Talk to us
All solutions
All industries
All teams
How we work
Assess, build and run, with one accountable team before and after go-live.How we work 2 weeksfrom scattered ideas to a ranked, delivery-ready plan.The AI Assessment Data scientists and engineers shipping to production, then maintaining it.Data and AI engineering Your data stays yours, governed from the first day.Trust and governance
Featured stories
See all success stories
4.2 hto collect 2,800 financial reports with fifty agents in parallel, down from 210 hoursRead the story 35%more coverage of points of sale for Heineken field teamsRead the story 50k+cable designs tested virtually for CoficabRead the story 800K+calls a month transcribed, summarised and classified at NOSRead the story
Resources
Information Hygiene: The Security Discipline That Matters More Than EverRead the blog Artificial Intelligence Made in Portugal: ‘We grew more than 60% in one of the most competitive markets in the world,’ says DareData Partner.All news
Company
2019Founded in Lisbon by engineers. Today we build and run AI for large companies across Europe.About DareData 130+engineers, data scientists and principals in Lisbon, Porto and LondonMeet the team We are hiring engineers who want to build AI that runs.See open roles 50+customers served. Tell us about yours.Talk to us

CUF forecasts hospital demand 20% more accurately than its manual method

Daily demand forecasts across every hospital unit, on their first ML platform

Industry
Life Sciences & Health
Solution
Optimisation
Built in
2 months to MVP

Daily demand forecasts across every hospital unit, on their first ML platform

Industry
Life Sciences & Health
Solution
Optimisation
Built in
2 months to MVP

CUF forecasts hospital demand 20% moreaccurately than its manual method

Client
CUF
Industry
Life Sciences & Health
Solution
Optimisation
Built in
2 months to MVP
Technology
Machine Learning
Team
4 DareData engineers
20%better forecast accuracy than the manual method, validated against the October to December hard close
46models running in production
9budget cycles a year, instead of one manual round in October

CUF planned hospitalisation and emergency demand with a simple rule, produced by hand once a year during the October budget cycle. A forecasting system now produces three-month rolling forecasts at daily granularity for every hospital unit, clinical scope and emergency typology, with a model chosen per segment and a monthly pipeline that picks the champion. Validated against the October to December hard close, it is 20% more accurate than the manual method.

With DareData, CUF:

  • Forecasts 20% more accurately than the manual budgeting method, validated against the hard close
  • Produces three-month rolling forecasts at daily granularity for every unit and typology
  • Runs 46 models in production, with a champion picked every month
  • Keeps absolute error at 4.2% for hospitalisations and 4.4% for emergency demand
  • Supports 9 budget cycles a year instead of one manual round
  • Delivered a complete MVP across both workstreams in 2 months

The challenge

One manual forecast a year

CUF runs 11 hospital units with around 10,000 staff. Planning beds, staffing and budgets across that network depends on knowing how many patients will be admitted and how many will walk into emergency care, unit by unit and day by day.

A yearly forecast made by hand

Demand was forecast with a simple rule, produced manually once a year during the October budget cycle. Individual judgement shaped both the forecast and every revision that followed, with no standard method behind it.

No daily or unit-level view

The forecast had no daily granularity and no breakdown by unit, clinical scope or emergency typology. Planning teams could not see how demand would differ between a large medical unit and a small one, or between adult, paediatric and obstetric emergencies.

Starting machine learning from zero

There were no models in production and no infrastructure to run them. Any forecasting system had to arrive with its own production discipline, and leave CUF's team able to build on it.

The solution

A model chosen for every segment

The system forecasts daily hospitalisation volumes and emergency demand across all CUF hospital units, broken down by unit, clinical scope (medical or surgical) and emergency typology (adults, paediatrics and obstetrics). It produces three-month rolling forecasts and delivers them on a business dashboard.

The work ran as two parallel workstreams on one shared foundation. Hospitalisations were led by DareData, while emergency demand was developed by CUF’s own team with DareData mentoring. Before any model could train, data from two hospital registry systems had to be unified in CUF’s Databricks warehouse, then aggregated from individual patient records into daily admission counts per unit, scope and typology, using history from 2019 onward.

Segments behave differently enough that one model does not fit them all. Statistical, machine learning and heuristic approaches were compared for each combination, and the best one was kept per segment: Prophet for scheduled surgery, which follows a strong planned pattern, ensembles for large medical units, and simpler heuristics for small units where volume is too low to learn from.

Production discipline was built from scratch alongside the models. MLflow handles the model registry and experiment tracking, a monthly pipeline runs every model automatically and selects the champion, and unit, integration and data-quality tests cover every layer. Everything runs on Databricks, with PySpark for large-scale aggregation.

How it works

From data to decisions, in 4 steps.

  1. 01Unify the dataTwo hospital registry systems are brought together in the Databricks warehouse.
  2. 02Count demand dailyPatient records become daily admission counts per unit, scope and typology, from 2019.
  3. 03Pick a model per segmentStatistical, machine learning and heuristic approaches are compared for each unit and scope.
  4. 04Run it every monthAn automated monthly pipeline selects the champion model and feeds a business dashboard.

The impact

Sharper forecasts, nine budget cycles a year

  • 20%better forecast accuracy than the manual method, validated against the October to December hard close
  • 46models running in production
  • 9budget cycles a year, instead of one manual round in October

Forecasts are now 20% more accurate than the manual budgeting method, measured against the October to December hard close: the real accounting close, not a test set. Absolute error sits at 4.2% for hospitalisation forecasts and 4.4% for emergency forecasts.

Instead of a single manual round in October, the business can run 9 budget cycles a year, each starting from a data-informed forecast. 46 models run in production across every unit and typology, and the complete MVP across both workstreams reached the business in 2 months.

The project was also CUF's first machine learning platform. Emergency demand was built by CUF's own team with mentoring, on the same MLOps foundation, so the capability stays in house. The client rated the project 9 out of 10.

The team

The people who built it.

  • Leonid KholkineSenior Data Scientist & Project Manager
  • Alexandre ChambelSenior Data Engineer
  • Davide FarinatiData Scientist
  • João MirandaData Scientist
The solution behind this storyData and AI engineeringSee how we build it

Your AI story, next.

Tell us what you want AI to do for your business. We design, build and run it with your teams, and measure it like the stories above.

  • 01Use cases chosen for value, not hype
  • 02Systems built into your stack and data
  • 03Run, monitored and improved after go-live
Talk to usBook an assessment

More from production.

All success stories
Rui MendesHead of GrowthHave a similar problem? Book 30 minutes.No slides. We look at your use case together.Pick a time