Pricing & Optimization Systems
Turning historical business data into pricing and inventory decisions with measurable financial impact.
Selected professional work at Thrasio, an e-commerce business with a large product portfolio. This was work undertaken in my employee role, not a Kogan AI Consulting client engagement.
Results at a glance
| 15% | 20% | 10,000+ |
|---|---|---|
| Reported profitability improvement | Reported reduction in excess inventory | Products in the elasticity modeling scope |
The first two outcomes relate to an inventory-aware pricing initiative. The product count relates to a separate DML elasticity workstream. Figures reflect previously reported results, pending final public-attribution review.
The challenge
Pricing decisions depend on more than current demand. Higher prices can improve margin but slow sales; lower prices can clear inventory while sacrificing profit. Stock levels, purchasing commitments, marketing spend and cash flow change which decision is best.
The business needed to move beyond forecasts or isolated pricing rules toward a system that could evaluate competing actions against financial and operational objectives.
My contribution
At Thrasio, I led and implemented work connecting demand forecasting, price-response estimation, simulation and optimization.
The approach combined:
- Demand forecasting to estimate baseline sales patterns.
- Price-elasticity modeling, including double machine learning (DML), to evaluate likely responses to price changes.
- Inventory-aware optimization using an NPV-based objective that considered margin, sell-through, cash recovery and capital tied up in stock.
- Scenario testing and business guardrails so recommendations could be validated and used responsibly.
The central idea was straightforward: a prediction becomes valuable when it improves a decision.
Outcomes and lessons
The inventory-aware pricing work was reported to increase profitability by 15% while reducing excess inventory by 20%. Related elasticity models covered a product catalog exceeding 10,000 items.
Other pricing and marketing initiatives also produced reported improvements, discussed separately in the technical deep dive; they should not be added to or conflated with the results above.
The larger lesson is that commercial ML must reflect the decisions it supports: financial objectives, inventory constraints, model uncertainty and real validation—not prediction accuracy alone.
Relevance to your business
This approach applies to companies managing pricing, inventory, purchasing, promotions or other recurring decisions where historical data can inform profitable action.
Explore the technical approach · View services · Discuss a similar project