What Is Entity Resolution?

Entity resolution is the process of deciding whether two or more records that look different actually describe the same real-world thing, and then linking or merging them into a single trusted representation.

Definition

So, what is entity resolution? At its core, it answers a deceptively simple question: are these records the same entity or not? In product data, the “entity” is usually a physical product, a manufacturer, a supplier, or a location. The same item can arrive in your systems a dozen different ways. One feed lists it as “Hex Bolt M8x40 Zinc,” another as “M8 x 40mm Hex Head Bolt, ZP,” and a third only carries a manufacturer part number with no description at all. Entity resolution determines that all three point to one product, even when no shared identifier exists.

It works by combining identifier logic (deterministic matching on keys like GTIN or MPN), similarity scoring (probabilistic and fuzzy matching on names, attributes, and specs), and clustering (grouping all the matched records together). The output is a set of resolved entities, each with a confidence score and a link back to every source record it absorbed. Done well, entity resolution is reversible and auditable: you can always see which inputs produced a resolved entity and unwind a bad merge. The term overlaps with record linkage and underpins the canonical, golden record that downstream systems consume.

Why entity resolution matters for product data

Without entity resolution, the same product fragments across your catalog, and every fragment carries its own price, stock level, and description. Analytics double-count. Procurement buys from three “different” suppliers who are really one. AI assistants answering “do you stock this part?” return contradictory results because the underlying records were never reconciled.

Consider an industrial distributor consolidating fifty supplier feeds. Vendor A sells a bearing as “6204-2RS,” Vendor B as “6204 2RS C3,” and a private-label sheet calls it “Deep Groove Ball Bearing 20x47x14.” Entity resolution recognizes these as one product, attaches the best attributes from each source, and produces a single sellable SKU. The same pattern repeats across MRO (overlapping fastener and fitting catalogs), CPG (the same SKU described differently by each retailer’s data pool), and furniture (variant explosions where color and size options masquerade as separate products).

The downstream payoff is large. Clean, resolved entities feed accurate confidence scoring, reliable deduplication, and complete enrichment. They also make product data legible to generative engines: an AI assistant can only cite a product confidently when there is one authoritative record to point at, not five conflicting ones. This is why resolution is foundational to a canonical product-data layer rather than a one-off cleanup task. Teams that try to do it with hand-tuned scripts usually hit a wall, which is exactly the failure mode described in why fuzzy-match scripts break at scale.

Related terms

FAQ

What is the difference between entity resolution and deduplication? +

Entity resolution is the decision step: it determines which records refer to the same real-world entity. Deduplication is the action that follows: once records are resolved into a group, you merge or collapse them into one canonical record. You cannot deduplicate reliably without resolving entities first, because removing “duplicates” you have not actually confirmed risks deleting genuinely distinct products.

Is entity resolution the same as record linkage? +

They are closely related and often used interchangeably. Record linkage is the older, broader academic term for connecting records that describe the same entity, especially across separate datasets. Entity resolution typically emphasizes the full lifecycle: matching, clustering, merging into a canonical entity, and maintaining it over time as new sources arrive. In product data, you will see both terms describe the same core capability.

Can entity resolution work without a shared ID like GTIN or MPN? +

Yes. That is precisely when it earns its keep. When a clean identifier exists, deterministic matching handles most of the work. When it is missing, malformed, or reused across products, resolution falls back to probabilistic and fuzzy matching on names, attributes, units, and specifications, scoring each candidate pair and clustering the confident matches. Most real catalogs need both approaches because identifier coverage is never complete.

How does entity resolution improve AI and search results? +

Generative engines and search systems answer best when each product maps to one authoritative record. If a product is split across several conflicting duplicates, an AI assistant cannot tell which price, spec, or availability is correct, so it either hedges or returns inconsistent answers. Entity resolution collapses those fragments into a single, well-attributed entity, giving AI a citable source and making your catalog far more legible to generative search.

Is entity resolution a one-time project or an ongoing process? +

It is ongoing. Every new supplier feed, price list, and catalog import introduces records that must be resolved against your existing entities. Treating resolution as a permanent layer, with confidence thresholds, audit trails, and reversible merges, keeps the catalog clean as it grows, rather than letting duplicates re-accumulate after each onboarding.