Every company that operates in the physical world runs on a plan, and most of them still build it in a spreadsheet. Someone decides which of 200 field technicians visits which of 800 customers tomorrow, in what order, with which parts are loaded in the van. Someone builds next month's roster for a hospital ward so that every shift has the right skill mix and nobody works a night followed by an early start. These look like administrative chores, and mathematically they are monsters that cost real money every day they are solved badly.

That gap is the one Timefold closes.

Last month the Ghent-based company announced a $13 million Series A led by Alstin Capital, with KOMPAS VC co-investing alongside existing backers Lakestar and Smartfin. Timefold is building what we believe will become the default technology stack for planning optimisation: the software that decides which technician drives where, which nurse works when, and which task runs on which machine, and does it measurably better than any human planner or general-purpose AI. Here is why we invested.

The hardest problems are the ones that look mundane

These look like administrative chores. Mathematically, they are monsters. Take a single delivery driver with fifteen stops. The number of possible orders in which to visit them is fifteen factorial - roughly 1.3 trillion. Checking one ordering per millisecond, evaluating them all would take more than forty years. By sixty stops, the possible orderings for that one driver already outnumber the atoms in the observable universe - roughly 10⁸⁰ - and that is before assigning jobs across a fleet, or accounting for skills, time windows, traffic and labour rules. 

The plan is also never finished. A technician calls in sick at 6:47am, a customer cancels, a job overruns, and the whole puzzle has to be re-solved in seconds, not hours.

Nurse rostering is the same monster wearing a different coat. Before a hospital planner even thinks about efficiency, European law imposes hard constraints: a minimum of eleven consecutive hours of rest in every 24-hour period, limits on night work, mandatory breaks. Then come skill coverage on every shift, contracted hours, fairness between colleagues, and the individual preferences that determine whether your best people stay or quit. Many hospitals employ full-time planners who spend days producing a roster that is merely legal, let alone optimal - while the suboptimal schedule quietly bleeds money through overtime and burns out staff through unfairness.

This is why planning optimisation has stayed a maths-driven consulting business, accessible mainly to large enterprises with dedicated operational research teams. Most of the world's mid-market companies still plan in spreadsheets.

That is the gap Timefold closes.

Why your favourite chatbot cannot fix this

It is tempting to assume large language models will simply absorb this problem, as they have absorbed so many others. They will not, and the reason matters.

An LLM predicts plausible output. Plausible is precisely the wrong standard for planning. A roster that puts one nurse back on shift ten hours after her last one is illegal, however right it looks on the screen.. A route plan that sends an uncertified  technician to a gas-boiler job burns the whole trip for nothing. Planning problems are defined by hard constraints, where a near-miss carries real cost, and by search spaces so vast that pattern-matching on training data is no substitute for exploring them properly. LLMs are also non-deterministic: ask the same question twice and you may get two different answers, which is the last thing anyone wants from the system that decides whether Saturday’s night shift is covered. 

This is a category difference that no model release will close. Generative AI is built to surprise you. Scheduling software must never surprise you. Timefold's solver is deterministic: the same input produces the same output, every time, with a transparent score that shows which constraints were satisfied and which trade-offs were made, and why. When a planner asks "why is Maria on the Tuesday shift?", there is an answer.

Far from making LLMs redundant, this is the reason demand for them keeps rising. Agents that act in the physical world need a tool that computes feasible plans, much as they need a calculator for arithmetic. Timefold's API-first design makes it exactly that tool, and the platform's copilot already lets planners interrogate schedules in natural language. So AI adoption pulls demand toward Timefold’s API rather than away from them. 

Twenty years of compounding, hiding in an open-source repository

Timefold is led by a highly experienced team that has been tackling exactly these kinds of optimisation problems for two decades. Geoffrey De Smet started writing OptaPlanner, an open-source constraint solver, in 2006 and led its development at Red Hat for over a decade, during which it was deployed in more than 5,000 organisations. In late 2022, he and serial entrepreneur Maarten Vandenbroucke forked the codebase to found Timefold, carrying nearly twenty years  of algorithmic refinement into a company built to commercialise it properly. The open-source solver is downloaded more than 700,000 times a month, and its users include some of the world's largest technology companies and public institutions.

The insight that turns a great solver into a great company is Timefold's recognition that every scheduling problem is a unique combination of the same set of constraints. Rather than treating each customer as a bespoke consulting engagement, Timefold has productised the domain into standardised, configurable APIs - Field Service Routing, Employee Shift Scheduling, Pickup & Delivery Routing, and Task Scheduling - each covering the overwhelming majority of real-world requirements out of the box. Integration is deliberately boring: JSON in, JSON out, over REST, from whatever software stack a customer already runs. What used to be a multi-year consulting project becomes an API call.

Customers can embed the engine into their own products - software vendors in field service and workforce management increasingly do - or consume it directly to solve their own planning problems. Either way, the results are tangible: customers include ADP, CBRE, Deutsche Bahn, Lufthansa, NEC and IBM; the platform has generated over a million schedules, eliminated more than 30 million hours of overtime, and removed over 10 billion kilometres of unnecessary travel. Productivity gains of up to 25 per cent are the kind of number CFOs can check against their own numbers, and that value is now showing up in Timefold’s: annual recurring revenue grew fourfold in 2025.

At KOMPAS, we back enabling technologies that make physical industries - manufacturing, logistics, infrastructure, and the built environment - more productive, more resilient, and more sustainable. Timefold is a near-perfect expression of that thesis.

The timing is deliberate on our part. Skilled labour is the binding constraint across the sectors we invest in. There are too few technicians, nurses, or drivers, and the supply is shrinking.. The only durable answer is to get more from the people you have, and scheduling is where that leverage lives. At the same time, ten billion kilometres of avoided travel translates straight into fuel saved and emissions cut, which makes optimisation one of the rare technologies where efficiency and decarbonisation are the same feature. The open-core model gives Timefold something rare in enterprise software: a global developer community that works as both moat and pipeline, in a market Timefold sizes at roughly $50 billion.

Above all, we invested in the team. Geoffrey De Smet is one of a handful of people worldwide who have spent twenty years on this exact problem, and Maarten Vandenbroucke has built and exited SaaS companies before. Together they have assembled the core OptaPlanner team and pointed it at a commercial opportunity that is finally ready for them.

The Series A will fund Timefold's expansion into the US and the continued build-out of its platform. We are delighted to partner with Maarten, Geoffrey and the whole team in Ghent. When scheduling works, everything works.

To learn more, visit: www.timefold.ai

***
Written by Sebastian Peck, Partner at KOMPAS VC.