How do you automate planning with AI?

What does automating planning mean?

Automating planning means that software makes a proposal for your planning or roster. The software does this with your rules and your data: who can work, who can do what and what needs to be done. The planner looks at the proposal, adjusts what is needed and decides.

Making a planning is a puzzle with many pieces. A technician has a day off, a customer only wants mornings, a job requires a certificate and an urgent job comes in between. An experienced planner keeps all of that in their head. Software can solve that puzzle too, if the rules are written down. The software does the calculating, the planner keeps an eye on it. In short: AI does the groundwork, you decide.

It is a form of workflow automation: work that comes up often and follows fixed rules is prepared for you by software.

Which planning can you automate?

Almost any planning that is always made according to the same rules. For SMEs, it is mainly about four types:

• Work planning for technicians and crews. Which technician goes to which job? Think of installers and contractors. See AI in construction (in Dutch).
• Project planning. Which job or order is up when, with which people and machines? See AI for manufacturing.
• Rosters. Who works which shift? In a healthcare practice or restaurant, it is about staffing levels, opening hours and employees’ wishes. See AI for healthcare practices and AI for hospitality.
• Trips. Which delivery goes in which van, and in what order? That is a puzzle of its own, with addresses and travel time. Read automating route planning and AI for transport.

The technology is roughly the same for all four. The difference lies in the rules.

What does AI do in planning, and what does ordinary software do?

Ordinary software does the calculating, AI does the reading. Those are two very different things.

The puzzle itself is calculation. You give the software conditions, such as: every shift gets one person, nobody works two shifts in one day and the shifts are divided fairly. Special software, a so-called solver, then looks for a roster that meets all the conditions. That is a well-known technique. Google explains it in the OR-Tools guide, with a roster for nurses as an example.

AI in the sense of a language model, as in ChatGPT or Copilot, is good with language. It reads an email, a text message or a job sheet and understands what it says. But a language model is not a calculator. If you ask it for a whole roster, it may miss a rule or make up something that looks convincing. That is called a hallucination.

A good set-up therefore uses both:

1. AI reads what comes in and turns it into data: a new job, a change, a cancellation.
2. Planning software applies your rules and makes a proposal.
3. The planner looks at the proposal and approves or adjusts it.

Which rules need to go into an automated planning?

All the rules a planner now has in their head. There are often more than you think. A list to start with:

• Availability. Contract hours, fixed days off, leave and absence.
• Skills. Who is allowed to do which work? Think of a diploma, a certificate or a driving licence for a trailer.
• Travel time and location. Where does someone start, where is the job, and how far apart are the jobs?
• Customer agreements. A fixed day, a time slot, a deadline or a technician the customer already knows.
• Resources. A van, machine or treatment room that can only be used by one person at a time.
• Working and rest times. What the law and your collective labour agreement (cao) say about shifts, breaks and rest.

Distinguish between hard and soft rules. A hard rule is one the system must not break, such as a mandatory certificate. A soft rule is a wish, such as a fixed crew or a preference for early shifts. The software takes those into account as far as possible.

What does the law say about rosters?

The Dutch Working Hours Act (Arbeidstijdenwet) has rules for rosters. A few you can build into software:

• Announce on time. You communicate a roster as early as possible. What counts as on time may be set out in your cao. If it is not, you agree it with the employee. If nothing has been arranged, you announce the roster at least 28 days in advance. If that is not possible because of the nature of the work, an employee must hear at least 4 days in advance at what times they have to work. They must then still hear 28 days in advance on which day their weekly rest begins (Article 4:2).
• Take people into account. You take employees’ personal circumstances into account, as far as is reasonable. You also record the fixed patterns of working and rest times in writing, so that employees can consult them (Article 4:1).
• Limits on shifts and rest. According to the Dutch government (Rijksoverheid), someone may work a maximum of 12 hours per shift and a maximum of 60 hours per week. Over 16 weeks, that is an average of at most 48 hours per week. After a working day, someone has at least 11 consecutive hours of rest. Once every 7 days that may be 8 hours, if the work requires it.

A cao may contain different agreements. Separate rules apply to young people under 18, to pregnant employees and employees who have just given birth, and to lorry drivers. Put the rules that apply to you into the software as checks. Then the system gives an alert if a proposal goes against them.

This is an explanation in plain language, not legal advice. For your own situation, check the law and your cao.

Where does AI help with planning?

