The problem

Dutch construction companies spend an average of eleven hours a week per buyer on manual comparison work. This is what that means.

What it costs in time

Eleven hours, every week, per person.

11 hrs

a week per buyer on comparison work

Remarcable / BuildMate 2025

Three buyers, 46 working weeks, 65 euro an hour. That is 148,500 euro a year of unproductive time, spent only on laying suppliers side by side. Before the cost of the wrong decision.

That decision costs money too. In an internal Skanska experiment a generic enterprise AI system produced errors of up to 8,354 euro on a single tender line. Comparing by hand in Excel offers no guarantee either. Neither does a system that generates answers.

What leaks away

The knowledge sits in a head, not in a system.

<5%

of staff see an infrastructure project of 500 to 600 million euro through from start to finish

Rockx 2023, TU Delft

Most procurement knowledge is written down nowhere. It sits with the buyer who has known for twenty years which supplier is reliable, which certificates hold up, and which quote looks good but is not.

That makes the loss structural rather than unlucky. Eight in ten Dutch contractors already face staff shortages. When that buyer leaves, the knowledge leaves too, and no system holds on to it.

Rockx interviewed Dutch practitioners with decades of experience. Several reported never having encountered a systematic risk assessment approach. Not that they had rejected one. They had never seen one.

NL Times / Euroconstruct 2025 · Rockx 2023, TU Delft

Why now

Regulation is making it urgent.

The documentation obligation already existed. Three developments are making it concrete and enforceable:

Aanbestedingswet

Every supplier decision by a municipality must be documented, consistent and defensible before the ACM.

CPR 2024/3110

From January 2026 new certification requirements apply to construction products. The obligation exists. Most organisations lack the tooling to keep up with it.

CSRD Scope 3

Rising reporting requirements for supply chain emissions demand structured supplier data.

The spreadsheet is no longer a legal answer.

Precision

Generic AI does not solve this.

8,354 euro

of error on a single tender line using a generic enterprise AI tool

Hamppi 2025, Aalto University. Case study based on an internal Skanska experiment.

In that experiment a generic enterprise AI system produced different numerical outputs for identical tender inputs. That is not an edge case. That is the current state of generative AI in procurement: the system generates an answer but cannot reproduce the calculation.

A system that generates answers is a different thing from a system that computes them. Dundir uses a mathematical optimisation engine. Every recommendation is deterministic, reproducible and carries its full reasoning trail.

What this does not prove The experiment shows that a generic system does not compute reproducibly. It does not show that our optimisation engine buys more cheaply in practice than an experienced buyer. That validation has to come from pilot data, and we do not have it yet.

Four ways in

Where you come in depends on what you need.

Open now

The problem

Eleven hours, every week, per person.

The knowledge sits in a head, not in a system.

Regulation is making it urgent.

Generic AI does not solve this.

Where you come in depends on what you need.

Become a research partner

Pilot as a design partner

Procurement scan

A 30 minute conversation