Blog
Retailers aren’t failing to invest in AI. They’re investing in AI before their retail data infrastructure is ready for it. According to Gartner, 63% of organisations either do not have or are unsure if they have the right data management practices in place for AI.
Yet the platforms keep getting bought, the pilots keep getting launched, and the outcomes still fall short.
The problem isn’t that retail data is messy; it’s that most retail data infrastructures were never designed for machine-led decision making in the first place.
AI models inherit every inconsistency, duplication, and structural weakness in the underlying retail data, then scale those flaws across pricing, assortment, and forecasting decisions.
In retail, where product attribution, pricing, assortment, and competitor data are fragmented across systems, that amplification becomes commercially expensive.
Successful AI adoption is less about the sophistication of the model and more about the commercial architecture of the retail data underneath it.
“Garbage in, garbage out” has been retail’s data warning for years. AI makes it an entirely different kind of threat.
In a traditional reporting environment, bad data produces a bad dashboard. A merchandiser notices the numbers look off. They flag it. Someone fixes it. The damage stays contained.
In retail AI, bad data trains models to learn the wrong patterns, then applies them across thousands of pricing decisions, assortment calls, and demand forecasts, without ever flagging that something is wrong.
By the time the issue becomes visible, the decisions have already been actioned at scale.
That’s why data scientists report spending 80% of their time preparing, cleaning and structuring data before AI models can generate usable insights.
In retail, where data is seasonal, fragmented, and structurally inconsistent, the challenge is even greater.
Every industry has data problems. Retail is structural, and AI amplifies it quickly.
A single product exists across hundreds of SKUs, attributes, and naming conventions, described differently by every supplier feed, competitor site, and internal system that touches it.
A price point tells you almost nothing without knowing whether it sits within a markdown cycle, a promotional window, or a full-price strategy. A trend signal might be real momentum or isolated market noise, and acting too late can be commercially worse than not acting at all.
Generic AI tools and large language models (LLMs) weren’t built for the complexity of retail merchandising, pricing, and assortment analysis. They were built on broad data and trained to find patterns in it. Bolt them onto a retailer’s fragmented data estate, and you don’t get retail intelligence. You get faster outputs – but not better retail decisions.
The 2025 Data Landscape Retail & CPG Survey found that more than a third of organisations regularly receive inconsistent or incorrect answers to basic data questions – a direct result of fragmented definitions and siloed models. Not edge cases but standard operating conditions.
No amount of AI investment repairs this while the foundations remain broken. It only makes the cracks more visible, faster.
The phrase gets used loosely. Here’s what it requires in practice.
First, the data must be structured before it ever reaches a model. A product like a “midi dress” should resolve consistently across your own assortment, competitor products, supplier feeds, and retail analytic platforms. That only happens when products have been cleaned, matched, enriched, and normalised at scale.
Then there’s context. A competitor’s price in isolation means very little without understanding the surrounding markdown cadence, seasonal trajectory, promotional activity, and assortment movement driving it.
Finally, explainability must be built into every AI recommendation so merchandising and buying teams can validate the retail signals behind each decision. Because if merchandising and buying teams cannot trace a recommendation back to the underlying retail signals informing it, adoption breaks down quickly.
These are not advanced capabilities. They are the baseline requirements for AI that teams actually trust and use.
Agentic AI is the next shift in merchandising: systems that don’t just surface recommendations but also autonomously cleanse data, run scenario analyses, and adapt to changing retail conditions in real time.
McKinsey identifies this as the capability that could finally automate up to 60% of the low-value tasks currently consuming merchandisers’ time. But while the operating model is changing, the underlying dependency remains the same.
When AI systems begin making continuous decisions across pricing, assortment, and promotions without human validation at every step, the quality of the underlying retail data becomes exponentially more important.
Every inconsistency, duplication, or structural weakness in the data layer scales downstream into commercial decision-making.
In retail, autonomy without structured intelligence isn’t acceleration. It’s an amplification of existing flaws. Strengthen the foundation, and the value of the entire system rises with it.
Most retail AI tools are built on data that was never designed for buying and merchandising decisions. Scraped feeds, unvalidated competitor pricing, and inconsistent product categories were passed to a model and presented as intelligence.
EDITED was built differently because the retail data foundation came first.
Over 17 years, EDITED has built and structured one of the retail industry’s deepest datasets for merchandising, pricing intelligence, assortment analysis, and competitor benchmarking.
That foundation enables AskEDITED to generate AI-powered retail insights that teams can trace, validate, and act on confidently.
The future of retail AI depends on structured retail intelligence – not just bigger models.
Retail AI projects don’t fail because the technology is incapable. They fail because the data underneath it was never built for the decisions being asked of it. PwC found that companies pairing increased AI use with stronger data foundations see 2x performance improvement.
Retailers succeeding with AI are not necessarily using better models – they are operating on better retail intelligence.
Speak to EDITED to see how retail-grade data intelligence makes your AI investment actually work.
