Time series forecasting
Every plan starts from an estimate of what comes next: next month's sales, tomorrow's energy use, how busy a network or warehouse will be.
We build forecasting systems that learn from your history and what drives it, and give a range of likely outcomes alongside the central estimate.
What a forecasting system does
A forecasting system learns how a quantity has changed over time and what influences it, such as calendar effects, weather, prices or campaigns, and estimates its values for the periods you plan for.
A single number hides how uncertain the future is. A range shows how far actual values may land from the most likely outcome, and that spread is what stock levels, staffing and reserves are usually sized against.
What we build
Demand forecasting
Expected sales or orders per product, location and period, including items with little or no history.
Load and consumption forecasting
Short- and medium-term forecasts of energy, water or resource use, taking weather and calendar into account.
Capacity planning
Expected traffic, workload or occupancy. Teams use it to plan people, space and equipment ahead of time.
Sensor and equipment forecasting
Projections of machine and sensor readings that show when a value is likely to cross its limit.
Cash flow and revenue forecasting
Incoming and outgoing payments or revenue expected in each period, for the finance team's planning.
When it's a good fit
Forecasting earns its keep when a decision based on expected volumes comes up again and again, and getting the number too high or too low costs money. A simpler approach may be enough when:
- patterns are stable and a simple seasonal average already comes close
- there is very little history and no comparable items to learn from
- the forecast would not change any decision, because plans are fixed long in advance
How a project runs
-
Start from the decision
We clarify which decision the forecast feeds, for which periods and at what level of detail. This sets how accuracy will be judged.
-
Measure today's method
We measure your current forecast or planning method on past periods. A new model is only worth using if it beats that method on the same data.
-
Build and back-test
We test each model as if it had been used in the past, one period at a time, with only the data available at that point.
-
Run alongside current planning
The new forecast runs in parallel with the existing process for a while. Planners compare both before switching over.
How we check forecasts
A forecast can score well on past data and still fail once conditions change.
Ranges are checked too
Besides the central estimate, we check whether actual values fall inside the stated ranges as often as they should.
Unusual periods get their own check
We evaluate holidays, campaigns and disruptions separately, because that is where forecasts most often go wrong.
What you receive
- Forecasts delivered on a regular schedule into your planning tools
- A comparison against your current method on past data
- Code and documentation, handed over to your team
Frequently asked questions
How accurate can a demand forecast be?
It depends mostly on how stable the pattern is and how much history there is. Comparing new models with your current method on past periods shows what accuracy to expect for your data.
What data do we need for time series forecasting?
The history of what you want to forecast, plus whatever you already know affects it, such as prices, campaigns or holidays. We can start with what exists and show where more data would help.
How long does a forecasting project take?
The first comparison with your current method usually comes early in the project. The rest depends on data quality and how many things you need to forecast.
Where does our data stay during a forecasting project?
It can stay with you. We can train and run the forecasting models on your own servers or private cloud.
What happens when patterns change after go-live?
We track accuracy after go-live and retrain or adjust the model when patterns shift. Your team can run the retraining itself, or we can keep supporting it.
Work with us
Tell us in a few lines what you're working on. We'll get back to you.
From our blog
All articles
What AI evaluation metrics miss in production
Strong AI evaluation scores can hide production failures when the chosen metric does not reflect business risk, user outcomes, or operational cost.
Read article
Will a proven AI method work on your data?
A proven AI method can still fail on your data and workflow. Testing shows whether published evidence transfers to your specific constraints.
Read article
When should you use synthetic data?
Synthetic data can help when real data is scarce, sensitive, or unavailable. When is it worth using, and how can you tell whether it improves results?
Read article