On-demand retail loss: The hidden cost of rapid delivery

Delivery in 30 Minutes, refunds in 30 Seconds. But who’s counting the cost?

On-demand retail loss is becoming one of the most urgent unanswered questions in modern retail.

Customers want convenience, certainty, and increasingly they want groceries delivered within the next 30 or 60 minutes.

For retailers, the opportunity is equally compelling.

Faster fulfilment can increase conversion, encourage repeat purchases and create new reasons for customers to shop.

Rapid delivery has moved beyond an interesting pilot for many businesses and is becoming a material sales channel.

But behind that growth sits a question that has received remarkably little independent attention:

How much money are retailers losing in the race to deliver faster?

At a recent ECR Retail Loss research meeting, one leading retailer told us: “I’d probably categorise it as the largest emerging risk to shrink that we’ve got.”

That does not make rapid delivery a bad proposition. It makes it a proposition whose risks now need to be understood with the same energy that went into its growth.

ECR Retail Loss, working with Prof Michael Townsley and the team at Griffith University, is launching a major international research project into on-demand retail loss.

The study will bring together retailers in Australia, the UK and USA.

Its purpose is to establish a picture of the scale and causes of loss associated with on-demand retail.

And, critically, to identify how retailers can improve profitability without destroying the speed and convenience that make the proposition attractive.

Why on-demand retail loss is becoming impossible to ignore

The challenge at the heart of on-demand retail is simple: speed creates value, but speed can also create risk.

An order may need to be received, picked, packed, substituted where necessary, handed to a driver and delivered to the customer in as little as 30 minutes.

Each stage has to happen quickly, often across several teams and organisations.

When something goes wrong, the response is expected to be faster still.

When a customer reports a missing item, damaged product or incomplete delivery, the retailer may have only seconds to decide whether the claim is genuine and issue a refund.

The pressure to protect customer satisfaction can be intense, particularly when the evidence needed to make a good decision sits in another system or with another organisation.

The result is an uncomfortable paradox.

Retailers invest millions in making products move faster, while potentially giving themselves less time, evidence and visibility to understand where those products went.

Early signals shared with the research team suggest that stores offering on-demand delivery may experience losses 5-10% higher than stores without the proposition.

This is not yet a settled industry benchmark. It is precisely the sort of early observation the new research needs to test, explain and place in context.

The difficulty is that on-demand loss can be absorbed into a store’s broader shrink number.

Products are picked from the same inventory, handled by the same store teams and recorded through systems that may not isolate the channel’s impact.

A new revenue stream can therefore grow quickly without an equally clear view of the losses attached to it.

This is why retailers need evidence that separates normal operating costs from preventable loss, and genuine service failures from deliberate abuse.

Seven hidden risks behind rapid delivery

There is no single cause of on-demand retail loss. The research launch identified at least seven points where loss, error or uncertainty can enter the journey.

These pressures turn rapid delivery risk into a wider problem of evidence, ownership and operational control.

Effective quick commerce loss prevention therefore has to examine the whole journey, not just the customer’s final claim.

1. The product was never available

An inventory record may show that a product is in the store when the shelf is empty. The picker cannot find it, a substitution is made, or part of the order goes unfulfilled.

Pressure to hit both pick-rate and fill-rate targets can create a further temptation.

If a customer orders a small bottle of ketchup but only the larger size is available, a picker may put the larger bottle in the order without recording it as a substitution.

The order appears to have achieved a 100% fill rate, but the store has unknowingly given away the extra value and made its inventory record less accurate.

What begins as an attempt to protect service measures becomes a source of loss.

2. The product was available but not picked correctly

Speed targets can compress the time available to check an order.

A product may be missed, the wrong variant selected, an item scanned but placed in another bag, or a chilled product left outside the correct temperature-controlled area.

These are not necessarily fraudulent events. They are fulfilment failures, but they still erode profit and can be difficult to distinguish from a false claim after delivery.

3. The order was packed, but the handover was weak

Multi-bag orders create a deceptively simple risk. The store may complete three bags, but only two are collected.

Without a reliable handover record, the retailer may know that the products were picked but not whether the complete order left the store.

The precise handover point matters: who collected the order, which bags were included, when responsibility changed and what evidence was captured?

