Product · for any sector

Custom Models

Bespoke prediction models trained on your data - demand, claims volume, production, footfall, traffic. Built by the team behind our weather engine, and back-tested before you commit.

Back-test of a custom prediction model: predicted values closely tracking actual target data over three months, with the forecast horizon extending beyond the last observation.
Trained on your data Weather & market data included Back-tested first API · dashboard · reports
What it does

Your data has a model inside it

We join your history with our weather, market and socioeconomic data - and train a model for the one number your business runs on.

Your data + ours

Sales, claims or production history joined with 20 years of weather and market context.

Back-tested before you buy

The prototype phase ends with numbers on held-out seasons - you see the accuracy first.

Any target

Demand, claims volume, footfall, output, occupancy, risk - if it has a pattern, it can be forecast.

Probabilistic by default

Distributions and confidence, so you can plan on the risk level you choose.

Delivered your way

API feeds, dashboards or scheduled reports - matched to how your team works.

Maintained & retrained

Models keep learning with every season; we watch the accuracy so you don’t have to.

How it works

From sample data to production

STEP 01

Share sample data

A season or two of history is usually enough to scope the model - under NDA, in any format.

STEP 02

We prototype & back-test

A working model on held-out data within weeks - with honest accuracy numbers, not slides.

STEP 03

Production & handover

We run it as a service, or build it and hand it to your data team - your call.

Two ways to run it

We run it - or we build it and hand it over

Model as a service

We operate it for you

You consume forecasts; we own the pipeline, monitoring and retraining.

  • Forecasts via API, dashboard or reports
  • Retraining and monitoring included
  • No ML team needed on your side

Build & transfer

For your data team

We design, train and document the model - then hand it over with the playbook.

  • Full documentation and code handover
  • Training for your team included
  • Support agreement optional
At a glance
Inputs
Your historical data plus our weather, socioeconomic and market datasets.
Targets
Any measurable KPI with history - demand, claims, production, traffic, occupancy.
Validation
Back-tests on held-out seasons; accuracy reported before production.
Timeline
Scoping in days, prototype in weeks, production per integration.
Ownership
IP and data ownership fixed per agreement; your data never trains other clients’ models.
Delivery
API, dashboards, reports - operated by us or handed over to you.
Questions we get

Before your data team asks

How much data do we need?

Usually one to two seasons of history. We’ll tell you quickly - and honestly - if it isn’t enough for a reliable model.

What if the model doesn’t beat our baseline?

That’s what the prototype phase is for: you see back-test numbers against your baseline before committing to production.

Who owns the model?

Per agreement - from a service you subscribe to, up to full IP transfer with handover.

Is our data safe?

Processing runs under NDA and a data-processing agreement; your data is never used for other clients.

Why us
“The hardest part of a custom model isn’t the algorithm - it’s the data that has to surround it. We bring twenty years of it.”
Forecasts.cloud · 15+ years of machine learning for European clients
15+ yrs
of machine learning for European clients
40+
weather models already in our stack
20 yrs
of historical data to enrich yours
weeks
from kickoff to a back-tested prototype
The same team and stack behind our weather engine, claim tools and pricing hub
More products

One engine, many tools

Bring us the number you wish you knew in advance.

Tell us what you’d like to predict and what history you have - we’ll scope a prototype and tell you honestly whether it’s worth building.

Discuss my use case