The Evolution of Product Master Data Management

The barcode's hard part was never the scanner

Product master data has always advanced by agreement rather than by invention. RFID is the current proof.

At a minute past eight on the morning of 26 June 1974, a cashier at a Marsh supermarket in Troy, Ohio passed a ten-pack of Wrigley's Juicy Fruit over a laser scanner. It rang up at US$0.67. That was the first Universal Product Code ever scanned in a shop, and the equipment behaved exactly as intended.

The technology is the short part of the story. The committee that made the moment possible had been convened at the turn of the 1970s, and very little of what it did was engineering. It brought in consultants to define a numeric format. It argued over how many digits the code should carry, who was entitled to issue one, what a manufacturer prefix signified, and what should happen when someone wanted to change a number already in circulation. Several competing symbol designs were submitted by technology firms; IBM's linear version was approved on 3 April 1973, and a council was established the following year for the sole purpose of administering it.

Then adoption stalled. A shop gained nothing from installing scanners unless manufacturers had already printed codes on most of what it sold, and the working threshold was somewhere near 70 per cent of the products on the shelf. Three years after that first scan, fewer than 200 American grocery stores had scanning equipment installed. The invention was finished. The agreement was not.

What was actually being built

The UPC usually gets filed under automation, which undersells it considerably. What that committee produced was a governance instrument: a rule about who owns a number, who may change it, and what the number is permitted to refer to, backed by an organisation whose entire job is enforcing that rule. The printed symbol is the visible consequence, not the achievement.

The arrangement still runs. GS1 administers it internationally, and GS1 South Africa issues the 600 prefixes that appear on goods produced here. A number means one thing rather than two because an institution exists to prevent the second meaning from taking hold.

That is product master data management. It predates the term by about thirty years.
The sequel that never quite materialises.

Which brings us to RFID, a technology that has been eighteen months away from mainstream adoption for roughly two decades.

Walmart announced the first serious mandate in 2003, instructing its top 100 suppliers to tag cases and pallets within two years. It was extended to all suppliers in 2005 and quietly put on hold by 2007. The usual explanation at the time was cost: tags ran to about US$0.25 each, which is impossible arithmetic on a tin of beans. The mandate returned in 2020 for apparel, expanded into home, toys, electronics and sporting goods in 2022, and took in more than a dozen further categories by February 2024.
The cost explanation has since expired. Passive UHF inlays now sit between US$0.03 and US$0.05 in volume, which is roughly R0.50 to R0.80 or three to four euro cents a tag, and retailers running item-level programmes in apparel report stock accuracy in the high nineties against a baseline that rarely clears seventy. Those are dollar prices at the factory gate, so a South African buyer should add freight, duty and whatever the rand is doing that month before treating the conversion as a landed cost. Even after all of it, the business case works, and it has worked for several years.

Yet outside apparel and general merchandise, RFID remains a pilot technology. Grocery and fast-moving consumer goods, where the inventory accuracy problem is arguably worst, are largely untouched. If the technology is mature and the economics are sound, something else is holding it.

The something else is the same thing the grocery committee spent four years on. The supplier buys the tag; the retailer banks the benefit. The UPC had precisely that split, and resolving it took an industry body, a shared numbering standard and the better part of a decade in which competitors agreed to be bound by the same rule. RFID never got that. It got leverage instead, in the form of one very large buyer telling its suppliers what to do.

Leverage and agreement produce different results. A mandate buys compliance up to its own boundary and nothing past it. A supplier tagging for one retailer's requirement applies the tags at the end of the line, encodes them to that retailer's specification, and changes nothing upstream. The tag becomes a shipping condition rather than a piece of master data. Ask that supplier what its item identifiers look like internally and you will usually find the same spreadsheets that were there before.

And the leverage itself is weakening. Amazon has just ended Walmart's thirteen-year run as the world's largest company by revenue, largely on the strength of cloud, advertising and marketplace services rather than retail. That qualification is the interesting part rather than a defence of Walmart: retail demand is now distributed across marketplaces, discounters and direct channels, and no single buyer sits where the American grocery industry collectively sat in 1973. Whatever replaces the barcode will have to be agreed rather than imposed, and there is currently no forum doing the agreeing.

