Product data normalisation is the work of turning every spelling and variant of the same value into one agreed value, across the whole catalogue. Years of manual entry leave hundreds of spellings for the same thing. SKULaunch finds inconsistent values, groups the variants and cleans them up in bulk, so the data is consistent before you enrich, migrate or publish it.
The problem
Inconsistency builds up slowly: different people, different suppliers, different years. Colour has "Grey", "Gray", "GRY" and "Slate grey". Brand has three spellings of the same manufacturer. Size has "XL", "X-Large" and "Extra large". Each one looks harmless on its own product page.
Together they break the things customers use. Filters show five options that mean the same thing. Search misses products. Reports undercount brands. Nobody owns the clean-up because it feels endless, and every downstream project, from a new website to a marketplace launch, inherits the mess.
Before and after
An illustrative example from a catalogue audit at a tools and hardware distributor.
Before: Brand values in use
"Bosch", "BOSCH", "Bosch Professional", "Bosch Pro", "Robert Bosch". 2,300 products across five spellings.
After: cleaned in SKULaunch
- Brand: Bosch, with "Professional" held as a product line attribute
- All variants grouped and mapped in one action
- Similar clean-ups applied to colour, finish and material
- Mappings kept, so new data arrives clean
How it works in SKULaunch
- Find the variants. Filter an attribute to see every value in use and how often.
- Group and map. Map variants to the correct value, or to a controlled list.
- Apply in bulk. Review the mapping, then update every affected product at once.
- Prevent recurrence. Saved value mappings clean new imports automatically.
What product data normalisation covers
"Inconsistent values" is several different problems, and each needs a slightly different fix:
- Case and spelling. "BOSCH" and "Bosch", "Gray" and "Grey". Pick one form and map the rest.
- Abbreviations. "SS", "S/S" and "Stainless" become Stainless steel.
- Synonyms. "Charcoal" and "Anthracite" may be the same colour family, or may not. That is a business decision, not a find and replace.
- Units and number formats. "2m", "2000mm" and "2,0 m" are one length. See how to normalise units of measure.
- Combined values. "Bosch Professional" holds a brand and a product line. Split them into two attributes.
- Yes and no fields. "Y", "Yes", "1", "Included" and "TRUE" all mean the same thing in a battery included column.
Clean the attributes customers use first
Do not try to clean everything. Start with the attributes behind filters and search: brand, colour, material, size and type. Those deliver visible improvement on the website quickly. Leave rarely used free-text fields until last, or retire them if nobody relies on them.
A practical order for a first pass:
- Brand. Few distinct values, used everywhere, and errors are obvious.
- Colour and finish. High customer visibility, and usually the messiest.
- Size and dimensions. Needs units fixed first, then values.
- Material and type. Often needs a controlled list agreed with buyers before mapping.
Work one attribute at a time, sorted by product count. Fixing the five spellings that cover 2,000 products beats fixing fifty spellings that cover thirty.
A worked example: colour in one category
An illustrative colour clean-up across 1,800 kitchen taps from six suppliers. Filtering the attribute shows 41 distinct values. Grouping them takes four decisions:
- "Chrome", "CHR", "Polished chrome" and "Chrome plated" map to Chrome
- "Brushed SS", "Brushed stainless" and "Stainless brushed" map to Brushed stainless steel
- "Matt black", "Matte black", "Black matt" and "BLK" map to Matt black
- "Gunmetal" and "Anthracite" stay separate, because the buyer confirms they are different finishes
The 41 values become 9, the Finish filter on the website shows nine clean options, and the mappings are saved against each supplier so the next file is cleaned on arrival. The last decision is the important one. Normalisation is not about forcing values together. It is about making the decision once, deliberately, and applying it everywhere.
Common mistakes
- Losing information. Mapping "Slate grey" to "Grey" throws away the shade. Keep a colour family for filters and a colour name for the page.
- Find and replace in a spreadsheet. Replacing "Grey" also changes "Greyhound". Map whole values, not fragments.
- Normalising to a list nobody agreed. If buyers do not recognise the values, they will type their own again.
- Cleaning once. Without saved mappings, the next supplier file brings every old spelling straight back.
What changes
A consistent catalogue that every project, channel and team can build on, and that stays consistent as new data arrives.
Common questions
What is product data normalisation?
Converting the different ways a value has been written into one agreed form, so the same thing is always recorded the same way.
What is the difference between normalisation and a controlled value list?
Normalisation cleans the values you already have. A controlled list defines the values allowed from now on. You normally do both. See controlled value lists.
Can we see how bad it is first?
Yes. Filter any attribute to see every distinct value and its count.
Is the clean-up reversible?
Changes go through review, so you approve the mapping before it is applied.
Should we clean before or after a migration?
Before. Clean data migrates once; dirty data gets cleaned in a live system. See cleaning data before a PIM migration.
How do we stop inconsistent values coming back?
Save the value mappings so new imports are cleaned on the way in, and use controlled lists for the attributes behind your filters.
More on mapping and normalisation, how to audit catalogue quality, and the data behind faceted search for distributors.

.avif)