What is product data enrichment? A guide that starts with the supplier’s spreadsheet

What product data enrichment means

Product data enrichment is the process of taking an incomplete product record and adding what it is missing, so that each product can be found, filtered, compared and bought. On a real catalogue that comes down to four kinds of data.

• Identity. The product’s real name, its brand, its barcode number (the GTIN) and the manufacturer’s part number.
• Specifications. The measurable facts a shopper filters on: dimensions, capacity, power, material, colour.
• Categorisation. Where the product sits in your own navigation, and which products it is a part or accessory for.
• Content. Images large enough to zoom into, a title and description, and the page title and snippet that appear in search.

It is usually mentioned in the same breath as data cleansing, and the line between the two is real but thin. Cleansing fixes what is already there: one colour spelt three ways, a width in centimetres beside one in millimetres, a duplicate row. Enrichment adds what is not there: the capacity nobody filled in, the photograph the supplier never sent.

On a supplier file you need both, and in that order, because you cannot tell whether a value is missing until the values that are present have been made to mean the same thing. A width recorded as “60” in one row and “595mm” in the next is not two widths. It is one width written twice, and a third row with no width at all is the real gap.

What a supplier actually sends

Guides written by software vendors tend to picture enrichment as polishing a listing that is nearly finished. The file that lands in a buying or ecommerce team’s inbox is usually further from finished than that. These turn up in almost every supplier sheet:

• a dealer code where the product name should be;
• three dimensions in one cell, as 1850x595x655, with no unit;
• a category column in the supplier’s words, not the words in your menu;
• blank cells for the specifications shoppers filter on most;
• image links to small gallery thumbnails, or to pages that no longer exist.

Here is one row, before and after. It is the same illustrative product the homepage uses, and every change in the right-hand column is one of the kinds of work described below.

Most of that row is plain filling in. The energy rating is the instructive one. It was not wrong when the supplier wrote it down: fridges and freezers moved to a new A to G label on 1 March 2021, which did away with the A+, A++ and A+++ classes.¹ The same appliance can carry a different letter on the new scale without having changed at all, so an old rating has to be looked up again, never converted. Enrichment is full of values like that, which were true once and have quietly stopped being true.

Where the missing data comes from

Almost everything a supplier leaves out has been published somewhere: on the manufacturer’s own product page, in the datasheet or manual attached to it, and on the pages of other retailers that stock the same item. Enrichment is mostly the work of finding those pages and reading them correctly.

Finding them is the easy half. The hard half is proving that the page you found describes your product. Codes are reused across regions, a range comes in five widths with near-identical pages, and another retailer’s listing may describe last year’s model. A value copied from the wrong page is worse than a blank, because it looks finished.

It is just as easy to give up too early. On one of our recent jobs, for a percussion and pro-audio retailer, 212 products came back empty on the first pass. We put twenty of them straight back through, with no code changes and nothing retyped, and fourteen came back with a real source attached. They were never missing, just badly searched.

That is why every value added during enrichment should keep a link to the page it came from. It turns a figure nobody can check into one anyone can check in one click, and when two sources disagree it records which one was chosen.

The work, in the order it has to happen

Enrichment is a sequence rather than a checklist, because each step needs the one before it. You cannot categorise a product you have not identified, or write a description from specifications you have not filled.

1. Identify each product. Match the supplier’s code or barcode to the real product, and check that every source page shows that product. Barcodes and part numbers do most of this work; the guide to GTINs, EANs and MPNs covers how.
2. File it into your categories. Your category tree, not the supplier’s, with parts and spares kept apart from the products they fit. See mapping supplier categories.
3. Fill and split the specifications. Find the missing values, split compound cells like 1850x595x655 into their own fields, and put every measure in one unit.
4. Make the values consistent. Fold every spelling of a value onto one word, so a filter offers one option per real thing. The post on attribute values has a worked example.
5. Link the accessories. Match the filters, seals and spares a supplier lists against items you already stock, so they can be sold together.
6. Fix the images. Find the full-size original behind each thumbnail, drop logos and duplicates, and convert formats your store will not import. Google’s new minimum makes this urgent; see the 500 by 500 pixel rule.
7. Write the copy from the values. Titles, descriptions, page titles and search snippets, written from specifications already in the data rather than from imagination.
8. Review before anything goes live. Check each number against the rest of its category and each title against its own specifications, and flag whatever does not add up.

If the supplier file is the starting point, the guide to supplier data onboarding walks through the same sequence from the moment a spreadsheet arrives.

Why it matters to sales, search and returns

The case for enrichment is usually made with surveys. The most useful recent one is Syndigo’s 2025 State of Product Experience report, which asked more than 8,500 shoppers in six countries, the UK among them. 44% said they had abandoned a purchase because the product information was insufficient, and 21% had returned a product that did not match what its content led them to expect.² Syndigo sells product content software, so read the figures as a vendor’s survey. The direction is not in dispute.

The less visible cost is in navigation. A filter can only offer the values that are in your data. On one bathroom retailer’s catalogue, a single Type column held twelve spellings for eight actual types of bath, so shoppers were offered twelve filter options for eight real things. How that happens, and the fix.

Then there are the channels you do not control. Google asks for the barcode number of any product that has one, and says products without a correct GTIN may get limited visibility.³ From 31 January 2027 it will require every product image to be at least 500 by 500 pixels.⁴

And a growing share of product questions are now answered by AI assistants rather than by a page of results. They work from the same structured facts. OpenAI’s product feed specification for ChatGPT requires a brand and a factual description for every item, and takes a GTIN, a manufacturer part number and a category path as optional fields.⁵ A blank specification cell gives an assistant nothing to quote; the post on AI shopping assistants goes further.

How to tell whether it was done well

Enrichment that cannot be measured cannot be trusted, so decide how you will check it before it starts. Three measures cover most of it.

• Completeness. Does every product have the values a shopper filters on in its category? Not every column in the file: the ones that matter for that kind of product.
• Consistency. Is each value written the same way everywhere: one unit per field, one word per value, one format per identifier?
• Confidence. How much of the data is traced to a named source, and how much is unverified?

The third is the one most often left out, and the most useful. On a catalogue of 901 products we checked 874 against a named source and listed the other 27 as unchecked rather than guessing at them. Knowing what could not be verified is what makes the number at the top of a report worth reading.

A good check also warns rather than blocks. Some gaps are acceptable, and exporting a catalogue with them should be a decision made with the gaps in view. The product data quality checklist turns these three measures into checks you can run.

By hand, in a PIM, or with a service

There are three common ways to get enrichment done, and they are not mutually exclusive.

By hand, in spreadsheets

Workable for a few hundred products from a couple of suppliers. It stops scaling when every new supplier file means days of copying values out of PDFs, and it rarely records where any value came from.

In a PIM

A product information management system, such as Akeneo, Pimcore or Salsify, is where product data is stored, governed and sent out to your channels. It is very good at holding data once you have it, and at showing you what is incomplete. Many now include AI features that draft descriptions or suggest values. Their core job, though, is to hold the data you give them. Tracking down a missing capacity on a manufacturer’s datasheet, and proving it belongs to your product, is work that happens before the data arrives.

With an enrichment service

A service takes the supplier file as it arrived and returns it enriched, formatted to import into your store or PIM. That is what RefynData’s product data enrichment service does: you send the spreadsheet exactly as your supplier shipped it, and it comes back with the gaps filled, the formats fixed and a source on every value.

Whichever route you take, the tests do not change. Is it complete, is it consistent, and can you see where each value came from?