ACCENTURE · 2025Python · Streamlit · Claude Code

The analyst's
co-pilot.

Demand-planning analysts were spending 5–6 hours every single day manually pulling, matching, and reconciling data across five sources and three platforms. I built an internal app — in 3 days — that does it for them.

200+
man-hours reclaimed every month
01 — THE SITUATION

Five sources. Three platforms.
Five to six hours. Every day.

The demand-planning team at Accenture was the last mile between raw data and decisions — but they were spending the bulk of their day not planning, but plumbing. Every morning meant opening five different data sources, manually cross-referencing them across three separate platforms, reconciling mismatches by hand, and then assembling outputs for reports and system uploads.

Five to six hours. Daily. For a task that was entirely deterministic — the same steps, the same logic, the same pain — just with different data each time. There was no single source of truth. No automation. No tooling. Just analysts, spreadsheets, and a growing backlog of ad-hoc requests on top.

5
data sources to pull from manually
3
platforms to match and reconcile across
5–6 hrs
lost to reconciliation every single day
02 — WHAT I BUILT

A Streamlit app that
does the plumbing.

I built the app over three days using Python, Streamlit, and Claude Code — starting from a blank repo and iterating fast with Claude Code handling the boilerplate, data-wrangling logic, and UI scaffolding while I focused on the business rules and validation.

The app ingests from all five data sources, runs the matching and reconciliation logic automatically, and produces three types of output that previously required hours of manual assembly:

Upload-ready files
Formatted exactly for the target systems — no manual reformatting before upload.
📊
Daily · Weekly · Monthly reports
Automated report generation on schedule — analysts review, not rebuild.
🔍
Ad-hoc analysis
Analysts can run custom queries and slices on demand — without touching raw data.

The interface is intentionally minimal — a Streamlit dashboard with clearly labelled actions. The goal was zero learning curve: anyone on the team could pick it up without a walkthrough.

PythonStreamlitClaude CodePandas
03 — THE IMPACT

200+ hours back.
Team adopted it without being asked.

The clearest signal that it worked wasn't a metric — it was behaviour. The team started using the app without being told to. No rollout deck, no change-management session. They just switched.

At 5–6 hours of daily reconciliation eliminated across the team, the app reclaims over 200 man-hours every month — time that now goes into actual analysis, exception management, and planning decisions.

The product is still live and evolving. New data sources, new report types, and new ad-hoc modules get added as the team's needs grow. Three days to ship v1; the roadmap is open-ended.

"The clearest proof: the team adopted it without being asked. That's the only product metric that really matters."