Does better product information sell more? What the evidence actually shows

Ten things that get called the same thing

Half the confusion here is that "does it work" is asked about ten different quantities, and evidence about one gets offered as evidence about another. They are not interchangeable and several of them move in opposite directions.

A vendor case study that reports engagement and a business case that promises sales are not talking about the same thing, and the step between them is the step nobody has evidenced.

The three numbers, and what each one actually is

"Sixty per cent of returns are caused by poor product information"

And its close relative, that seventy odd per cent of returns are due to inaccurate descriptions.

Both trace to consumer surveys asking whether somebody has ever returned an item because the description was wrong. That is a share of people who have ever done a thing. It is then quoted as a share of returns, which is a different quantity with a different denominator, and the conversion between them was never performed by anybody. It was restated until the restatement became the citation.

Both figures originate with vendors selling the remedy.

"One in four product records is wrong"

No primary document exists for this one. The nearest real source is the GS1 UK Data Crunch Report of October 2009, which found data inconsistent in well over eighty per cent of instances.

Note what happened in transmission. A different construct, inconsistency between two systems rather than wrongness against the product. A different population, UK grocery. A different figure, well over eighty rather than twenty five. And seventeen years old.

The lifetime value multiple

A Bain thesis that product lifetime value could double and that consumers could capture up to sixty five per cent of the new value created. Bain labels it as its own analysis, which is to say a model, and that label is accurate.

A model is a legitimate object and this is not a criticism of building one. It is a criticism of quoting one as though a measurement had occurred.

What controlled evidence actually exists

A dedicated sweep of eight evidence pools found two controlled field experiments in the entire territory. Both concern item level radio tagging, one from 2008 and one from 2013. Neither is in apparel.

Their combined finding is the useful part: the effect ranges from zero to 81.6 per cent depending on the product category. A range that wide, across two studies in categories that are not yours, cannot be used to project anything. It is evidence that the answer depends on the case, which is a real finding and a useless planning input.

Every other step in the chain from operational accuracy to money is asserted by adopters and measured by nobody.

Controlled evidence outside item-level tagging

The sweep above asked one question: does item-level tagging move sales. Widen it to product information generally and controlled evidence does exist, in categories that are not passports.

Read the column on the right before the column in the middle. Each of these is a specific treatment on a specific decision, and the honest generalisation from five of them is that product information sometimes changes behaviour measurably, not that a passport does.

Deployment scale is not a return. Decathlon tagged 100 per cent of its products by January 2019 and describes an estate of roughly 50,000 readers, and sold 1.23 billion products in 2025. Neither number is a denominator for anything. Products sold are not reads, reads are not benefit events, and a company case with no counterfactual cannot isolate causal return. The same applies to any vendor's adoption figure: scale evidences that a thing was built, not that it paid.

The most honest sentence in the literature is a refusal

CIRPASS, the European Commission funded project that studied passport costs for smaller businesses directly, reports that all potential providers were reluctant to give quantitative figures for commercial reasons, and sometimes did not know yet.

Not a market that measured and got a disappointing answer. A market that has not measured, said so to a regulator's researchers, and continues to sell against numbers it does not have.

Where numbers do exist, and why

There is one place in this whole territory where hard percentages are published.

Retail till and inventory deployments produce quantified benefits. Woolworths reports waste reduction of up to forty per cent on fresh categories and inventory productivity gains of around a fifth from two dimensional barcodes. Those figures exist because the retailer was already running a system that would notice a change, with a baseline in it, before the programme started.

Nothing on the consumer side is measured because nobody has a system that would notice. That is a fact about instrumentation and not about consumers, and it is the single best explanation for why this evidence base looks the way it does.

Two consequences follow. The measured benefits in this category accrue to operations rather than to marketing. And they accrue to the retailer, who is usually not the party being asked to pay for the product data work. The scanning numbers that do exist, with their denominators, are on what a scan actually tells you.

How to test a return figure in a meeting

Three questions, in this order. They cannot tell you whether a figure is true. They tell you whether it is a figure.

What was the comparison? Not what happened after the programme. What happened to the comparable thing that did not get the programme, over the same period. Without a comparison group there is no way to separate the effect from everything else that was happening that year, and almost nothing published in this territory has one.

What is underneath the percentage? A share of what, measured against what population. A share of surveyed people who have ever done something is not a share of transactions, and the two get swapped in this territory more often than not.

