fragmented governance erodes trust at the moment of choice. More dashboards do not produce more decisions when the workflow is broken.

KPMG reframed the problem as a product design challenge. Instead of layering on more BI, the team created three connected surfaces that share one identity and one Unity Catalog, so context, trust, and action live in the same place.

A three-surface design on one governance plane

The architecture follows the medallion pattern, landing sources in bronze, conforming in silver, and curating in gold. Unity Catalog spans the lake and every surface, enforcing lineage, role-based access, row and column level security, and a single audit. Finance users see only curated gold, and no surface bypasses Unity Catalog.

Enter: a unified entry surface with one search across workspaces, apps, dashboards, and certified queries. Developer controls like Databricks Notebooks, Spark UI, and cluster configuration are intentionally absent.
Ask: a conversational layer that sits on pre-joined, certified gold tables augmented with a plain-English business glossary.
Act: apps built for recording decisions, each inheriting user permissions and writing every action to an auditable table from day one.

From conversational analysis to recorded decisions

Conversational analytics centers on trust. Genie is scoped strictly to curated gold tables, so finance questions map to finance-grade data and definitions. Stewards in finance pin certified queries for high-traffic questions, and user feedback flows back to those stewards. Trust comes from curation and stewardship, not from the model alone.

Decision capture lives in Streamlit apps. On‑behalf‑of‑user authentication ensures each session inherits the user's permissions. Submissions include reason codes and justifications, then flow to approvers. Every proposal, approval, and denial is recorded in an audit table that is queryable immediately, producing an end‑to‑end trail without leaving Databricks.

A real workflow and measurable outcomes

The build targets KPMG's engagement management controller process for ERP adjustments and engagement closeouts, performed roughly 75,000 times per year by a team of 170 controllers. The prior approach relied on SAP and HANA extracts, a homegrown VBA tool, spreadsheets, and email approvals, which made reconciliation slow and audit trails incomplete.

In the new flow, a controller tests ERP rate scenarios in natural language, reviews revenue and margin impacts, then submits an adjustment in the app with a reason code. The engagement partner reviews a dedicated dashboard and approves or denies the request, with every step recorded. The firm expects to return roughly 25,000 hours of capacity annually and collapse five systems into a single governed flow.

Rollout principles that built trust

Four choices shaped adoption: curate the gold layer before exposing Genie, pilot with a small group and iterate weekly, embed apps directly in the workflow, and make the unified entry point the default home. The broader lesson is that adoption is a trust problem, not a model problem. Certified, stewarded queries outperform model tweaks, and conversational analytics is only as good as the semantics beneath it. The team's two rules: apps are for recording decisions, not for views, and success is measured by actions submitted, not logins.

Design for decisions, not dashboards

The gap between insight and action is a workflow problem. Treating governance, semantics, and decision capture as one continuous surface let KPMG move from analytics that inform to systems that execute, with a clean audit trail and capacity returned to the business.

FAQ

Unity Catalog is the single governance plane under the entry surface, Genie, and the apps. It provides one permission model, one audit log for queries and app actions, lineage across medallion layers, federated identity, attribute-based access tags for line-of-business scoping, and network controls via Private Link.

Genie queries only curated gold tables so answers draw from certified, pre-joined data connected to a business glossary. That constraint makes responses trustworthy for finance users. Exposing bronze or silver would surface raw or partially conformed data that lacks shared semantic definitions.

Streamlit apps use on‑behalf‑of‑user authentication, so each session inherits the user's permissions. Every submission, reason code, justification, approval, and denial writes to an audit table that is queryable immediately, creating an end‑to‑end record from proposal to decision.

The team built the gold layer first, piloted with 12 controllers, gathered weekly feedback, and only then rolled out apps and the unified entry point as the default home. Adoption is measured by actions submitted in the apps, not by logins, aligning success to real decisions made.

For ERP adjustments and engagement closeouts, the firm expects to return about 25,000 hours of capacity per year, consolidate five systems into one governed flow, and produce a complete audit trail for every transaction for internal review and external audit.