Request audit
Independent catalogue analysis · May 2026

Marks & Spencer:
a £1.8M monthly
data problem

An independent ontology-based analysis of marksandspencer.com — identifying 10 structural issues across taxonomy, sub-brand consistency, AI-generated content, and the food/clothing category boundary.

⚠️

Critical finding: 80% of M&S product descriptions are AI-generated. When the underlying product data is structurally inconsistent, AI generation amplifies those inconsistencies across thousands of listings simultaneously.

View all 10 issues Request full catalogue audit
Catalogue health score
36
/100
Needs work — urgent
SHACL violations34
Sub-brand conflicts12
Category redundancy76%
AI-amplified errors3,200+
Monthly revenue at risk
£1.8M
Across full catalogue
Online sales target
50%
Of clothing by 2028 — data quality is the blocker
AI-generated descriptions
80%
Amplifying structural data errors at scale
Tech investment FY26
£225M
Without clean data, infrastructure investment underperforms
🤖

The AI description amplification risk: M&S uses AI to generate 80% of product descriptions — a genuine competitive advantage for speed and scale. But AI generation is only as good as the structured data it draws from. When product type attributes are inconsistent, category assignments are ambiguous, or sub-brand schemas conflict, the AI generates plausible-sounding but inaccurate descriptions that propagate those errors across thousands of listings simultaneously. Clean catalogue ontology is the foundation that makes AI-generated content accurate — not just fast.

Sub-brand exposure

Revenue at risk by sub-brand

M&S Collection
£420k
est. per month
Autograph
£310k
est. per month
Per Una
£265k
est. per month
Goodmove
£195k
est. per month
M&S Food
£240k
est. per month
Jaeger
£165k
est. per month
Rosie
£125k
est. per month
eBay Resale
£75k
new — unvalidated feed
Findings

10 priority issues,
ranked by revenue impact

Key contact

Why David Clark matters

Direct connection
David Clark — Director of Online, Omnichannel & Partnerships, M&S

David Clark joined M&S in 2026 after four years at Frasers Group, where he led digital platform transformation including the group-wide commercetools migration. He has direct, personal experience of exactly the catalogue data quality problems described in this report — he lived with them at Sports Direct and Flannels. He will read this analysis with immediate recognition. He is the ideal first contact for this conversation at M&S.

M&S is investing £200–250 million in digital infrastructure in FY2025-2026, targeting 50% online sales share in clothing by 2028. Every pound of that investment underperforms if the underlying product data is structurally inconsistent. Catalogue quality is not a cost — it is a multiplier on every other digital investment M&S is making.

Ongoing solution

Catalogue governance for
AI-first retail

For a retailer using AI to generate 80% of product descriptions, catalogue governance is not optional — it is the quality control layer that makes AI generation accurate. CatalogueScore validates the structured data before it reaches the AI layer, ensuring that what the AI generates is grounded in consistent, complete product data.

— 01
Pre-AI validation gate
Before any product data is passed to the AI description generator, SHACL validation checks category assignment, sub-brand schema, and required attributes. Products with structural errors are flagged for human review before AI amplifies those errors into live descriptions.
— 02
Sub-brand schema enforcement
Eight active sub-brands (Collection, Autograph, Per Una, Goodmove, Jaeger, Rosie, Blue Harbour, Boutique) each require different attribute schemas. The ontology engine enforces brand-appropriate rules at ingestion — Autograph requires provenance and fabric composition; Goodmove requires performance attributes.
— 03
Food/clothing boundary
M&S runs food and clothing in a single catalogue with fundamentally different attribute requirements. SHACL product type detection automatically routes food products to food schemas and clothing to clothing schemas — preventing cross-category attribute contamination.
— 04
eBay resale feed validation
The new eBay resale partnership introduces third-party pre-loved product data with uncontrolled quality. The ontology engine validates every incoming resale item against M&S canonical product names and categories — flagging mismatches before they pollute the main catalogue.
— 05
Sparks personalisation accuracy
44% of online sales are linked to the Sparks platform. Personalisation accuracy depends entirely on consistent product attributes — if the same cardigan has three different "colour" values across variants, Sparks cannot reliably recommend "more like this." Attribute normalisation directly improves Sparks revenue.
— 06
Monthly health score report
The catalogue health score is recalculated monthly. The trend line is a reportable metric for the CDO and digital leadership — showing whether the £225M technology investment is translating into improved data quality and measurable revenue recovery.
Next step

Run the full analysis on your live catalogue

All we need is your Google Merchant Center product feed. No custom data pull. No IT involvement. Preliminary findings within 24 hours. NDA available on request.

Get in touch directly or visit