StakeInsights Consulting studio · deep tech

Case 01 · Oil & gas

From last year's audit to a live decision layer

A listed lubricants distributor was steering by audited statements and a hand-built monthly MIS. We built the decision layer its finance, sales and leadership teams now run on — refreshed from the ERP through the day, with every figure traceable to the query that produced it.

Client
Listed lubricants distributor, India
Desks
CFO · Sales · Boardroom
Status
Live in production
Name
Withheld · reference under NDA

The situation

The numbers existed. Nobody could see them in time.

Leadership read the business from last year's audited statements and a management report assembled by hand each month. By the time it was read, the quarter had moved on.

  • Finance and sales each held their own version of revenue, and argued about whose was right.
  • Collections, payables and inventory lived in separate ERP reports that nobody reconciled against each other.
  • Statutory obligations — paying small suppliers within 45 days — were tracked, if at all, in spreadsheets.

What we built

One reconciled set of numbers, and agents that act on it.

  • Automatic ingest. The ERP's report files are picked up from the client's cloud storage through the day, checked for change, archived, and staged. Most report types load straight through; purchase and vendor data still wait for a person to sign off.
  • A decision layer across the management P&L, balance sheet, collections, payables, inventory and listed peers. Every figure opens to its definition, formula and SQL.
  • Agents with owners. Alerts for small-supplier dues past 45 days and overdue payables go to the payables team, not a shared inbox. A performance report emails itself to every salesperson and regional manager each morning.
  • A board-grade stakeholder report and a short strategic read, both written from the same numbers finance signs.
  • Enterprise plumbing. A private database with no public endpoint, encrypted, with a standby in a second data centre and credentials that rotate automatically.
Fig. 1 — What runs every day: ERP files arrive, are checked and computed, and go out as screens, owned alerts and scheduled reports.

What was hard

The problems weren't the dashboards.

01

Two definitions of revenue. The management P&L used net revenue; an internal analysis used a different basis for its own purposes. We made finance's definition the only one any AI-written read is allowed to see, and kept the internal analysis clearly labelled as internal.

02

Files that arrive late, or twice. ERP drops get re-sent through the day. Every load had to be safe to repeat — re-running it can't double-count — and any load can be reverted if a file turns out to be wrong.

03

An AI read that must never invent a number. The written read may only quote figures already shown on the page. If a number isn't on screen, it can't appear in the narrative.

What changed

Decisions moved closer to the facts.

  • Leadership reviews the quarter from live, reconciled numbers instead of last year's audit.
  • A breach of the small-supplier payment limit reaches its owner the day it happens, not at the next monthly review.
  • Every salesperson starts the day with their own numbers, without anyone building a report.

Client name withheld at the client's request. A reference call can be arranged under NDA.

Start with one decision

Have a decision that looks like this one?

Tell us what you're trying to steer. We'll say plainly whether it needs a build like this — or something smaller.

Tell us the decision you're trying to make.

We'll tell you what it takes — and whether you need a custom build at all.