Data audit and problem framing
We check whether your data holds a signal to forecast from. If there is no signal, we will say so before you spend budget on a model.
Forecasting · ML · Data analytics
Predictive analytics is useful exactly as far as you can trust the forecast when you make a decision. We build models whose accuracy is measured, whose limits are known and whose behaviour is understood when the data fails.
For companies where decisions on purchasing, production or hiring rest on an estimate of demand, load or risk, and where a forecasting error costs money.
We check whether your data holds a signal to forecast from. If there is no signal, we will say so before you spend budget on a model.
Time-series and gradient-boosting models that account for seasonality, promotions, the calendar and external factors. We compare every model with a simple baseline: if it does not beat the baseline, it is not needed.
We predict failures from telemetry: we label events, select features and work out the cost of errors. A false alarm and a missed failure cost different amounts, and the model is tuned with that difference in mind.
Scoring models with explanations: why the model gave this score, which features drove it and in which cases its decision cannot be trusted.
The model in production: a data pipeline, regular retraining, drift monitoring. Forecast accuracy drops over time, so it has to be watched after handover too.
We look at what there is: completeness, depth of history, labelling quality. Before work starts, we estimate what accuracy can be reached.
We build the simplest possible solution. It becomes the bar: anything more complex has to beat it, or there is no point in it.
We improve a metric tied to your economics: we calculate what a forecasting error costs in money.
Data pipeline, monitoring, retraining. We hand it over with documentation and train your team.
We compare any model with a simple baseline. If the model does not beat it, the model is not needed, and we will say so.
If the data holds no signal, you will learn that from the diagnostics, before the main spend. In our own research we also publish negative results alongside positive ones.
One team handles the data, the model, the infrastructure and the interface, so responsibility does not get lost between contractors.
It depends on the task and the seasonality. A demand forecast usually needs a history covering 2–3 full seasonal cycles. We give an exact answer after the data diagnostics: its results show whether the required accuracy can be reached at all.
BI answers the question “what happened”; predictive analytics answers “what will happen, and how likely is it”. BI describes the past. A model estimates what has not happened yet, and it is always off by some amount that the decision has to take into account.
The model will be wrong; the question is how often and at what cost. So from the start we calculate the cost of an error and design the system so that expensive errors are rare, even if cheap ones become more frequent.
Yes. The model can be deployed inside your own infrastructure if the data cannot leave the company perimeter.
R&D · Research · Development
Multi-agent architectures · LLM · Autonomy
AI implementation · Process automation · Integration
The AVA calculator works out the economics of your process from your own figures, with no form and no price from us. If automation does not pay off, it says so.
Estimate the paybackDescribe the problem in your own words. If it can’t be solved or won’t pay off, we’ll say so before any work starts.