Marko Tomic

Flagship 03 · Decision infrastructure

Commercial Intelligence

Two locally built applications for the economics of the business - because the logic is the asset, and renting it from SaaS buys neither flexibility nor understanding.

Applications2 · both local Scale1,153 SKUs DependencyNo external SaaS

Measured

2.1% Full-year forecast variance

Forecast versus actual generated net revenue. Quarterly forecasts landed ~6% below actual.

74 s ~1,000 real PDF invoices parsed

Benchmark of the invoice-ingestion pipeline.

1,153 SKUs under margin intelligence

Cost, VAT, shipping, price and advertising context per SKU.

Context

The economics lived in spreadsheets

Pricing decisions ran on scattered information: supplier invoices in PDFs, advertising performance inside disconnected Google platforms, margin math in spreadsheets that drifted out of date. Revenue planning was intuition dressed up as a forecast.

The business didn't lack data. It lacked one place where cost, price, margin and cash expectations meet.

Decision

Build local, own the logic

The obvious answer was a SaaS BI stack. The chosen answer was two locally deployed internal applications. The data is sensitive, the logic is the competitive asset, and the scale is modest - ongoing rent for generic tooling buys neither flexibility nor understanding. Complexity should be earned; here, a local application is the simpler system that fully fits the problem.

System

Not dashboards - decision infrastructure

A dashboard shows you a number. These systems hold the logic that produces the number - inspectable, testable, and validated against what actually happened.

  • Input
  • Processing
  • Application
  • Scenario gate
  • Human judgment
Margin & Pricing answers "what should this cost the customer?"; Revenue & Cash-Flow answers "what should we expect, and when?" Both stop at scenarios - the decision stays human.

Margin & Pricing Intelligence

Parses supplier PDF invoices at scale, computes current margin per SKU, recommends prices, and models pricing and margin scenarios with advertising-spend context. Mathematical optimization runs operationally. Inspect the system →

Revenue & Cash-Flow Intelligence

Mathematical forecasting for revenue planning, seasonal expectations, cash-flow thinking and scenario analysis - validated against actual subsequent net revenue. A natural-language querying layer exists as an MVP. Inspect the system →

Ownership

Who owns what

I owned

  • Both applications designed & built end to end
  • Data ingestion
  • Modeling
  • Application & deployment

AI / system executed

  • Mathematical optimization runs operationally
  • Invoice parsing & normalization

Outcome

What changed, measured

Measured: the forecasting system's full-year prediction landed approximately 2.1% below actual generated net revenue; quarterly forecasts landed approximately 6% below actual. The pricing system has informed real pricing decisions and contributed to improved margins.

Not claimed: no invented margin percentage. The pricing claim is stated without a number because an honest one doesn't exist.

Reflection

Why this architecture holds

The value was never the interface; it was forcing the economics of 1,153 SKUs into explicit, inspectable logic. A forecast that is 2.1% off for a full year is useful precisely because you can see how it reasons. The natural-language layer stays an MVP until it earns more - the same discipline described in Complexity should be earned.

Related