325,000-Item UNSPSC Classification Using Agentic AI
Overview
A global IT solutions provider with operations in the UK and EU engaged AICA to unify its product taxonomy across regions. The project involved classifying over 325,000 hardware, software, and support SKUs to UNSPSC v26.
We combined automation with human-in-the-loop quality control to deliver a consistent, high-confidence catalogue that improved sales reporting, procurement visibility, and operational efficiency.
.
Objectives
The client’s key goals were clear:
- Classify 325,000 SKUs to UNSPSC with ≥90% confidence.
- Minimise low-confidence outputs (≤80%) that would otherwise require manual rework.
- Deliver results quickly, in weeks rather than months.
- Ensure smooth integration through CSV in/CSV out with batch releases for client QA.
Implementation
Baseline Testing
- Initial sample runs achieved ~93% average confidence, highlighting enrichment needs for sparse records.
Context Restriction
- UNSPSC segments were narrowed to a relevant subset, reducing false positives and boosting precision.
Consistency Engine
- Human-verified examples seeded rules to stabilise classifications across bundles and near-duplicate products.
Iterative Batching
- The catalogue was processed in ~50k-item tranches, with rapid feedback loops and targeted tuning between runs.
Quality Controls
- Outlier detection, segment allow-lists, and regression checks ensured accuracy improved progressively.
Achievements
Coverage
- 325,000 items classified and enriched to UNSPSC v26.
Confidence
- The vast majority of records were mapped at ≥90% confidence, with <2% under 85 in early batches.
Consistency
- Commodity-level coherence improved across look-alike SKUs, bundles, and renewals.
Exceptions
- Only 5–7% of items required enrichment or triage due to sparse source data.
Throughput
- Weeks-scale delivery allowed QA without delaying the overall programme.
Business Impact
Spend Visibility
- Unified taxonomy across regions enabled global category analytics and supplier consolidation.
Sales Reporting
- Reliable segment/family/class roll-ups supported margin analysis and cross-sell insights.
Operational Efficiency
- Manual QA effort fell sharply, while onboarding of new SKUs became faster and more predictable.
Governance
- QA controls and exception packs provided a repeatable framework for ongoing data stewardship.
Words from the Client
Conclusion
By combining AI-driven automation with human oversight, AICA delivered a fully classified, high-confidence catalogue of 325,000 SKUs. The client gained faster sourcing insights, consistent reporting across regions, and significant reductions in manual rework. This case demonstrates how structured data, underpinned by UNSPSC, directly drives efficiency and measurable business value at scale.
More importantly, this project highlights how AICA’s Agentic AI changes the rules of classification. What was once a multi-month effort can now be completed in a matter of weeks, with accuracy levels exceeding 90%.
This is another testament to the power of AICA’s approach: making data classification no longer a bottleneck, but a strategic enabler of growth, governance, and operational efficiency.
We specialise in product data & services cleansing, enrichment, and comparison utilising AI and ML to detect a wide array of errors and inconsistencies in your data.
Ensuring that your data works for you. Feeding your Master Data Management system with pristine and enriched data.