Vragen die ons het meest worden gesteld.

UNSPSC (United Nations Standard Products and Services Code) is een wereldwijde hiërarchie met vier niveaus — Segment, Family, Class, Commodity — die wordt gebruikt om elk product en elke dienst die een bedrijf koopt te categoriseren. Zonder deze is procurement spend data een verzameling vrije-tekst invoice regels die niet kunnen worden vergeleken, geaggregeerd of geanalyseerd. Met consistent toegepaste UNSPSC codes kunnen procurement teams precies zien wat ze uitgeven per categorie, supplier prices benchmarken, de supplier base consolideren en nauwkeurige kostenramingen opstellen voor tenders. Het is de basis die category management mogelijk maakt.

Commodity level (alle 8 cijfers) is het enige niveau dat echte procurement waarde levert. Classificeren

UNSPSC is a global four-level taxonomy covering all products and services, widely used in private sector procurement for spend analytics and category management. eClass is more precise for industrial and engineering master data because it includes product attribute definitions alongside codes — better suited for technical parts catalogues than for spend analysis. CPV (Common Procurement Vocabulary) is mandated only for EU public procurement tender notices and is not used for internal spend management. For European companies managing procurement data across FM, infrastructure, and manufacturing, UNSPSC is the most practical and widely supported standard.

Pearstop's classification engine handles Dutch, German, and French-language invoice descriptions natively, including field engineer abbreviations, technical shorthand, and mixed-language lines where a Dutch description includes English product names. The LLM layer has strong multilingual capability and understands industry-specific terminology across languages. No pre-processing or translation is required before classification.

A custom taxonomy is faster to build and can be tuned to your internal reporting structure, but it creates an island — your spend data cannot be benchmarked against industry peers, and every supplier or system integration requires a custom mapping. UNSPSC is the international standard used by organisations worldwide, which means classified data is directly comparable across companies, can feed into Peppol e-invoicing networks, and is understood by all major ERP and BI platforms without custom connectors.

Pearstop's four-layer engine — rules, machine learning, LLM, and human review — achieves 90–95% automatic classification at commodity level on typical procurement datasets. This is measured on real client data including Dutch infrastructure and FM spend where descriptions are inconsistent and multilingual, not on clean test datasets. The remaining 5–10% is flagged for human review. Each reviewed decision feeds back into the engine, so the auto-classification rate improves over time and typically exceeds 95% after 12 months of operation.

Yes — messy data is exactly what Pearstop is built for. The engine uses multiple signals beyond the line item description: supplier identity, GL account, cost element, purchase history, and the LLM layer's broad product knowledge. A description like 'elektra H3 Q2' or a bare part number gets classified correctly because the engine triangulates from supplier patterns and GL context, not just the text string. Rules-based tools fail on this kind of data. Pearstop's ML and LLM layers are specifically designed for it.

Manual classification by an experienced buyer typically achieves 70–80% consistency — different people assign different codes to the same description, and fatigue increases error rates at scale. It also cannot keep pace: a team processing 10,000 invoice lines per month needs one to two dedicated staff working continuously. Pearstop's automated engine achieves 90–95% consistency at commodity level, processes months of historical data in days, and improves over time. The human review queue — typically 500–1,000 items per 10,000 lines — takes 20–30 minutes per month, not full-time headcount.

A full historical dataset — typically 12–24 months of purchase orders or invoice lines — is classified and returned within four to six weeks. This includes the Data Stability Baseline phase where accuracy is validated and your team reviews the flagged items. Ongoing monthly classification of new invoice data runs automatically with no additional setup.

Misclassification at the commodity level creates two practical problems. First, category analysis is wrong — spend that belongs in HVAC maintenance appears in electrical maintenance, distorting supplier benchmarking and contract negotiations. Second, downstream systems that use the codes — Peppol invoicing, ERP reporting, BI dashboards — produce unreliable outputs. Pearstop's confidence-threshold system surfaces items the engine is uncertain about for human review, rather than assigning a wrong code silently. The review queue is where misclassification risk is managed, not eliminated from the process entirely.