Who measured it, and what do they sell? Not a reason to dismiss a figure. A reason to ask for the method, which in this category is usually unavailable. Where we chased vendor scan and outcome figures, the sample size, the fielding period and the question wording were not obtainable.

The same discipline applied to scanning specifically, along with the four field measurements that do carry populations, is on what a scan actually tells you.

What this page will not publish

The most specific outcome claims available anywhere in this category are a re-engagement multiple and a resale value multiple, published by a vendor about its own customers. No baseline. No time window. No comparison group. They are marketing and they are not carried here, in either direction, because reporting the size of a gap requires both sides of the arithmetic to have a denominator and one side has none.

No conversion, returns or revenue figure attributable to product data appears on this page. No cost per product or per unit appears either, because no figure survived checking anywhere and that absence has its own page.

What this is not an argument for

It does not say better product information has no commercial value. Several things in this territory work reliably, get paid for and have done for years. What they have in common is set out on what a passport programme looks like when it works, and none of them is a consumer engagement story.

It does not say the effect is unmeasurable. It is measurable and it is cheap to measure. A programme rolled out to part of a range, against a matched part that does not get it, with a stated period and a stated outcome, would produce the first real number in this category. Nobody has published one.

It does not say vendors are lying. It says the numbers in circulation are the wrong shape for the argument they are offered in support of, which is a different and more common failure.

What it would actually take to answer this

If a return figure is required before this work can be funded, the cheapest honest route is to produce one on your own range. The design is not difficult and it is the reason the absence in this category is so striking.

Pick one outcome from the table above and commit to it in writing before you start, because a study that decides afterwards which metric moved has measured nothing. Split a range into two groups that are comparable on the things that drive the outcome anyway, which for apparel usually means category, price band, season and channel rather than a random split across the whole catalogue. Apply the work to one group and not the other. Run it for a period long enough to cover a normal buying cycle for that category, decided in advance. Then compare the two groups over the same period, and state the size of each group beside the result.

Four things will be tempting and each one destroys the answer. Rolling out to everything and comparing with last year, which measures the year. Choosing the metric after seeing the data. Counting engagement and reporting it as demand. And stopping early because the direction looks right.

None of that requires a vendor, a platform or a passport. It requires deciding in advance what would count as an answer. That is the part this category has not done, and there are ordinary reasons why: a controlled roll-out is awkward to run inside a live range, the results belong to the business rather than to the supplier who would have to fund the work, and nobody has an obligation to publish a disappointing one. The absence is more likely a collective action problem than a conspiracy. It is still an absence, and the first business to close it will be quoted for years.

That is a measurement problem rather than a content problem, and it is the problem ActivateMeasurement is built for: it can size that test and state the smallest effect it could honestly detect before any budget is committed.

What one publication would change

A controlled experiment in apparel, or any consumer category, with a stated method, a comparison group and a period, published by anybody. One would change this page. There are currently none.

An operator publishing outcomes against units placed on the market rather than totals. The absence of denominators is the structural defect in this whole evidence base and one publication would begin to close it.

Sources

• A sweep of eight evidence pools for controlled evidence on product data return, completed 27 August 2026
  Field research

  The load bearing source for the central absence. It located two controlled field experiments, both on item level radio tagging, dated 2008 and 2013, neither in apparel, with a stated effect range of zero to 81.6 per cent by product category.

• A UK grocery product data quality report, October 2009
  Industry research

  Cited only as the traceable origin of a widely restated figure, and for what it actually measured, which is inconsistency between systems in a named sector at a date seventeen years ago.

• Consumer survey figures on returns and product descriptions, published by vendors of product data services
  Vendor material

  Cited for their construction rather than their content: a share of consumers who have ever returned an item, restated as a share of returns.

• A European Commission funded study on passport service provision
  Institutional research

  Cited for one proposition: that potential providers were reluctant to supply quantitative figures for commercial reasons and in some cases did not have them.

• A consultancy thesis on product lifetime value
  Consultancy publication

  Cited as a model, on the strength of its own authors' labelling of it as analysis. No figure from it is carried.

• Retail operational figures from till and inventory deployments
  Operator reporting

  Cited for the existence and general magnitude of measured operational benefits and for the reason they exist, which is pre-existing instrumentation with a baseline.

Sources as at 30 August 2026.