Operations Research for better decisions

What could your organization achieve if trade-offs were quantified instead of debated?

Many organizations solve a complex puzzle every day when allocating their limited resources, such as:

– Which customers should we group into one route for grocery delivery?
– How do we efficiently schedule our maintenance activities and personnel?
– Where should we locate new warehouses for our logistics operations?

These puzzles are often solved manually, relying on people with years of experience. This is complex and time-consuming. There is also no guarantee that a good solution has been found.
Even when an automated process is already in place, having the absolute best solution is often not the most important objective if it takes days or weeks to compute. A feasible solution that can be found within a reasonable time and actually be implemented delivers far more value.

This approach even allows you to calculate and compare multiple scenarios.

These types of challenges fall within the field of optimization or Operations Research. At Dataworkz, we know that solving this puzzle is not only about smart and mathematically optimal solutions. The entire chain—from raw data to end product—is at least as important, ensuring that the solution can be executed reliably and automatically on a daily basis.

All while your data scientists continue to test and implement new models in the background, without disrupting the business.

In addition to our experience in “pure” optimization, Dataworkz has extensive experience in setting up the full end-to-end chain, which we also refer to as DecisionOps (as the counterpart to MLOps in the Machine Learning domain):

– Automated ingestion of data from your business databases or data lake
– Determining and safeguarding input data quality
– Iterative development
– Controlled and repeatable scenario testing
– Versioning and controlled deployment of improved optimization models
– Monitoring decision quality and observability
– Confident handover to your organization

There is significant overlap here with our expertise in Machine Learning, Data Science, and Data Engineering.

Our cases

Below are several concrete cases within the Operations Research domain that Dataworkz has worked on:

An trading application that operates fully autonomously, calculating the optimal charging schedule for a large-scale battery every minute, controlling the battery accordingly, and executing trades on the intraday market. (Essent)
Determining the optimal charging schedule for electric vehicles based on market prices, sending control signals, and executing the required trades on the intraday market. (Essent)
Determining the optimal grid layout for all home connections and medium-voltage infrastructure in a new residential area. The application reads environmental data and generates a plan for the placement of electricity cables in the most efficient way possible. (Alliander)
Creating a multi-quarter planning schedule for contractors and technicians in the energy transition. The goal is a more stable and reliable workload and sufficient contractor and technician capacity. (Alliander)
Optimally using solar panels and batteries when charging delivery vans, ensuring they have enough energy for next-day parcel deliveries. (DHL)
Optimizing the placement of parcel lockers depending on demand in an area and other regional characteristics. (DHL)
Optimizing personalized offers to maximize relevance within a set of business rules. Using an optimization model, business rules—such as variation and budget constraints—are translated into constraints. Customers do not receive repetitive offers, while distribution across product categories remains intact. The result is a balanced offer that complies with business rules. (Jumbo)
An AI tool (LiMiTS) to optimize aircraft maintenance planning at the gate. The tool intelligently assigns engineers based on tasks, skills, availability, and real-time flight data, making the planning process more efficient and balanced. As a result, workload peaks are reduced, planners save time, and aircraft become available faster with fewer delays. (KLM)

Our Proposition

Dataworkz is your ideal partner for Operations Research and optimization challenges:

We strive for well-founded, data-driven decisions.
We leverage the years of business experience your employees bring.
Decisions driven by algorithms with measurable objectives and constraints.
Specialists with a mathematical background and extensive experience in optimization across various industries.
With our data science and data engineering expertise, we support your organization from start to finish.

Our Approach

Together with Your Business Experts

In all our projects, the starting point is collaboration. This is especially true for Operations Research. The strength of Operations Research lies in translating business knowledge into mathematical models. The years of accumulated knowledge and experience of your employees are essential in this process. We therefore prefer to collaborate very closely with the business. An added benefit is that your organization will feel comfortable with the model results because it’s their knowledge translated into a model.

Clearly Formulating the Problem

First, together with the business, we formulate the problem as concretely as possible. Objectives must be clearly defined, and all constraints, limitations, rules, and interactions described.

First Version of a Model

Once this is clear, we translate all objectives and constraints into a mathematical model and develop an initial version. The goal is to quickly assess expectations, identify key challenges, and detect problems early.

Moving to Production

When results are sufficiently satisfactory, we further develop the first version into a production-ready application. This is another area where Dataworkz excels. In Operations Research, this is often referred to as DecisionOps—the OR counterpart to the better-known MLOps. Models must not only work on a single data scientist’s laptop but must become full production applications that all your employees can use at any time.

Monitoring

We implement systems to continuously monitor whether everything functions properly and maintains high quality. We integrate with your existing observability platforms, such as Datadog, Grafana, or MLflow. Changes to data transformations or the OR model can be properly evaluated before going into production, ensuring output quality remains safeguarded.