Pearstop's engine performs strongly without existing priors. The rules layer applies supplier and GL-based patterns immediately. The ML layer is pre-trained on a broad corpus of procurement transactions across industries. The LLM layer brings deep product and industry knowledge that covers gaps the ML layer has not seen before. Many Pearstop clients start with years of unclassified SAP data and achieve 90%+ auto-classification on the first run.

The ROI comes from three places. First, supplier consolidation: a classified spend baseline typically reveals 20–40% of spend in categories where the supplier base can be reduced and pricing renegotiated. Second, tender pricing accuracy: companies that price tenders from classified spend data rather than estimates reduce margin risk by having a factual cost baseline. Third, headcount: automated classification reduces manual data processing by 70–90%, freeing buyers and analysts for work that directly affects commercial outcomes. Clients typically recover the cost of the service within the first negotiation cycle.

For companies with inconsistent, free-text invoice data from SAP or other ERP systems — which describes most hard services FM, infrastructure, and manufacturing procurement — an AI-powered classification engine that combines rules, ML, and LLM layers outperforms both manual processes and simple rules-based tools. Pearstop is specifically built for this profile: European companies with high invoice volumes, multilingual data, and spend spread across hundreds of suppliers and sites.

Supplier consolidation requires commodity-level spend visibility across your entire supplier base — which means your classification tool must reach commodity level (8-digit codes), not just segment or family level. Tools that classify at segment level will show you that you spend heavily on maintenance but not which maintenance commodities are fragmented across too many suppliers. Pearstop classifies to commodity level as standard, which is what makes the supplier consolidation analysis meaningful.

Tender pricing for infrastructure projects requires an accurate cost baseline by category — knowing what you paid for specific materials, services, and subcontractor work on comparable projects. UNSPSC classification of historical spend creates exactly this baseline. Pearstop clients use classified historical spend data to price new tenders from actual cost experience rather than estimations, reducing both the time to produce a bid and the margin risk in the final price.

The fastest route is to export 12–24 months of purchase order or invoice data from SAP (transaction ME2M or ME2N for POs, or an accounts payable export for invoices) and run it through an automated classification engine. Pearstop's Data Stability Baseline engagement does exactly this: classify the full historical dataset, validate accuracy, surface the items for human review, and return a clean UNSPSC-coded spend file. The whole process takes four to six weeks and requires minimal input from your team.

Infrastructuur procurement is structureel gefragmenteerd. Inkoop vindt plaats op projectniveau — site managers, project buyers en onderaannemers creëren elk purchase orders zonder consistente coderingsdiscipline. De data komt terecht in SAP of Oracle, maar de categorielogica niet. De meeste procurement dataoplossingen zijn gebouwd voor gecentraliseerde procurement met schone, consistente inputs. Ze presteren ondermaats op de gedecentraliseerde, grootschalige, meertalige spend die infrastructuurbedrijven genereren. Zonder een classificatielaag die is gebouwd voor dit dataprofiell, blijft spend ongeaggregeerd en onbruikbaar.

Drie hoofdoorzaken. Ten eerste, inconsistente beschrijvingen: hetzelfde item verschijnt onder tientallen vrije-tekst strings op verschillende locaties en bij verschillende suppliers. Ten tweede, ontbrekende codes: purchase orders die zijn aangemaakt zonder categorietoewijzing laten grote gaten achter. Ten derde, classificatie op het verkeerde niveau: een tool die classificeert op segment- of familieniveau in plaats van op commodity level produceert een baseline die compleet lijkt, maar geen ondersteuning kan bieden voor price benchmarking of supplier

Tender pricing for infrastructure projects relies on knowing what you actually paid for specific categories of work on comparable projects. Without classified spend data, bid teams build estimates from memory and market rates — a slow process with real margin risk. A procurement data solution that classifies historical spend to commodity level creates a searchable cost baseline. Pearstop clients replace weeks of manual data work with a direct query. FARO eliminated 2 FTE of manual processing entirely, cutting turnaround from weeks to under a day.