AI helps mainly with everything that comes in as text. Together with the planning software, it also helps to spot problems. For example:

• Reading requests. A customer emails about a fault. AI takes the address, the type of job and the preferred day from the email and prepares a job. See also processing email automatically with AI.
• Processing changes. An employee calls in absent, or a customer moves their appointment. AI reads the message, and the software shows which jobs or shifts are affected.
• Taking job sheets into account. Does a digital job sheet say a job is not finished? Then AI prepares a follow-up appointment for the planner.
• Flagging conflicts. Two jobs at the same time, a shift without anyone holding the right certificate, or too little rest between two shifts. The software checks every proposal against your rules and reports what is not right.
• Explaining. With each proposal, AI sets out in plain language what has been moved and which alerts the software gave. That way the planner sees what they need to check.

What do you not leave to AI?

You do not leave decisions about people to AI. A roster determines when someone works and with whom. Sometimes it also determines how many hours someone gets, and therefore what they earn. That affects people.

The GDPR has something to say about this. A person has the right not to be subject to a decision taken solely by a computer, if that decision has legal effects or affects them significantly in another way (Article 22). Whether a roster is such a decision depends on the situation. The safe route is clear: the system makes a proposal, the planner decides. This is also called human in the loop. Do make sure that the planner actually assesses the proposal and can adjust it, and does not just click approve.

The EU AI Act is strict here too. AI that takes decisions about terms of employment, that allocates tasks based on employees’ behaviour or personal traits, or that monitors and evaluates their performance and behaviour, is on the list of high-risk AI (Annex III, point 4). Additional requirements apply to it from 2 December 2027, including for the business that uses such a system. Do you want to allocate work based on how someone performs? Then first find out which rules apply.

Also be careful with the data. Data about health is a special category of personal data, with stricter rules (Article 9 GDPR). So only record in the planning that someone is absent, not why. Does your planning run with a supplier in the cloud? Then you sign a data processing agreement with that party.

And there are things software does not know. That a customer prefers a technician they know, that a job turns out bigger than expected, or that someone is having a tough week at home. That is what the planner is there for.

What does that look like in practice?

Below are three made-up examples. They are not my clients, and they deliberately contain no figures.

How does that work for trips and deliveries?

For trips it works the same way: applying rules and making a proposal. For De Rijplaten Specialist and De Slowfeeder Specialist, two online shops from Ede, I built automatic transport planning. An order comes in through Shopify. AI chooses between DHL and own transport, and with own transport puts the order straight into the delivery driver’s route. That choice is about parcels, not people. With a roster, that is different. Read more in automating route planning and in the case.

Planning package, integration or Excel for now: what fits?

That depends on how large and how complex your planning is. An overview:

First see what your current package can already do. If it does what you need, you do not need to have anything built. Is your planning in an Excel file full of tabs and colours that only one colleague understands? Then read Your Excel lists have got out of hand.

How do you start automating planning?

Start small: one planning, one set of rules, and a person who approves every proposal. This is how to go about it:

1. Choose one planning. For example the weekly planning for the technicians, not everything at once.
2. Write down the rules. Sit down with the planner and keep asking: why do you put this technician here? Is the planning mainly in one person’s head? Then read The planning is in one person’s head and mapping processes.
3. Put the data in one place. Availability, skills and jobs, in a system that software can read, usually via an API.
4. Let the system run alongside for a while. The software makes proposals next to the normal planning. That way you see where rules are missing.
5. Have the planner approve every proposal. Even when things go well. Record who decides what.
6. Tell your employees. How is the planning made, and who can they go to with questions or objections?
7. Only expand after that. A second planning, or an integration with quotes and invoices.

Does it involve a lot of employee data? Then also check whether you first need to carry out a DPIA.

How do I help you automate planning?

I start with a business analysis. I look at how your planning is made now, which rules are in it and where the data is. After that you receive a proposal with a fixed price. Then I build it myself, in the systems you already use: the system proposes, the planner decides.

I receive no commission from suppliers, so I first look at what your current package can already do. Read more about integrations and automation and about how your price is set, or book a conversation.

Frequently asked questions

Sources: Dutch Working Hours Act (Arbeidstijdenwet) (in Dutch) (Articles 4:1 and 4:2, text in force from 1 July 2025), Dutch government (Rijksoverheid): What are the legal rules for my working and rest times? (in Dutch), GDPR, Regulation (EU) 2016/679 (in Dutch) (Articles 9, 22 and 28), EDPB: Automated decision-making, profiling and online tracking (in Dutch), EU AI Act, Regulation (EU) 2024/1689 (in Dutch) (Annex III, point 4), Regulation (EU) 2026/1744 (in Dutch) (date for high risk under Annex III), Dutch Works Councils Act (Wet op de ondernemingsraden) (in Dutch) (Article 27(1) and (3)) and Google OR-Tools: Employee Scheduling (a roster as a calculation puzzle with conditions). Consulted on 22 October 2026.