Confident Handover to Your Organization

This means, among other things, that the model runs independently within your systems, is robust, well-tested, secure, and easy for the business to use. We prefer to do this together with your data scientists and engineers. This ensures continuity and allows us to step back with confidence once the project reaches sufficient maturity. Your IT department can then maintain and further develop the solution independently.

Safeguarding the Delivered Work

It is important to note that Operations Research models are generally reproducible and explainable. In other words, they are not black boxes like many ML models. It is possible to clearly explain why certain decisions were made. Moreover, if your business would have made a different choice, it can often be calculated whether that choice would have been better or worse—and why. This is a powerful way to build trust in a new model.

Result

We begin by investigating feasibility through the rapid development of a first, simple version of a model. It will not yet be production-ready but will respect all business requirements. The goal is to quickly assess whether a strong business case exists, rather than discovering after a year that objectives cannot be achieved.

Depending on the size of the challenge, the number of departments involved, and external factors, this phase typically takes one to three months. After this phase, more concrete goals can be set regarding cost savings or more efficient resource utilization.

One result that can already be stated in advance is that your organization will be able to make better-founded decisions based on all business factors and constraints defined together beforehand. Decisions will be more explainable, monitored, and validated. Your organization will also become more agile. Changes in the market, legislation, regulations, and objectives can be easily incorporated into the model.

Why Dataworkz

At Dataworkz, we believe in implementing simple solutions. We prefer to do this together with your IT department. No buzzwords or glossy PowerPoints—just writing solid code to solve real problems.

We combine data scientists and engineers in one team, ensuring we deliver not just a model that works on a laptop, but a mature application that can be widely used across your organization.

At Dataworkz, we share knowledge internally and with our sister companies under the Thisworkz organization. This means you are not just hiring a single data scientist or engineer—you benefit from the collective expertise of many professionals across different disciplines.

How to get started

Do you recognize these challenges? Then Operations Research is likely a suitable solution for your organization.

Contact us and we can organize a brainstorming session to explore how we can support you. You are welcome to visit our office, where together with several of our Operations Research experts, we will review your challenges. We can present examples of solutions implemented at other organizations and determine the best approach for you.

And if it gets intense, you can always relax in the massage chair—or unwind with a game of Mario Kart.

Operations Research or Machine Learning

You may wonder why your problems cannot simply be solved using AI or Machine Learning. After all, AI is a hot topic. Below is a comparison.

Key Aspect	Machine Learning (ML)	Operations Research (OR)
Main purpose	Predict or learn patterns from data	Optimize decisions under constraints
Best used when	You have lots of historical data and want forecasts or scoring	You need the best plan, schedule, or allocation
Typical questions answered	“What will happen?” “Who is likely to do X?”	“What is the best action given our limits?”
Handling constraints (rules, limits)	Weak — constraints are hard to enforce strictly	Strong — constraints are central to the model
Explainability to stakeholders	Often limited; models can be hard to explain	High; logic and trade-offs are transparent
Reliability & guarantees	Probabilistic; no guarantee of optimal decisions	Strong guarantees if the model is solvable
Common business uses	Forecasting, customer scoring, demand prediction	Scheduling, routing, capacity and resource planning

In summary:

Use Machine Learning when making predictions and when you have high-quality historical data.
Use Operations Research when the challenge focuses on decision-making involving rules, trade-offs, and costs.
In practice, a combination is common. Dataworkz also has extensive experience with MLOps.

A combination often occurs in practice as well. Dataworkz also has a great deal of experience with MLOps!

Another overview of typical ML use cases:

Use-Case Aspect	ML: Why it fits / doesn’t fit	OR: Why it fits / doesn’t fit
Demand forecasting	✅ Excellent at learning complex patterns from historical data❌ Poor if data is sparse or non-stationary	❌ Not designed for prediction✅ Can incorporate forecasts as inputs
Scheduling & resource allocation	❌ Hard to enforce strict constraints	✅ Natural fit; constraints and objectives are explicit
Routing & logistics	❌ Can approximate decisions but lacks guarantees	✅ Classic OR domain (TSP, VRP); strong optimality results
Pricing & revenue management	✅ Learns price–demand relationships	✅ Optimizes prices given demand models
Fraud detection / anomaly detection	✅ Strong at pattern recognition in large datasets	❌ Not designed for classification problems
Strategic planning	❌ Data alone may not capture policy constraints	✅ Strong for scenario analysis and trade-off evaluation
Dynamic decision making	✅ Reinforcement learning excels with enough data	✅ Dynamic programming works well for structured problems
Highly constrained environments	❌ Constraints are awkward to encode	✅ Core strength of OR
Unstructured inputs (text, images)	✅ Core ML strength	❌ OR cannot directly handle unstructured data
Explainability & auditability	❌ Often difficult to justify decisions	✅ Clear rationale and traceability