4. The last mile sits outside the retailer’s view

Retailers may control customer relationships and store operations while outsourcing the final mile.

Alternatively, a delivery platform may own the customer interface, courier relationship and claim decision while the retailer fulfils the order.

In either model, one party can hold evidence that another cannot see.

The retailer may have excellent pick data but limited proof of final delivery. The platform may see driver signals and customer history but have little insight into what happened inside the store.

5. Cancelled and failed orders do not complete the return journey

If a delivery fails or is cancelled, what happens to the goods? Were they returned to the store? Was every bag returned? Was the stock booked back into a retailer-controlled system?

For grocery products, there is also a food-safety question. Once an item has left the retailer’s control, it may not be possible to return it to sale. Even where a return is recorded, the value may still become shrink.

6. Refund decisions are separated from loss ownership

The person or system approving a refund may not own the resulting loss.

A platform can make a rapid customer-service decision while the cost is allocated back to a store.

The store may see higher shrink without access to the claim evidence, decision rule or customer history behind it.

Commercial agreements can make this more complicated. A retailer may accept a complaint or refund rate as part of a platform relationship without giving individual stores the visibility or control needed to improve it.

7. Deliberate abuse exploits gaps between the stages

Some claims will be dishonest, with customers falsely reporting missing items or damaged goods. While drivers or other participants may deliberately interfere with an order.

Repeat abuse may be hard to identify when activity is split across several platforms and systems.

Collusion creates another exposure.

If the person placing a home-delivery order knows the person picking it, there is a greater temptation to add extra items that have not been scanned or paid for.

Weak checking and pressure to complete the order quickly can make that theft harder to detect.

Previous ECR Retail Loss work on missing-item claims from third-party grocery deliveries has already highlighted the consequences of limited customer data, weak courier identification, fragmented accountability and refund decisions made with minimal verification.

The new study will go wider to examine the complete journey and answer:

• How much loss comes from deliberate abuse;
• how much comes from process failure;
• and how much remains unclassified because no one has the evidence to tell the difference.

The core problem is chain of custody

At the centre of the research is a familiar concept from crime prevention and evidence management: chain of custody.

An on-demand order can pass through numerous stages and several different organisations:

1. The customer places and pays for the order;
2. The retailer or platform receives it;
3. A store colleague picks the products;
4. Substitutions or unavailable items are recorded;
5. The order is packed into one or more bags;
6. The bags are handed to a driver;
7. The driver transports and delivers the order;
8. The customer raises a claim;
9. A retailer or platform decides whether to refund or reimburse;
10. The cost and feedback are allocated somewhere in the system.

At every handover, evidence can be captured, fragmented or lost.

• Was the product actually available?
• Was it picked?
• Was it placed in the correct bag?
• Were all the bags handed to the right driver?
• Did the driver deliver the complete order?
• Was a cancelled order returned?
• Is the customer’s account consistent with the evidence?
• Most importantly: who can answer each question?

One party may have detailed information about its part of the journey and almost no visibility of the next.

Claims data, driver information, reimbursement decisions and store losses can sit in different systems, under different ownership and subject to different commercial rules.

This matters because a scan only proves what it was designed to prove.

Scanning an item can establish that a picker handled it, but not necessarily that it entered the correct bag, that every bag left the store or that the order reached the customer intact.

The objective is to identify the minimum useful evidence required at each handover, make it accessible to those who need it, and use it to prevent the same failure from happening again.

When seeing is no longer believing

Artificial intelligence adds a new dimension to the problem.

At the research launch, Professor Townsley demonstrated how an ordinary photograph of packaged chicken could be transformed in about a minute into a convincing image suggesting that the product had leaked and been damaged.

For a retailer trying to resolve a complaint quickly, the implications are significant.

Photographic evidence can no longer be treated automatically as independent proof that a claim is genuine.

That does not mean every photograph is suspect, or that genuine customers should face an obstacle course to obtain a fair refund.

Rather, it means images need to be assessed alongside other evidence:

• The order history;
• pick-and-pack records;
• bag handovers;
• delivery confirmation;
• claim patterns;
• and any relevant product information.

Some retailers are already exploring image metadata and automated checks as additional signals.

