Episode 12: The Five Layers of Product Data Optimisation

Most product data work starts in the wrong place. Marketing needs descriptions for launch, so somebody writes descriptions. Or the pressure is on to be visible in AI search, so the team jumps straight to structured data.

Both are real priorities. Neither works on its own.

Product data optimisation runs in five layers, and the word layer is doing the work there. This is not a checklist you can take in any order. Attempt layer four or five before one, two and three are in place and you hit a wall, because the foundation each layer needs is not there.

Layers, not a checklist

One assumption before we start. This is an approach for a catalogue that already exists and needs reworking, not a rebuild from nothing. If we were starting a product data programme from scratch we would structure the data first and build up from there. Here we are assuming a structure exists and you are improving on top of it.

There was genuine internal debate about the ordering, particularly between relations and structured data. What is not up for debate is the direction of dependency.

Each layer needs the one beneath it. Skip a level and the work above it quietly fails.

1. Basic data

This one is what it says on the tin. Product identity, core features, product type, and the family or category the product belongs to.

The question this layer answers is how complete your product data is, and the answer lives in your attributes. Every category should carry a set of mandatory attributes. That might be five, it might be ten, depending on how complex the product is, and every one of them should be filled for every product in that category.

So the minimum is a name, a type, a category and the must-have attributes for that category, on every SKU. If your categories have been defined properly, there should be no blanks at all.

Completeness first. Everything above this layer is built out of what you capture here, which is why nothing else works when it is patchy.

2. Readability

This is the richness layer. Formatting, length, and the words a customer actually reads.

The reason it sits second rather than first is practical. Writing a good description, a feature list or an FAQ is genuinely difficult when the mandatory attributes are empty. Even writing a decent title is hard. You are describing a product you hold no structured facts about, which is how businesses end up with pages of confident, useless copy.

With layer one in place the standard is clear. The title should say what the product is in one line. The description should answer the top three buying questions. Keywords should be the terms your customers actually search for, not filler.

A good product description is the answer to the buyer's question. A bad one is noise.

3. Product relations

Variants, similar products, substitutes, accessories. This layer is about how the products in your catalogue connect to one another.

It is one of the easiest wins available, provided you know your products. It is also one of the most direct levers on average order value, because it powers customers also viewed, related accessories, and the recommendation that appears when something is out of stock.

That last one matters more than it looks. Without substitute relationships an out of stock is a dead end and a lost sale. With them it is a redirect.

Most search and merchandising tools need this level of information before they can do any of it. They will not infer relationships you have never recorded.

4. Structured data

If we were rebuilding a catalogue from scratch, this is where we would start. Category by category, identify the optimal structure: the attributes, the features, the units, consistent values and lists of values. Then go back and fill them.

Working on an existing catalogue, you can rework what you already have to optimise the structure rather than starting again.

This is the layer that makes AI work. Agentic systems scan enormous numbers of pages very quickly, and the more context and structure you give them, the more authority a page carries. Without structured data you probably will not be surfaced at all.

It is worth separating the two jobs cleanly. Readability is how you describe the product to a person. Structure is how the machine reads it.

Which is why skipping straight to this layer does not work. Layers one to three have to be complete first.

5. Product expertise

The final layer is the knowledge that sits around the product rather than in it. Compatibility. Installation. Usage conditions. The questions that come up again and again.

Think about what your best salesperson tells a customer out on the road. That is this layer. Most businesses hold an immense amount of that expertise internally and almost never surface it, because everybody is busy and marketing ends up inventing copy rather than extracting what the product experts already know.

It is also the difference between being findable and being chosen. Layer four gives AI a consistent, structured list of attributes, enough to decide your product is a possible match. Layer five explains what those attributes do and which buyer they suit, so the match becomes obvious. One is data. The other is judgement.

The sources are all internal. Product experts, product managers, the sales team, and customer service, which is a gold mine. So are your return reasons. Much of it is anecdotal and appears in no data sheet anywhere.

The mistakes we see most

Four failure patterns come up repeatedly.

Starting at layer two, because marketing needs descriptions before a product can go live, without ever defining the key attributes underneath them.

Buying an AI search tool before layers one to three are fixed. The tool struggles, and the money is already spent.

Treating relations as a one-off project. Products arrive constantly and lines get discontinued constantly, so relationships decay from the day you finish. It has to be an ongoing audit.

And confusing layer four with layer five. Four is structure: cleaning, normalising, making values consistent. Five is context: what the product is actually used for. Doing one of them well does not cover the other.

Score one category out of five

Pick a single category and score it against each layer. How complete is the basic data? How readable are the pages? How well do the products connect to one another? Is the structure consistent across the whole category? And are you surfacing the product expertise you already hold?

You can run that audit today. It takes a category, not a programme.

Most teams score around two out of five. That is not a failing, it is the normal result of building each layer only when something downstream demanded it. Knowing which layer breaks first tells you exactly where the next piece of work sits.

Listen to the full episode

Episode 12 of Product Data Weekly is available now. For more episodes and the weekly newsletter on operational issues inside product data and ecommerce teams, visit productdataweekly.com.