Barry's Job Just Got Replaced By An E-Ink Screen

Barry's job just got replaced by an e-ink screen.

Walmart has digital shelf labels live in roughly 2,300 US stores, with the rest of the chain due to follow.

A Supercenter can carry more than 120,000 products, so this is not a tasteful little pilot beside the avocados.

It is hundreds of millions of tiny screens, wired back to systems that can change a price in minutes instead of sending someone down an aisle with a printer and a box cutter.

Wendy's put US$20 million behind digital menu boards for the same basic reason.

Woolworths has rolled electronic labels through more than 170 New Zealand stores, and Foodstuffs has been using them across much of its network for years.

Every company says some version of the same thing.

This is about accuracy, efficiency and giving store teams more time for customers.

It is.

It is also the start of something larger.

Retailers are quietly building an operating system for the physical store.

Not the ERP with the black screen that takes two days of training before it will admit a product exists.

The other kind: the layer that tells a business what is happening on the shelf, around the shelf and because of the shelf.

Ecommerce has had this visibility for years.

Every click, abandoned cart, search and product-page wobble produces a trail somebody can put in a dashboard.

Physical retail has mostly had a door counter, last week's sales and a manager saying Saturday felt busy.

That gap is closing across retail, QSR and FMCG.

The interesting problem is that it is closing one system at a time.

The shelf label knows the price.

The camera knows the traffic.

The loyalty platform knows the customer.

The point of sale knows what got bought.

The loss-prevention platform knows who it is worried about.

They all live in the same store and often behave as if they have never met.

The Barry Problem

For decades, a price change in a physical store meant a decision at head office, a file sent to the branch, a tag printed somewhere out the back and Barry walking the aisle to replace it.

New price, new rectangle of paper, same box cutter.

Do that across tens of thousands of products and thousands of weekly changes and Barry stops looking like a rounding error.

The labour case for electronic labels is obvious, but it is not always instant.

The screens, rails, gateways, installation, software and integration all cost money, while the savings arrive one avoided task at a time.

The better case includes fewer shelf-to-till errors, faster markdowns, easier online picking and a reliable link between a product record and its physical position.

That is why the New Zealand rollout matters more than another futuristic concept store in Las Vegas.

This is already normal infrastructure here.

Woolworths said it had 1.2 million labels operating in its New Zealand network by late 2023, while the Grocery Commission's first annual report confirmed that every major grocery group had begun deploying them.

Vusion, the French company supplying Walmart, finished 2025 with €1.53 billion in adjusted revenue, up 51 percent, and 375 million labels connected to its cloud.

Its total installed base was larger again: 650 million labels across 75,000 stores.

That is not a screen business pretending to be software.

It is a hardware footprint becoming a data platform.

A label can be tied to a product, position, price file and picking workflow, while newer Bluetooth-enabled rails can support navigation and proximity services.

Barry's paper tag did one job.

The thing replacing it is an endpoint.

One Store, Five Versions of the Truth

This is where the connected-store pitch starts to get untidy.

Retail technology is usually bought against one budget and one measurable problem.

Pricing buys electronic shelf labels.

Operations buys footfall and queue monitoring.

Visual merchandising buys floor-planning software.

Loss prevention buys cameras, number-plate recognition and incident intelligence.

Marketing buys loyalty, location and offer tooling.

Each decision can be sensible on its own.

Together they create five versions of the same Tuesday afternoon.

The traffic system says 600 people entered.

The merchandising system says the front fixture underperformed.

The point of sale says the product missed plan.

The loyalty system says regular customers ignored the offer.

The loss-prevention system says the spirits aisle had a problem.

The useful question is not whether each dashboard is correct.

It is whether the retailer can connect the sequence without exporting four CSVs and inviting everyone to a meeting.

A person entered, passed a fixture, stopped at a shelf, saw a price, picked something up and either bought it or did not.

Online, that is a funnel.

In a store, it is currently a committee.

Cameras, Man

There is now an entire industry devoted to turning a floor plan into something that looks like Hotjar.

RetailNext, FootfallCam, Xovis and others count entries, measure dwell time and show where people cluster or keep walking.

Red where the store is alive.

Blue where a fixture is quietly paying rent to hold six jackets nobody has touched.

The cleaner products use overhead depth cameras, radar or other sensors to measure movement without trying to identify a person.

That distinction matters, although “anonymous” is a design choice rather than a magical property of a camera.

What gets captured, retained, combined and exported decides how anonymous the system really is.

Flagship, the Sydney company founded by Simon Molnar, attacks the same visibility problem without pretending every answer needs another camera.

It connects floor plans, product placement and sales so a retailer can see which fixtures are earning their space.

That is more interesting than the original hardware story because it treats the store as a commercial model rather than a surveillance project.

It also exposes the integration problem nicely.

Knowing a fixture is weak is useful.

Knowing whether it was weak because of placement, stock, price, traffic or the promotion beside it requires data owned by somebody else.