Those tools may help, but no single technical test should carry the whole decision.

Evidence can be incomplete, stripped of context or misunderstood.

A stronger answer is a layered evidence model that makes it harder for both fabricated claims and genuine service failures to disappear into the same refund process.

Combine easily manufactured evidence with automated refunds, pressure to protect customer satisfaction and a business model built around speed, and retailers potentially face a new generation of loss.

The response must be proportionate. The goal is not to slow every claim down, but to recognise which claims can be resolved automatically, which need additional evidence and which patterns indicate a wider control failure.

Not all on-demand loss is fraud

The most damaging mistake would be to treat on-demand retail loss as another name for customer fraud.

On-demand grocery loss can begin well before a customer or driver becomes involved.

It may start with an inaccurate stock record, a poor substitution decision or an item placed in the wrong bag.

Losses can originate from:

• inaccurate inventory records;
• genuine service failures;
• picking, packing or substitution mistakes;
• incomplete multi-bag handovers;
• driver errors or misconduct;
• failed returns;
• weak processes and controls;
• fragmented data;
• automated or poorly informed refund decisions;
• commercial agreements that separate decisions from consequences;
• unclear ownership and escalation;
• deliberate customer abuse.

The boundary between accidental and deliberate loss is not always clean.

The same operating conditions that make genuine mistakes more likely, such as speed, weak checks, unrecorded substitutions and fragmented evidence, can also create opportunities for theft and collusion.

In other words, an environment where things can easily go wrong without malicious intent can also make deliberate loss easier.

Each of these causes demands different responses.

• A training or scanning intervention may reduce repeated picking errors.
• A stronger driver handover process may address missing bags.
• Better data sharing may allow a retailer to identify repeat-claim patterns.
• Revised commercial agreements may clarify who owns a loss and who must provide evidence.
• A risk-based claim process may reserve human review for the small proportion of cases where the available signals conflict.

ECR Retail Loss’s existing framework for measuring e-commerce loss distinguishes intentional from accidental risks and places losses across fulfilment, returns and logistics.

The new on-demand research will build on that thinking in an operating model where decisions are compressed, and responsibility is often shared.

Without that separation, retailers risk solving the wrong problem.

Extra checks intended to deter dishonest claims could add cost and friction while leaving the underlying fulfilment failure untouched.

Equally, treating every claim as a service issue can allow repeat abuse to scale.

The first task is to establish what happened. The second is to assign the right intervention.

From anecdotes to international evidence

The new research will move on-demand retail loss from anecdote to evidence.

The project will compare two prevalent operating models.

In the first, the retailer owns the customer relationship and controls much of the order journey.

The last mile may be outsourced, but retailer rules and systems shape the proposition.

In the second, a platform owns the customer front end and manages delivery while the retailer picks and assembles the order inside a platform-led journey.

The platform may also adjudicate claims and hold more of the customer or driver evidence.

Neither model is inherently good or bad. Each changes who can see, decide and act at different points.

The research will examine how those differences affect loss, accountability and the ability to intervene.

Initial fieldwork will focus on rapid grocery delivery. The project will compare experience in Australia, the UK and the US, with scope to include other markets where strong participation makes that useful.

The work will combine:

• store visits and structured observation;
• interviews with store, head-office, commercial, loss-prevention, customer-care and platform teams;
• surveys of customers and gig workers;
• analysis of order, sales, claim, refund and operational data at a level feasible for each retailer;
• comparisons across stores, markets, platforms and operating models.

The strength of the conclusions will depend on the data available. Aggregated data can establish broad rates.

Repeated store or channel data can show how loss changes over time and help account for volume and seasonality.

Before-and-after comparisons may provide stronger evidence about what happens when a proposition launches or a control changes.

With strong participation, the research aims to produce:

• a robust estimate of loss associated with on-demand delivery;
• evidence on whether loss rises as adoption or utilisation grows;
• comparisons across stores, markets, platforms and operating models;
• a practical breakdown of the main loss drivers;
• a chain-of-custody framework for the order and claims journey;
• a roadmap for reducing loss without killing the proposition.

This complements our wider work on e-commerce loss and its exploration of the economics of quick-commerce delivery.

Why retailers should take part

The quality of the answers will depend on the quality and diversity of retailer participation.