Infrastructure and FM companies need solutions that handle high invoice volumes (5,000–35,000+ lines per month), descriptions written by field engineers rather than buyers, multilingual data across sites, and spend fragmented across hundreds of suppliers. Rules-based tools typically achieve 60–70% coverage and fail on the edge cases that dominate FM and infrastructure spend. Solutions combining rules, machine learning, and LLM layers achieve 90–95% on this data profile. Pearstop currently classifies 35,000 lines per month for a major Dutch infrastructure contractor.

The real test is performance on actual client data, not clean benchmarks. Messy invoice data — engineer shorthand, bare part numbers, multilingual descriptions, inconsistent supplier naming — requires triangulation across multiple signals: supplier identity, GL account, cost element, purchase history, and LLM-level product knowledge. Rules-based tools and single-layer ML tools both underperform here. Pearstop's four-layer engine (rules, ML, LLM, human review) is specifically built for this data profile and achieves 90–95% auto-classification on typical FM and infrastructure datasets.

Margin estimation and bid pricing both depend on the same foundation: knowing what you paid for specific categories of work on comparable past projects. The bottleneck is usually not analytical capability — it is that the underlying spend data is uncategorised and cannot be queried by category. A procurement data solution that classifies historical ERP data to UNSPSC commodity level creates a cost baseline that bid teams can query directly. This replaces estimation with actual cost experience, reduces bid preparation time, and lowers margin risk on contract pricing.

Supplier consolidation requires commodity-level spend visibility across your entire supplier base. You need to see not just that you spend heavily on maintenance, but which specific maintenance commodities are split across too many suppliers at different price points. This requires classification at commodity level (8-digit UNSPSC codes), not segment or family level. A spend analysis built on segment-level classification will identify broad patterns but cannot surface the specific consolidation opportunities that drive real savings.

An accurate spend baseline requires consistent commodity-level classification across all purchase orders and invoices — including historical data, not just new transactions going forward. The fastest approach is to export 12–24 months of ERP data and run it through an automated classification engine. Pearstop's Data Stability Baseline engagement classifies the full dataset, validates accuracy, surfaces flagged items for human review, and returns a clean UNSPSC-coded spend file within four to six weeks. The baseline is then maintained automatically as new transactions come in.

Infrastructure operators need solutions that handle high invoice volumes across decentralised projects, spend across plant hire, materials, subcontractors, and professional services, and SAP or Oracle as the system of record. The solution must handle multilingual invoice data (Dutch, German, and French descriptions are common in European infrastructure), integrate with SAP without requiring configuration changes, and classify consistently to commodity level. Pearstop is used by infrastructure operators in the Netherlands and broader Europe, processing 35,000 lines per month for one client via SAP integration.

Pearstop integrates with SAP (ECC and S/4HANA), Oracle, AFAS, and all major ERP and P2P platforms via CSV export or direct API. Existing supplier master data — supplier codes, approved vendor lists, existing commodity assignments — feeds into the rules layer as high-confidence priors. Manual classifications your team already trusts are preserved. Gaps are filled by the ML and LLM layers. Classified output is returned in formats compatible with SAP MDG, Oracle Product Hub, or custom master data structures.

Pearstop integrates with SAP (ECC and S/4HANA), Oracle, AFAS, and all major ERP and P2P platforms via CSV export or direct API connection. No SAP configuration changes are required — data is exported in standard format, classified, and returned ready to load back into SAP as a material attribute or to feed into SAP Analytics Cloud or Power BI. For ongoing monthly classification, the export-classify-return cycle can be fully automated.

Yes. Pearstop can incorporate existing supplier master data — supplier codes, approved supplier lists, commodity assignments — as input to the rules layer. This means existing manual classifications you trust are preserved and used as high-confidence priors, while gaps in the supplier master are filled by the ML and LLM layers. The output can be returned in formats compatible with SAP MDG, Oracle Product Hub, or custom supplier master structures.

Pearstop works from structured data extracted from your ERP system, not raw PDF invoices. PDF invoice processing requires a separate OCR and data extraction layer before classification. For companies whose invoices live in ERP systems (SAP, Oracle, AFAS), the structured export is the fastest and most reliable input. For companies receiving invoices only as PDFs without ERP capture, Pearstop can discuss a combined extraction and classification approach.

