1
Rules and machine learning
Supplier-specific rules and a machine learning layer trained on your own spend handle the clear majority of lines immediately, the way your most experienced buyer would.
Procurement
You cannot negotiate what you cannot see.
Most procurement teams can state their turnover to the euro. Few can state what they actually spent it on, category by category, across every site, entity, and bundled contract. Pearstop builds that baseline, so a negotiation, a tender, or a framework review starts from a real number instead of a guess.
The Problem
One price at tender. No way to prove it later.
Ask a head of commercial to defend what a bundled contract costs, service line by service line, and the honest answer is often a guess dressed up as a number. The same gap shows up a level higher. A construction group turning over hundreds of millions across a dozen entities can state its revenue to the euro and still not know, with any precision, what it spent last year or on what, because spend sits in different systems, coded differently, entity by entity.
“We say, well how much is this going to be. And we go, it feels like it should be three percent of the contract value. But in real terms that could be six or that could be two. We don’t know.”
Head of Commercial, multi-site cleaning services contractor
- ×One blended contract price, no way to break out what each service line actually costs
- ×Framework agreements signed years ago, with no live comparison against what is actually being bought today
- ×Group-wide spend spread across entities and systems, so nobody can add it up precisely
- ×No real baseline to negotiate a supplier or price a tender against, only last year’s guess
The Problem
Three layers handle classification. Your team handles judgment.
Most lines never need a human look at them. The ones that do feed back into the system, so the review queue gets shorter, not the same size, every month.
2
LLM layer for the edge cases
Ambiguous descriptions, multilingual line items, and one-off suppliers are resolved by a large language model with broad product and industry knowledge, not guessed at.
3
Human review, then it remembers
Anything still below a confidence threshold is flagged for your team, typically an hour or so a week. Every decision feeds back into the engine.
What a real spend baseline makes possible
A baseline to negotiate and tender from
Real spend, by category, across every site and entity. Not last year’s estimate.
Bundled contracts broken out by service line
Defend a renewal or a client challenge with a cost and performance record for each service, not one blended number.
Framework compliance checked, not assumed
See where live buying has drifted from an agreement signed years ago, across the whole business, not just the categories someone happens to be watching.
You cannot negotiate what you cannot see. Savings come third.
What is procurement data quality and why does it matter?
Procurement data quality is whether spend data, across invoices, purchase orders, and supplier records, is accurate and consistent enough to add up. For infrastructure, construction, and FM organisations buying across many sites, entities, and bundled contracts, poor data quality means nobody can state a real baseline to negotiate or tender from. Pearstop automates the cleaning and classification of procurement data for companies like Strukton, processing over 35,000 lines a month, so spend is visible by category, framework compliance is checked against live buying, and group-wide spend adds up to one number instead of several conflicting ones.
Frequently asked questions
What happens if we do not have existing classification data to train from?
Most classification systems rely on historical data to learn from. Pearstop combines rule-based assignment, machine learning, and an LLM layer that draws on broad product and industry knowledge, so it performs strongly even without existing priors.
Will our buyers still be in control?
Yes. Buyers review flagged items in a dedicated queue, typically one hour per week. Every decision they make trains the system further, reducing the review queue over time until manual input approaches zero.
Can Pearstop check our buying against existing framework agreements?
Yes. Once spend is classified consistently, current buying can be compared against the framework agreements already in place, category by category, so drift shows up as it happens rather than at the next audit.
Ready to build a real spend baseline?
Book a 7-minute discovery call. We will show you exactly where your spend data is hiding the number you need.