The research team wants to hear from mature operators processing significant volumes of rapid-delivery orders, retailers using several platforms or operating models, and businesses still designing or piloting their propositions.

Being early in the journey is not a disadvantage

Retailers that have not yet launched can help the study understand which decisions become fixed during design and where loss-prevention thinking can be introduced before weak processes become embedded.

Participation will be flexible

It may include interviews with operational and decision-making teams, selected store visits, discussion of processes and controls, and the sharing of operational data at a level that is feasible for the individual business.

Participation can be structured around commercial sensitivity.

The purpose is to develop a collective evidence base, not expose the performance of individual retailers or platforms.

For participants, this is a chance to:

• compare experience with retailers operating in other markets and models;
• understand where loss may be hiding inside existing store measures;
• identify evidence gaps before investing in additional controls;
• test assumptions about fraud, service failure and driver-related loss;
• help create practical standards for handovers, claims and returns;
• shape a roadmap grounded in real retail operations.

The research launch demonstrated the value of that collaboration.

Retail participants raised questions about incomplete multi-bag deliveries, validating handovers, driver returns, fragmented data, customer claims and the reputational risk created when third parties become representatives of the retailer’s brand.

Those are critical design questions for a channel expected to keep growing.

Speed and control do not have to be enemies

On-demand retail is growing. The customer proposition is compelling. The technology is improving, and the volumes are increasing.

But growth without visibility creates a problem. A retailer cannot control a loss it cannot locate, explain or separate from the wider store number.

The answer is not to burden every order with friction or treat every customer as a suspect.

Nor is it to accept unexplained refunds and shrink as the unavoidable price of convenience.

The opportunity is to build controls that move at the speed of the proposition:

• reliable inventory;
• accurate picking;
• visible handovers;
• shared evidence;
• clear decision rights;
• proportionate claim checks;
• and feedback that reaches the part of the operation able to act.

That all begins with better evidence.

Are retailers building a faster route to the customer, or a faster route to loss? This new research intends to find out.

Retailers in the UK, United States, Australia and other markets are invited to take part and help create the first international evidence base on loss, risk and profitability in on-demand retail.

To discuss participating in the research, contact Colin Peacock.

Because the future of rapid delivery should not be a choice between speed and profitability. The opportunity is to achieve both.

Frequently asked questions about on-demand retail loss

What is on-demand retail loss?

On-demand retail loss describes preventable events and outcomes that reduce profitability in rapid-delivery retail. It can include picking mistakes, missing products, incomplete handovers, failed returns, incorrect refunds, driver-related incidents and deliberate abuse.

It is broader than fraud because it also includes genuine process and service failures.

Is on-demand retail the same as quick commerce?

The terms on-demand retail, quick commerce and rapid delivery are often used for closely related propositions in which customers receive products much faster than through conventional e-commerce.

Definitions vary between markets, but the new research is primarily concerned with retail orders delivered within roughly 30 to 60 minutes.

Why can rapid grocery delivery create more loss?

Rapid grocery delivery compresses picking, checking, handover and claim decisions.

Goods and information may also pass between a retailer, delivery platform, driver and customer.

If those parties cannot see the same evidence, mistakes and abuse become harder to locate, challenge and prevent.

Is on-demand retail loss mainly customer fraud?

The short answer is no.

Dishonest missing-item or damage claims are one potential source, but losses can also arise from inaccurate inventory, picking errors, incomplete orders, weak handovers, failed driver returns, fragmented systems and unclear commercial responsibility.

The research will help retailers separate these causes and choose proportionate interventions.

What does chain of custody mean in on-demand delivery?

Chain of custody is the evidence showing who had the goods, where and when, across the order journey.

In on-demand delivery, it can cover picking, packing, bag sealing, driver collection, final delivery, returns and the claim decision.

Stronger custody evidence helps distinguish genuine failures from deliberate abuse.

How can a retailer participate in the research?

Participation can be tailored to the retailer. It may involve interviews, store visits, process reviews or the sharing of aggregated operational data.

Retailers with mature on-demand operations, businesses using third-party platforms, and companies still planning a pilot can all contribute.

Contact ECR Retail Loss to discuss the most appropriate level of involvement.