35k–50k
spend lines a month
Case Study
Classifying 35,000 to 50,000 procurement lines a month into UNSPSC, from zero classification history
Strukton is a major Dutch infrastructure contractor. Pearstop built this system together with their procurement department. It classifies every line of spend into UNSPSC, automatically, every month.
Infrastructure · Netherlands
Why this started
Strukton had identified a real cost-saving opportunity in procurement. To act on it, they needed higher granularity in their spend data. Their spend was already well organized. What they didn’t have was UNSPSC classification. This project started from zero classification history.
What we built
Pearstop and Strukton’s procurement team built a system with three layers. AI classifies each line automatically. A machine learning pipeline learns from the team’s own input and context, not just from a general-purpose model on its own. A human-in-the-loop review step catches anything the system isn’t confident about.
A common mistake in AI projects: reinventing the wheel. Teams start from scratch and don’t give the model enough context. Without that context, the model goes on a tangent. Strukton’s system avoids that by learning continuously from real decisions made by real buyers, not a generic prompt.
Four levels, every month
Every month, a new batch of spend comes in. Pearstop processes all of it. Each line gets four levels of UNSPSC classification: segment, family, class, and commodity.
A segment is the broadest category. A commodity is the most specific. Classifying to all four levels, not just the top one, is what makes the data usable for real decisions - benchmarking one specific product across suppliers, not just a broad category average. Read more about how UNSPSC’s four levels work.
4
UNSPSC levels per line
Weekly
buyer feedback cadence
The weekly feedback loop
Strukton’s buyers review what the AI classified. They flag what’s right and what isn’t. Every correction feeds back into the system. Pearstop and Strukton’s team meet weekly. The team gives feedback on what’s working and what isn’t. The system gets more accurate every month, not just more used.
Built and proven with a major Dutch infrastructure contractor, starting from zero classification history.
What changed
Before this project, Strukton had no UNSPSC classification at all. Now, 35,000 to 50,000 lines of spend are classified automatically every month, to all four UNSPSC levels. The cost-saving opportunity they identified at the start now has the granularity to actually act on it.
Want a case study built around your data?
We can show you what the same approach would look like for your procurement or asset data.
