Forecasts you can plan against
Bottom-up forecasts per station and customer, built with advanced machine learning on your own historical data. A built-in collaboration workflow lets station heads submit, country teams correct, and group sign off, without spreadsheet tennis.
What it does
The handful of things Forecasting earns its keep on.
Station-customer ML forecasts
Where the P&L lives, not a network-wide average. 95% accurate, back-tested every close.
Bottom-up collection workflow
Station heads submit the first number. The system collects, harmonises, and queues for review.
Correction and approval rounds
Country and group review, request changes, approve. Every override has an author, a reason, and a rollback.
Override the model where it matters
Local knowledge wins on contracts and one-offs. Everyone sees what changed and why.
Weekly or monthly refresh
Rolling, not annual. Plans refresh with the operation.
Cargo and ground handling on one engine
Ground handling and cargo share the same ML stack and the same workflow.
Models trained on your own history
The forecast is machine learning applied where it pays: per station and per customer, learned from your seasonality, your contracts and your disruption history rather than a network-wide average.
Two years of shipment history, grouped by aircraft type. Freighter, passenger belly and road feeder volumes occupy separate worlds, which is why a single network-wide model forecasts none of them well.
- Feature engineering, not a generic model Two decades of knowing which patterns matter in this sector is what gets the accuracy, not the algorithm on its own.
- Back-tested every close The forecast is scored against what actually happened, so 95% is a measured number rather than a claim.
- Local knowledge overrides the model Station heads can override any figure. Everyone sees what changed, who changed it and why.
Our forecasting work with WFS was recognised at the Air Cargo News Awards 2025.
There was no number to beat
When we ask what forecast accuracy was before Cohelion, there is rarely a figure. Not because anyone was careless, but because every miss had a genuine explanation, and a number that can always be explained away never becomes a number anyone keeps.
Every one of those is real, and several are now features in the model rather than excuses for it. The rest are why we publish a back-tested figure every close, measured the same way each time. A miss becomes something you investigate, not something you explain.
CFO first, CCO · COO too
For finance and commercial leaders who want a plan the stations stand behind, and for station heads who want to own their numbers.
"The shift from traditional planning to data-driven decision-making has not only improved our operational efficiency but also reduced costs."
How it fits with the others
Forecasting earns its keep on its own. Combined with the other three, it produces named outcomes you can't buy off the shelf.
Named outcomes it produces
Paired with the other products, Forecasting becomes one of these. Each is a decision somebody has to make, not a feature.
Talk to us about Forecasting
We'll set up a 30-minute walk-through. Forecasting on a dataset that looks like yours.