Steering media bids by product margin and forecast demand

Bid on the real margin of each product, then on forecast demand: higher ROAS and lower cost per lead across more than ten markets.

Challenge

In a tyre market with thousands of references, volatile margins and strong seasonality, classic bidding models optimised volumes, not profitability. Michelin wanted bids that reflect the real economic value of each product: more than 2,700 margins per market to inject into the platforms, safely and in real time.

Results

+53%

ROAS and -20% cost per lead with margin-indexed bidding (VBB v1)

+48%

ROAS and -78% cost per lead in Germany with forecast and margins (VBB v2)

+49%

ROAS and -33% cost per lead in Italy with forecast and margins (VBB v2)

10+

markets covered, from the United States to Australia

Grand Prix de la Data, Gold Award 2025 and 2026

Our approach

Proprietary margin data
Each Buy Now click sends a unique identifier, matched to a margin secured and indexed by Michelin. A server-side set-up (GTM) passes these margins to the platforms, so bids follow the real value of each product.

Predictive demand engine
A model built on Vertex AI combines history, weather, behavioural signals, seasonality and market trends into a forecast score that anticipates demand several weeks ahead.

Two value-based bidding models
VBB v1 optimises on product margins alone, VBB v2 crosses margins with forecast demand, so investment becomes dynamic and follows real market opportunities.

International roll-out
Workshops, training, playbooks and central support industrialised the approach in more than ten countries.

Let’s talk about your project

Tell us about your needs: we will get back to you quickly.