The heat map sees movement.

Flagship sees merchandising performance.

The till sees revenue.

The shelf label sees the price it was told to show.

Same aisle, four perfectly respectable partial truths.

The Face at the Door

The visibility argument becomes harder when the system stops measuring a store and starts identifying a person.

Bunnings switched on facial recognition in New Zealand after threatening incidents in its stores rose from 303 to 697 over four years, with repeat offenders linked to about a third of them.

Foodstuffs had already trialled the technology, and the Privacy Commissioner found that trial complied with the Privacy Act despite the high level of intrusion involved in scanning every face entering a participating store.

The company sitting behind much of New Zealand's retail-crime infrastructure is Auror.

Three University of Auckland students started it in 2012 because theft reports were slow, disconnected and largely useless across stores.

Auror now says its platform is used in more than 85,000 stores, and an NZ$82 million funding round led by Axon valued the company above NZ$500 million.

It launched its Subject Recognition product in October 2025, adding live facial-recognition workflows to a platform already used for incident reporting and automatic number-plate recognition.

Its ANPR network has also ended up inside a much bigger legal argument.

New Zealand Police search commercial ANPR platforms hundreds of thousands of times a year, and the Court of Appeal has been considering whether police access to historical Auror data should require stronger controls.

That is a long way from helping Barry report a stolen drill.

It is also how infrastructure expands: a narrow operational tool becomes useful to the next team, then the next agency, then a purpose nobody put in the first sales deck.

Flock Safety is the American warning label here.

Its vast camera network has helped police solve crimes, but it has also faced backlash over data sharing, access and security, including a 2026 investigation in which people who obtained one camera recovered an on-device encryption key and weeks of images and clips.

Different company, different country, same executive question.

Before connecting more data, who is allowed to use it, for what, and what happens when the answer changes later?

Pricing Gets Weird Quietly

The visible camera gets the sign on the door.

The less visible pricing system often gets a paragraph in the terms and conditions.

In 2023, a New Zealand Reddit user found that his McDonald's app showed prices up to $3 higher than his partner's for the same offer, at the same location, on the same day.

His account also had more loyalty points.

That does not prove McDonald's charged him more because he was loyal, and the company did not disclose the rule behind those two offers.

It did confirm that personalised deals had been part of the app since the loyalty programme launched.

That uncertainty is the point.

The customer could see the different number but could not see the decision that produced it.

Instacart built a larger version of the same trust problem in the United States.

It acquired pricing platform Eversight for US$59 million and let a small group of retailers run item-price experiments through its marketplace.

A 2025 investigation involving 437 shoppers found that nearly three quarters of the tested grocery items appeared at more than one price, with some differences reaching 23 percent.

Instacart said the tests did not use personal, demographic or behavioural data to set prices, but ended them after the report caused a political and customer backlash.

The distinction between a random price test and personalised pricing matters legally.

It matters rather less when two people compare phones beside the crackers.

Uber gets away with changing prices because the customer understands, however unhappily, that time, place and demand are moving the number.

Grocery pricing feels different because the shelf has trained us to expect one public answer.

New Zealand now has a specific biometric code requiring proportionality, safeguards and clear notice when facial recognition is used.

We do not have an equally obvious sign explaining when customer data changes an offer or price.

The Privacy Act and Fair Trading Act still apply, but neither gives a shopper a neat board at the entrance saying: “your price may be different from theirs.”

The more visible technology got the clearer rule.

The one changing the wallet is still mostly explained by lawyers.

The Shelf Does Not Know You

This is where the dystopian version of the story usually gets ahead of the engineering.

An electronic shelf label does not decide a price.

It shows the price sent by the retailer's pricing system, and the public screen is visible to everyone standing beside it.

A Bluetooth beacon does not know your loyalty number either.

It broadcasts an identifier, and a retailer's app can recognise that signal if the customer has installed the app and granted the right permissions.

The app may then connect location to an account and present a personalised offer on the phone.

That is already plenty powerful without inventing a shelf that changes from $4.50 to $5.20 depending on which of three people is looking at it.

A shared screen has a shared-state problem.

Two customers, one label, one number.

Minority Report solved this by having every billboard yell at Tom Cruise personally, which is one of several reasons it should not be used as a requirements document.

The credible near-term risk is quieter.

The public shelf price stays fixed while the app, coupon, delivery marketplace or checkout applies a different offer to each account.

The electronic label makes the base price easier to move, the loyalty system segments the audience and the pricing engine decides what each segment sees elsewhere.

Those systems do not need to merge into one evil machine.

They only need enough identifiers and timing to make the customer unsure which price is real.

That makes price integrity a product decision, not a legal footnote.

A retailer should be able to explain the source of truth, how often a price can move, which channels may differ and what a staff member does when the shelf, app and till disagree.

If the answer needs an architecture diagram, Barry is going to have a difficult afternoon.

