Data cleansing vs data enrichment

The two operations target different problems, which is why doing one does not accomplish the other.

Data cleansing is corrective. It works on information already in the record and fixes it:

• Deduplication — three supplier feeds contributed records for the same circuit breaker under three different SKU IDs. Cleansing merges them into one authoritative record.
• Standardization — one supplier wrote "15A," another wrote "15 Amp," a third wrote "15 AMP (AC)." Cleansing picks a format and applies it consistently across the catalog.
• Error correction — a transposed digit in a part number, a wrong voltage pulled from a bad PDF parse, a weight listed in grams in a metric feed and assumed to be ounces in the import. Cleansing catches and corrects the mismatch.
• Conflict resolution — two source records for the same industrial valve list different shipping weights. Cleansing identifies the conflict and establishes which value is authoritative.
• Format normalization — date formats, unit codes, case conventions, and leading zeros that vary by supplier or by import batch.

None of these operations add information. They fix what is already there.

Data enrichment is additive. It works on the gaps — information the record never contained:

• Missing structured attributes — the conduit fitting spec sheet was scanned, not structured; enrichment extracts material, thread size, pressure rating, and compatible standards into named fields.
• Descriptions written for a buyer — not a reformat of the manufacturer's copy, but content that reflects how an MRO procurement manager or electrical contractor actually evaluates the product.
• Granular taxonomy — moving a product from "Electrical" to "Electrical > Wiring Devices > Connectors > Conduit Fittings > Liquid-Tight, 1/2 in."
• Compliance and certification data — UL listing numbers, RoHS status, ANSI ratings, country of origin for tariff classification.
• Compatibility and application context — which panel brands accept this breaker, which trade applications call for this fitting, which industry verticals require that certification.
• Media — technical drawings, installation guides, spec PDFs, and photos beyond the manufacturer's single hero image.

The simplest way to keep them distinct: cleansing asks is what we have correct? Enrichment asks do we have everything a buyer needs? Both questions have to be answered, and they are answered by different work.

Order matters more than most teams expect. The right sequence is cleanse first, enrich second — and the reason is compounding.

If you enrich before cleansing, you are building on an unstable foundation. Consider a distributor with 40,000 SKUs and three supplier feeds that each contribute a version of the same 8,000 overlapping products. An enrichment pass on that raw data produces three versions of each enriched record — each slightly different, each looking authoritative, each competing with the others in search. The cleanup afterward is harder than it would have been before enrichment, because the records now look finished. Errors hide under good content.

The reverse mistake is more common and less obvious: cleansing everything and declaring the project complete. A catalog that comes out of a cleanup project is internally consistent and error-free. It is also exactly as thin as it was going in. A conduit fitting record with no errors and no missing GTINs still fails a marketplace listing check if it lacks the attributes that channel requires. It still ranks below competitors in search if it carries five attributes instead of twenty-five. It still fails to convert if the description says nothing a contractor would recognize.

Clean is the prerequisite. Complete is the goal.

There is a third step that most teams miss: keeping both running. Catalogs drift. New SKUs arrive from suppliers with thin, inconsistent data. Channels add required fields. Product lines get updated and the upstream changes do not always propagate cleanly. A one-time cleanse-and-enrich project has a half-life — quality starts decaying the week the project closes. The teams that stay ahead treat cleansing and enrichment as a continuous loop, not a point-in-time initiative.

Mistaking cleansing for a readiness milestone. After a cleanup project, it is tempting to treat "the data is clean" as meaning "the catalog is ready." Channel readiness and marketplace compliance depend on completeness, not just accuracy. A spotless record with five attributes still fails a Home Depot or Grainger data quality check that requires eighteen. Cleanliness is table stakes; completeness is what gets scored.

Enriching directly from the supplier, and only from the supplier. The most common enrichment approach pulls the manufacturer's copy and reformats it — longer title, cleaner spec table, smoothed description. The output looks different. It is not. Every other distributor selling the same SKU ran the same process on the same source. The content describes what the product is; it does not speak to the buyer who is deciding whether to buy it from you specifically.

Real enrichment in B2B starts from a second input alongside the supplier data: buyer signals. What terms do procurement managers search when specifying this category? What attributes do they filter on in your catalog? What compliance questions come up in the sales cycle? Enrichment that incorporates those signals produces content no one else has, because no one else built it for that buyer in that context.

Not knowing which problem the catalog actually has. Teams audit completeness when the problem is accuracy, and fix formatting when the problem is missing attributes. A quick diagnostic matters before committing to a scope: run an attribute fill-rate report (what percentage of SKUs have a value for each required field?) alongside a duplicate and error scan. The two reports reveal different problems and point to different work.

Scoping it as a project rather than a function. A one-time enrichment sprint raises quality for the SKUs in scope on the day it runs. New products onboarded the next month arrive thin. Existing products get updated by a supplier and the change arrives inconsistently formatted. The catalog that looked complete in January is visibly patchy by August. Cleansing and enrichment done at intake — as a standing function that every new SKU passes through — compound rather than decay.

Not all enrichment produces the same result. The gap between surface-level enrichment and genuine completeness is what determines whether a SKU gets found, chosen, and bought — or just stored.

Surface-level enrichment reformats what already exists. It pulls the manufacturer's spec table and maps it to your attribute schema. It takes the product title "3/4 in. Conduit Fitting" and expands it to "3/4 in. Liquid-Tight Straight Conduit Connector, Zinc Die Cast." This is real work and it matters. It is also the floor, not the ceiling.

The ceiling is enrichment that adds what no supplier PDF contains: the attributes a buyer uses to decide, not just to describe. In electrical distribution, that might mean the wire size range a connector accepts, not just its trade size — because the electrician specifying it needs to know if it fits the cable gauge they are already running. In safety equipment, it might mean the industry verticals where a particular ANSI rating is required — because the procurement buyer searching "required for construction" needs to know which SKUs qualify before they ever look at the part number.

This is the distinction between "what is this product" and "what does this buyer need to know to choose this product." The first comes from the supplier. The second requires knowing the buyer — their search behavior, their decision criteria, their vocabulary — and building the content around those signals rather than around the source document.

Anglera's approach is to run both layers: cleanse and normalize first, then enrich each SKU against buyer signals specific to that category and channel, and write the result back to the source of truth so every downstream system — PIM, marketplace feed, e-commerce platform, AI shopping engine — draws from the same complete record. The work happens upstream and persists everywhere, rather than being redone at each exit point.