FinancialanalyticswithAI
We deployed an AI system that unified financial data from multiple sources, automated management reporting and accelerated CFO decision-making. Instead of stitching reports by hand, the team got a single analytical loop with dashboards, deviations and ready-made conclusions.
NovaCapital — a holding with 6 legal entities. The finance team spent a week on a monthly report. We plugged in an AI analyst: data assembles itself, conclusions write themselves, the CFO reads the brief and decides.
The problem
Reports were assembled by hand from 1C, Excel and bank exports. A week of work for 3 analysts.
Risk
A mistake in a manual reconciliation = a mistake in the CFO's decision. Too expensive.
Goal
A daily CFO brief in 10 minutes. A weekly one in half an hour. With no manual work.
Speed of financial decisions
- 3 analysts — a week per monthly report
- Decisions made on week-old data
- 7 fragmented data sources
- Errors in manual reconciliation
- A report assembles in 30 minutes
- CFO gets the brief by 9 a.m.
- 7 sources → one window
- The model catches anomalies, not people
Shipped in 5 weeks
From first call to first production launch. No phase two, no endless iterations.
Discovery
Mapped 7 sources. Built a unified schema and a quality baseline.
Design
CFO panel: one screen, one decision lane, drill-down to source.
Build
Built the analyst agent, alerts and the nightly data rebuild.
Launch
Shipped the daily CFO/board brief. Locked the target data-freshness SLA.
How we built it
Data layer
DuckDB on top of 1C, ERP, bank exports. One schema, daily rebuild.
Analyst agent
Reads the data, finds anomalies, builds forecasts, writes commentary in business language.
CFO panel
A dashboard with one lane of "what changed" and "what to do". No 40 charts.
Alerts
Plan deviations land in the CFO's messenger — with a proposed action, not just a number.
Daily brief
The CFO opens one page and knows: what changed, why, what to do.
Data layer and alerts
DuckDB on top of 1C, ERP, bank exports. One schema, a nightly rebuild, an SLA on data freshness. Alerts arrive with a proposed action, not just a number.
What came out
“The weekly report now takes half an hour. The board decides on fresh numbers, not week-old ones.”
Yuki Tanaka
Vitaly Kust
James Park
Mariia Golikova
AI-CRM for a dry-cleaning network
A CRM that calls clients, reminds, cross-sells and closes tickets on its own. Before — 3 operators. After — one operator plus an agent.