The same problem, renamed each decade
The internal version has a similar lineage. Manufacturing planning systems in the 1970s introduced the item master file, on the reasonable theory that a business calculating requirements first needs one authoritative description of each part. ERP promised to extend that across the whole organisation: one system, one database, one version of every product.
It mostly did not work out that way, and not because the software was poor. Companies acquired other companies, divisions ran their own instances, and specialist systems were bought wherever a module was weak. By the middle of the 2000s the resulting mess had a name and a software category built around it. Since then, we have added integration platforms, data lakes, APIs that make any system reachable from any other, and now AI models pointed at the accumulated result.

Each generation has solved a transport problem. None has touched the agreement problem.

The cost of moving data between systems has collapsed to near zero, while the difficulty of getting two departments to accept a shared definition of "case" has not moved at all since 1973.

Where it lands in a warehouse

A warehouse management system sits downstream of nearly everyone else's product data. Records arrive from ERP, from procurement, from suppliers, from carriers. The warehouse is where all of it converges and, unhelpfully for whoever runs it, where the consequences become physical.

Unit of measure is the classic. A case is six in one system and twelve in another, both entered correctly by people following their own local convention, and the gap only surfaces as a stock variance three weeks later. Dimensions and weights are the quiet one: cubing, cartonisation, slotting logic and transport rating all depend on them, they are frequently captured once during implementation and never verified again, and no department considers itself the owner. Duplicate item codes appear when the same physical product is bought from two suppliers under two references.

Partial RFID adoption adds a category of its own. A tagged item carries a serialised identifier that has to reconcile with the GTIN on the barcode and with the internal item code, which means a half-adopted programme introduces a third identifier layer before it removes any work. Sites that treat RFID as a labelling project rather than a master data project tend to discover this after the readers are installed.

None of this presents as a data problem. It presents as a picker standing in an aisle holding something the system says should not be there, and the system takes the blame for a decision made in a spreadsheet upstream two years earlier.

The local complication

South African operations carry a version of this that global vendor material tends not to describe.

Trading partners here sit at very different levels of data maturity. A distributor may receive impeccably coded stock from a multinational supplier in the morning and take delivery in the afternoon from a smaller local producer whose barcodes are inconsistent, absent, or reused across variants. Relabelling at receiving becomes routine, and every relabel is a point at which somebody on shift makes a mapping decision that nobody records.

RFID does not transfer cleanly either. Local deployments operate in the ETSI band at 865 to 868 MHz, while the American mandates are written around 902 to 928 MHz, so hardware and specifications do not simply arrive ready-made with the standard. The retailers who have moved here have done so on their own terms rather than in response to anyone's mandate. TFG began rolling out item-level RFID across its fashion brands in 2019, and Woolworths deployed across more than two hundred fashion, beauty and home stores. Both are single-organisation programmes covering their own supply chains, which is exactly the pattern: agreement inside one business is achievable, agreement across an industry is not currently on offer.
Consolidation adds to it. A group that has acquired regional operators inherits their numbering conventions along with their warehouses, and each of those conventions was perfectly rational in isolation. Merging them is presented as a systems integration project when it is really an argument about definitions that needs someone senior enough to settle it.

Questions that tend to produce a pause

•Who in your organisation is permitted to create a new item code, and who reviews it?
•When two departments disagree about what a field means, is there a person whose decision ends the argument, or does it get resolved by whoever is more persistent?
•When last were product dimensions physically verified rather than inherited from an import file?
•When a supplier changes a pack configuration, how does the warehouse find out, and how long is the gap before the record catches up?
•If a customer mandated serialised tagging next year, would you be able to tell them what your item identifiers actually are?

If those questions produce hesitation rather than answers, the problem is already there. It has simply not been expensive yet.

Fifty years on

The grocery committee spent roughly four years arguing about a twelve-digit number, and it was the right allocation of effort. The scanner was ready in months. The agreement was the product.

That distinction has not been learned since. RFID has better physics, cheaper hardware and a working business case, and it has spent twenty years short of the adoption its supporters keep predicting, because a mandate is not an agreement and the industry has not yet built anything that is. The tooling available today is far better than anything available in 1973 and it will resolve nothing on its own. Someone has to own the definition. Someone has to be allowed to say no.