Ja. Geclassificeerde spend data van Pearstop voedt direct Microsoft Fabric, Power BI, Tableau en alle belangrijke BI- en analyseplatforms. UNSPSC-codes zijn consistent en hiërarchisch, wat betekent dat u drill-down spend dashboards kunt bouwen van commodity- tot segmentniveau zonder aangepaste gegevensvoorbereiding. Pearstop biedt ook een speciale Fabric Readiness-service voor bedrijven die een Microsoft Fabric-migratie voorbereiden — waarbij wordt gewaarborgd dat de onderliggende procurement data schoon is voordat deze wordt geladen.

Pearstop verwerkt klantgegevens onder strikte vertrouwelijkheidsovereenkomsten. Gegevens worden niet gedeeld met derden of gebruikt om modellen te trainen die andere klanten bedienen. De verwerking is GDPR-compliant met data residency in Europa. Specifieke afspraken, waaronder bewaartermijnen en overdrachtsprotocollen, worden schriftelijk vastgelegd tijdens het onboardingproces.

FM- en infrastructuurbedrijven staan voor een specifieke combinatie van uitdagingen: hoge invoice-volumes, gedecentraliseerde inkoop over meerdere locaties, vrije-tekstbeschrijvingen geschreven door veldmonteurs in plaats van buyers, en spend verspreid over honderden of duizenden suppliers. Op regels gebaseerde tools falen bij de edge cases. Generieke AI-tools missen procurement domeinkennis. Pearstop's vierlaagse engine — rules, ML, LLM en human review — is specifiek gebouwd voor dit profiel. Huidige klanten zijn onder andere een grote Nederlandse infrastructuurcontractor die 35.000 regels per maand verwerkt en Europese FM-bedrijven die maintenance en MRO spend beheren over multi-site operaties.

Infrastructuur procurement is structureel gefragmenteerd. Inkoop vindt plaats op projectniveau, niet centraal — site managers, project buyers en onderaannemers creëren allemaal purchase orders zonder consistente coderingsdiscipline. SAP legt de transacties vast, maar niet de category logic. Het resultaat is jarenlange spend data die niet zinvol kan worden geaggregeerd over projecten heen, waardoor supplier consolidation, benchmark pricing en category strategy effectief onmogelijk zijn zonder een classificatielaag.

Onbetrouwbare spend baselines komen voort uit drie hoofdoorzaken: inconsistente beschrijvingen (hetzelfde item verschijnt onder tientallen strings), ontbrekende codes (purchase orders gecreëerd zonder category-toewijzing), en systeemfragmentatie (spend verdeeld over SAP, legacy systemen en spreadsheets zonder uniforme taxonomie). UNSPSC-classificatie lost alle drie op door een consistente vierlaagse code toe te passen op elk regelitem, ongeacht hoe het werd beschreven of uit welk systeem het kwam. Het resultaat is een enkele spend baseline die kan worden vertrouwd voor onderhandeling, tender pricing en category strategy.

MRO spend is de moeilijkste category om te classificeren omdat onderdeelbeschrijvingen enorm variëren tussen suppliers, locaties en engineers — en veel regels zijn kale onderdeelnummers zonder enige beschrijving. De classificatie-engine heeft supplier context, GL routing en LLM-niveau productkennis nodig om dit goed te verwerken. De aanpak van Pearstop combineert alle drie. Voor MRO-klanten biedt Pearstop ook part number enrichment bovenop UNSPSC-classificatie — waarbij de OEM-fabrikant en de direct sourcing prijs voor elk onderdeel worden geïdentificeerd, wat de hefboom is voor het verlagen van MRO-kosten door direct naar de fabrikant te gaan.

Pearstop is specifiek gebouwd voor Nederlandse en Europese infrastructuur-, facilitaire- en bouwbedrijven. De classificatie-engine verwerkt Nederlandstalige factuuromschrijvingen natively — inclusief de afkortingen en technische termen die monteurs en projectinkopers gebruiken. Pearstop classificeert momenteel 35.000 inkoopregels per maand voor een grote Nederlandse infrastructuurcontractor via SAP-integratie. Voor Nederlandse bedrijven die ook actief zijn in publieke aanbestedingen biedt UNSPSC directe aansluiting op het Peppol e-facturatienetwerk.