AI Journey at Rallyday - UC #7: Financial Reporting Dashboard

Part of our ongoing series to document our data / ai with our friends at Nimble Gravity.

Overview & Business Value

An automated portfolio monitoring system that aggregates KPI data from portfolio companies into a unified analytical view, updated each time a portco submits their data. Databricks Genie provides the natural language query layer for the portfolio team. Portcos submit data via a monitored email inbox — no file drop folders, no SharePoint coordination. Output is a Databricks-powered dashboard and automated email digest; the architecture is not dependent on any single reporting tool.

Platforms & Integrations

Technical Approach

The data submission workflow is email-first. Portfolio companies send their monthly or quarterly data to a designated inbox (e.g., reports@rallyday.com). Power Automate watches the inbox, detects new emails with attachments matching expected formats, extracts the attachment, and drops it into a Databricks ingestion queue. This matches how portcos already communicate — email — and eliminates the SharePoint folder coordination burden.

The Databricks pipeline validates incoming files against the KPI taxonomy, normalizes for period consistency (LTM, QoQ, YoY), and runs variance calculations against prior periods and budget targets. Historical data for each portco is loaded during an onboarding sprint. Incomplete or malformed submissions trigger an automated bounce-back to the portco with specific guidance on what is missing.

The analytics layer is Databricks Genie, not a custom-built Claude agent. Genie handles natural language queries over structured portfolio data cleanly and without the overhead of a custom AI build. The portfolio team can query the full portfolio dataset in plain language. Formatted reports and executive summaries are generated as scheduled Databricks exports — delivered as email digests or accessible via a SharePoint-hosted report file. The architecture is not dependent on Zeck; if Zeck is adopted it can be added as an output layer later without rebuilding the pipeline.

Delivery & Integration

Agile Breakdown

Epic 1: KPI Taxonomy & Data Standards

• Workshop with portfolio team: define standard KPI set (financial + operational, by portco type)

• Design portco data submission template (Excel format, column-standardized)

• Define variance thresholds and alerting rules (what constitutes a flag)

• Portco onboarding plan: communication template and timeline for getting portcos onto standard format

Epic 2: Email Ingestion Workflow

• Set up reports@rallyday.com monitored inbox

• Build Power Automate flow: new email with attachment → extract → route to Databricks queue

• Build data validation: flag incomplete or malformed submissions; auto-notify portco with gap details

• Build submission confirmation: auto-reply to portco confirming receipt and validation status

Epic 3: Databricks Pipeline

• Build ingestion and normalization pipeline (attachment → KPI taxonomy)

• Build variance calculation engine (QoQ, YoY, vs. budget)

• Build historical data loader for portco onboarding (12+ months of historical data)

• Build pipeline monitoring and alerting (failed runs, malformed files)

Epic 4: Genie & Report Output

• Configure Databricks Genie over the unified portfolio dataset

• Define 10-15 example portfolio queries to seed Genie's understanding

• Build scheduled executive summary report (Databricks export → email digest)

• Build SharePoint report archive (optional formatted output)

Epic 5: Portco Onboarding & Go-Live

• Load historical data for each active portco

• Onboard 3+ portcos to standard submission template and inbox workflow

• UAT with portfolio team including a simulated full reporting cycle

• Go-live communication to all portfolio companies

Dependencies & Prerequisites

• KPI taxonomy agreement — cannot be driven by NG; requires Rallyday portfolio team ownership before Sprint 1

• At least 3 portfolio companies willing to adopt standardized submission template

• Portco historical financial data available for initial load (12+ months preferred)

• Power Automate or equivalent automation platform available in Rallyday's Microsoft environment

• Databricks environment available

Definition of Done

• All active portfolio companies submitting data to inbox on agreed schedule

• Dashboard data updated within 4 hours of each portco email submission

• Genie answers portfolio team natural language queries accurately on the unified dataset

• Malformed or incomplete submissions bounce back to portco with specific correction guidance

• Portfolio team has completed one full reporting cycle using the system

Risks & Assumptions

• Portco compliance is the primary execution risk — if portcos submit inconsistently or in non-standard formats, the pipeline breaks; the validation and auto-notification workflow is a key safeguard but human follow-up may still be needed

• KPI definition disagreements between portcos: some metrics (ARR vs. recognized revenue, Adj. EBITDA definitions) need explicit standardization in the taxonomy sprint

• Historical data quality: portco historical data may be inconsistent — plan one sprint of data cleanup per portco during onboarding