A Better Example Already Exists

QSR is producing a more useful version of the connected-store thesis because the fragmentation is impossible to hide behind a nice shop floor.

A multi-site operator can have a point of sale, rostering platform, inventory system, supplier portal, delivery marketplaces, invoice inbox and a spreadsheet quietly doing the job everyone thought the ERP did.

Meuze, co-founded by racing driver Jack Doohan, is building a decision layer across those systems for multi-location QSR operators.

It connects to the tools already in place, then uses sales history, weather, events, labour and supplier data to forecast demand and recommend or automate ordering, prep and staffing.

The Formula 1 joke writes itself, but the important bit is that Meuze starts with the disconnected operation rather than another sensor.

It is selling the joins.

The company says it reached US$9 million in contracted annual recurring revenue within eight weeks of relaunching and was operating across more than 2,000 locations by mid-2026.

Contracted ARR is not collected revenue, and startup numbers deserve the same raised eyebrow as every other vendor slide.

The demand still tells us something.

Operators do not need another dashboard telling them yesterday was busy.

They need today's forecast to change the order, the roster and the prep list while there is still time to do something.

That is the loop physical retail keeps failing to close.

Signal, decision, action, outcome.

Most store technology sells one of the four.

The value appears when somebody owns the whole sentence.

Nothing Is Actually Stitched Together

There are platforms that can connect parts of this stack, and every vendor now has an API page with enough arrows to make it look finished.

That is not the same as one operating model.

Vusion can connect labels, shelf position and picking workflows.

Flagship can connect floor plans, placement and sales.

RetailNext can connect traffic, dwell and conversion.

Auror can connect incidents, people of interest and vehicles.

Meuze can connect restaurant demand to operational actions.

The POS, ERP, loyalty platform and workforce system still keep their own records, permissions, clocks and definitions.

One system calls it a store.

Another calls it a location.

A third calls it site 0147 because somebody made a good decision in 2009 and then left.

This is the part that disappears in a demo.

The hard problem is not moving data between systems.

It is agreeing on what the data means, who is allowed to act on it and which system gets the final word when two models disagree.

A connected store needs a shared identity for products, locations and fixtures.

It needs event times that can be compared, a clear system of record for price and stock, and a way to measure whether an automated decision helped.

It also needs boundaries around customer identity, retention and secondary use before the loss-prevention feed quietly becomes a marketing input.

Those are architecture decisions, but they are also CEO decisions.

They decide where margin comes from, which risks the company accepts and whether the customer is part of the design or merely the thing being measured.

The Order Matters

The answer is not to wait for one enormous platform to own the entire shop.

That is how a sensible integration problem becomes a seven-year transformation programme with matching lanyards.

Start with the boring use case that has a clean operational owner and a visible result.

Price accuracy is good because the shelf and till can be compared.

Online picking is good because time per order and substitution rates can be measured.

Fixture performance is good when placement data can be tied to sales rather than a vague claim that the store feels more premium.

Queue prediction is good when somebody can change staffing before the queue appears.

Then build the joins deliberately.

• Give products, stores, zones and fixtures stable identities across systems.
• Decide which platform owns each decision before the models start making them.
• Keep identifiable customer data separate until there is a clear benefit, permission and retention rule.
• Measure the action against an outcome, including a control group where the business can support one.
• Give store teams a way to override the system and record why.

That final point is not a concession to people who fear technology.

The manager on the floor can see the freezer leaking, the school bus arriving and the local event the forecast missed.

If the software cannot absorb that context, it has not replaced gut feel.

It has merely become another opinion.

So What?

Physical retail is becoming programmable.

That does not mean every store is about to recognise your face, change your cereal price and send the result to Palantir before lunch.

It means the shelf, floor, till and customer account are finally producing enough usable signals to change how a store is run.

The retailers that benefit will not be the ones with the most sensors.

They will be the ones that can turn a signal into a decision, a decision into an action and an action into a measured result without losing the customer somewhere in the plumbing.

There is plenty to do before the grand platform arrives.

Replace manual price changes where the maths works.

Connect store layouts to sales.

Use traffic data to test a decision rather than decorate a board report.

Treat facial recognition and customer-level pricing as separate risk decisions, even if a vendor would quite like them on the same slide.

Most of all, decide what the store is allowed to optimise for.

Labour hours, availability, conversion, safety and margin can all improve while trust gets worse.

The dashboard will not flag that unless somebody asked it to.

Barry's job was not really replaced by an e-ink screen.

The box cutter was.

What happens to Barry's judgement is the more valuable question.

The connected store will be built shelf by shelf, camera by camera and integration by integration.

It should also be explainable one decision at a time.

Shoutouts: Walmart · Wendy's · Woolworths NZ · Consumer NZ · Grocery Commission · Vusion · RetailNext · Flagship · Bunnings NZ · Privacy Commissioner · Auror · New Zealand Police · Wired · NZ Herald · Consumer Reports · Instacart · Meuze · Apple iBeacon