Skip to content

Occupancy forecasting with data and AI

First, we measure actual occupancy, then we forecast it. For Port of Amsterdam, we measured historical berth usage.

For Port of Amsterdam, we developed a tool that measures historical berth usage and makes it easier to understand

In a port area where space is scarce, policymakers decide whether to increase or reduce the number of berths for certain types of vessels. The tool lets them base those decisions on data rather than gut feeling alone.

Read the case study

How can we know before today how full it will be tomorrow?

Beds, berths, rooms or vehicles: if you only see occupancy on the day itself, you have to scramble to organise staff and space. A model that forecasts occupancy by day and time of day gives planners time to make adjustments.

Book a call

An occupancy forecast starts with an accurate measurement

Discuss your occupancy

Measure what happened

For Port of Amsterdam, we measured historical berth usage so decisions are based on figures.

Count without manual work

For GVB, our software uses computer vision to count passengers over an extended period rather than relying on a snapshot.

Alert departments earlier

At St. Antonius, our algorithm supports doctors so departments hear sooner about a possible admission.

By day and time of day

An occupancy forecast with a range that planners can compare with their own estimates and adjust.

The measurement must match reality before we start forecasting

  1. Measure occupancy

    Use records, sensors or counts to build a reliable picture of actual occupancy for each part of the day.

  2. Model and test

    A forecast based on historical data, the calendar and known appointments, tested against periods the model has not seen.

  3. In the planning process

    The forecast becomes part of planning discussions, with a designated owner and regular checks for deviations.

Can we predict our occupancy well enough to build a model?

Let’s assess it together

A good fit if

Occupancy varies by day or season but follows a pattern, you have historical data on actual occupancy, and a planner needs to arrange staff or space in advance.

Not a good fit if

Occupancy depends almost entirely on fixed appointments already in the calendar. In that case, a clear overview of the schedule is enough, and we’ll tell you so.

What determines the scope

Whether actual occupancy is already recorded or first needs to be measured, how many locations and units there are, how far ahead you want to forecast, and who will manage the model.

What if something unexpected happens?

A model only knows what has happened before. That’s why it shows a range, the planner has the final say, and we flag any persistent deviation from actual occupancy.

Rule of thumb, schedule overview or predictive model for capacity planning?

A fixed rule of thumb

Choose this if occupancy is roughly the same each week. A fixed occupancy level for each day of the week is simpler and works just as well.

A schedule overview

Choose this if occupancy mainly comes from scheduled appointments. A report of what’s booked will show you what to expect.

A predictive model

Choose this if occupancy varies with the season, weather or unplanned arrivals. Our advice: make sure you measure actual occupancy first.

We’ve measured and predicted how busy it gets at the port, on public transport and in the emergency department

Data Science Lab has been building models that work in practice since 2016, as a Microsoft Partner and Databricks Partner. We’ve worked for Port of Amsterdam, GVB and St. Antonius.

Book an introductory meeting

In our first conversation, we’ll discuss the occupancy figures your planners currently see too late

Stef Stoelwinder

Stef Stoelwinder

Business Development

Stef is your point of contact. In our first conversation, we’ll discuss what capacity you want to forecast and how occupancy is currently recorded.

Which occupancy levels would you like to know a week in advance?

Tell us which capacity you mean and how you currently track occupancy. We’ll reply by email or arrange an initial conversation.

Enter a valid email address.

Let us know how we can help you.

Frequently asked questions

A model that estimates how much capacity will be in use in the future: beds, berths, rooms, workspaces or vehicles, by day or part of the day. Planners use that estimate to arrange staff and space in time.

You can find other uses of data and AI under AI applications.

Forecasting predicts demand in general, such as sales or patient numbers. Occupancy prediction focuses on the capacity that demand uses and how long it remains occupied.

That means the length of stay or use matters alongside the number of arrivals.

A time series of actual occupancy for each part of the day, including arrivals and departures. You also need calendar data, scheduled appointments and sometimes weather or event data. If occupancy is not yet measured, we first build a way to measure it, as we did with the berth tool for Port of Amsterdam.

How far back your data goes determines how well the model can learn seasonal patterns.

You can use existing records, such as check-ins and check-outs or hospital admissions and discharges, or use sensors and camera footage. For GVB, our computer vision software counts passengers; the footage is not viewed or shared, only analysed by the model.

You can read about the privacy requirements for these measurements under DPIA for AI.

Yes, using admissions, discharges and length of stay by department. At St. Antonius, our algorithm supports doctors in deciding whether to admit a patient, so departments can prepare sooner. You can find that case study among our case studies.

Read more about data in healthcare under healthcare.

By testing it on periods the model has not seen and comparing it with the current way of estimating occupancy. We look at the error for each part of the day, especially at busy times, when decisions matter most.

We keep making that comparison after launch, so you can see if the model starts to drift.

Where they already work: in the planning software, a report or a daily email. We include a range so the planner can see how certain the estimate is.

Read about how we keep models running in production under MLOps.

When occupancy comes almost entirely from fixed appointments, or is the same every week. In those cases, an overview of the schedule or a fixed rule of thumb is enough.

If your main concern is demand for products or services, see forecasting.