Forecasts you can defend — and plan on
A multi-model forecasting engine that proves itself on your history before it goes live, then feeds planning with distributions, not just a single number.
- 12+
- Forecasting models
- 55% → 85%+
- Accuracy at Decathlon
- P10–P90
- Distributions, not point guesses
Why Convect for demand forecasting
Demand Forecasting, done right
Every model is tested before it’s trusted
Before a model is used, it’s backtested on your own history. Accuracy, stability and fit are scored automatically, and only proven approaches go live.
- Historical backtesting
- Model performance scorecards
- Strategies by product segment
A state-of-the-art model library
Classical time series like ARIMA, gradient boosting like LightGBM, and deep learning like LSTM, N-BEATS and Transformers — chosen automatically or blended.
- Latest ML methods
- Deep learning models
- Continuously updated
Ensembles that reduce uncertainty
Consensus across models captures the strengths of each. Forecasting the full demand distribution gives inventory and production optimization a far more reliable input.
- Multi-model consensus
- Distribution forecasts
- Confidence intervals
Platform
How the three layers power demand forecasting
Demand Forecasting isn’t a standalone tool. It runs on the same ontology, optimizer and agents as production scheduling, so plans connect end to end.
AI Planning Agents
The Demand Sensing Agent monitors sales and external signals and triggers a re-forecast and plan update when deviations cross a threshold.
Contour Ontology
SKUs, channels, stores, promotions and holidays are modeled as entities and relationships — so a new product can borrow patterns from its look-alikes.
Optimization Engine
Forecasts flow into optimization as distributions and directly drive safety stock, replenishment and production — not just a report.
Scenarios
Typical scenarios
New products
Forecast items with under six months of history using look-alikes and attributes
Seasonal items
Detect and adapt to cyclical swings and catch the peak early
Promotions & events
Quantify promo lift to avoid both stockouts and leftovers
Omnichannel demand
Forecast by channel and region, reconciled top-down
Customer results
Proven by our customers
RetailDecathlon lifts forecast accuracy from 55% to 85%+
With new products making up more than a third of the range and strong seasonality, Decathlon used Convect’s Algoflow engine to automatically select and blend the right model for each item — giving replenishment and production a forecast they can trust.
55% → 85%+
Forecast accuracy
1/3+
Of range is new products
Auto
Model selection per item
Read the story
Consumer goodsUnilever cuts network inventory by 10%+ at the same service level
For a nationwide, multi-echelon distribution network, Convect combined probabilistic demand forecasting, network-wide deployment optimization and simulation to set the right safety stock for every SKU at every DC — cutting inventory by more than 10% in simulation, with service unchanged.
10%+
Lower inventory (simulated)
Same
Service level maintained
Every DC
SKU-level safety stock
Read the story
Further reading
Related articles

Technology·Oct 24, 2026
What is a supply chain ontology — and why do AI agents need one?
Your ERP stores data; it doesn’t understand it. A plain-English guide to supply chain ontologies, with a worked example of a line going down for maintenance.

Agentic AI·Nov 4, 2026
Why LLM agents alone can’t plan your supply chain
Understanding the factory, computing the plan and reacting to change are three different jobs. An agent that only has a language model is missing two of them.
Get started with demand forecasting
One ontology and one optimizer from demand to supply — so every plan connects to production scheduling.