TL;DR: Returns are not a customer-service problem; they are a reverse supply chain that most online retailers run on spreadsheets, courier portals and goodwill. The cost sits in four places: the refund itself, transport, handling and inspection, and the stock that sits in limbo before it can be resold. A returns architecture fixes this with five building blocks: a rules engine at the point of request, an RMA (return merchandise authorisation) record that follows the parcel, a disposition workflow at the warehouse, a refund and finance link, and a data loop back into the catalogue. This guide explains each block, the UK and Australian rules that shape them, and how to decide between configuring a platform and building something custom.
Why returns deserve their own architecture
Most shops are designed around the forward journey: browse, basket, pay, dispatch. Returns get bolted on later, usually as a form that emails a human. That works at 50 orders a week. It stops working when returns become a double-digit share of revenue and each one touches customer service, the warehouse, the courier, the payment provider and finance.
The scale is well documented in the US. The National Retail Federation and Happy Returns estimated that retailers would see $849.9 billion of returned merchandise in 2025, with an estimated 15.8% of annual sales returned and 19.3% of online sales. The same report found that 9% of all returns are fraudulent. Those are US retailer estimates and survey results, not UK or Australian figures, but they show the order of magnitude and the direction of travel.
For British and Australian retailers, the more useful benchmarks come from merchant data. Loop Returns analysed 13.8 million returns across more than 4,000 Shopify merchants and reported, in its 2025 State of Ecommerce Returns Report (as covered by Retail Times), a UK return rate of 17.5% and Australian rate of 10.9%. Two other numbers from that report matter more for architecture: 78.1% of UK returns ended in a refund and only 5.8% in an exchange, the lowest exchange rate of any region it measured. Australia's exchange rate was 13.2%. The report is a Shopify-merchant benchmark, not a census of all retail, but the gap tells a story: where the return flow nudges customers towards exchange or store credit, revenue is retained. Where it only offers a refund, it leaves.
If you are a founder or operations lead, the question is not "how do we stop returns?" You cannot, and you should not want to. The question is "what does each return cost us, and which of those costs are avoidable?"
Where the money actually goes
It helps to split the cost of a return into stages, because each stage has a different fix.
1. Decision cost. Someone, or something, decides whether the return is eligible. If a person does this by reading an email and checking an order, you are paying for labour on every request, including the obviously valid ones.
2. Transport cost. The parcel goes back, by courier, drop-off point, or the customer's own post. Who pays and which carrier is used are policy decisions, but also data decisions: you can only optimise what you can measure per return.
3. Handling and inspection cost. The parcel arrives, is opened, matched to an order, inspected, graded, and either restocked, refurbished, liquidated or written off. This is where most hidden cost lives. Loop's data shows a UK average of 9.97 days from return initiation to being processed. Every day stock is in transit or waiting in a returns bay is a day it cannot be sold, which matters most for seasonal or fast-moving lines.
4. Value leakage. The refund goes out, but the customer does not come back; the item is resold at a discount or not at all; or the return was never legitimate in the first place. Retail Gazette reported in October 2026 on an analysis of one million UK returns processed between July 2025 and May 2026, in which almost 39,000 were flagged as potentially suspicious, representing £29 million in losses. The article attributes the underlying work to Riskified and notes that suspicious returns took a median 18.8 days to come back, almost twice the 9.5 days for typical returns. That gap is itself a useful signal for a rules engine, which we return to below.
A useful exercise before you write a single line of code: take the last 200 returns and trace each one through those four stages. Record who touched it, how long it waited, and what happened to the item. Most teams are surprised by how much of the cost is waiting and re-handling rather than postage.
The legal frame: what you must do, and what is policy
Architecture should encode the law as configuration, not as tribal knowledge. The two markets behave differently.
United Kingdom
For online, mail and phone orders, GOV.UK guidance explains that customers can cancel within 14 days of receiving the item and then have a further 14 days to send it back. The seller must refund within 14 days of receiving the item back. You cannot deduct fees from the refund unless the item has been used or damaged, and standard delivery must be refunded (if the customer chose a more expensive delivery option, only the standard cost has to be refunded). No reason is required for the return. There are exceptions, including personalised items, perishables and sealed hygiene products, where a refund is owed only if the goods are faulty.
For a system, that means the return window is not one number. There is a statutory cancellation window, a faulty-goods position that sits outside it, and whatever longer window you offer as a commercial policy. Your rules engine should model these separately so a customer service agent never has to remember which clock applies. If you are also reviewing checkout and consumer-law obligations more broadly, our piece on engineering checkout for UK, Australian and US consumer law covers the front half of the same journey.
Australia
Australia works from consumer guarantees rather than a statutory cooling-off period for ordinary online purchases. The ACCC explains that the remedy depends on whether the problem is major or minor. For a major problem with a product, the consumer can choose a refund or a replacement. For a minor problem, the business must fix or repair it for free, although it can choose to offer more. A refund should be the full amount paid, and the business must not deduct an amount for the use the consumer has had of the product. If the product is confirmed as faulty, the business must reimburse reasonable return costs the consumer has already paid.
On change of mind, the ACCC's position is that consumers who simply changed their mind are not entitled to a repair, replacement or refund under the guarantees, but that if a business has a change-of-mind policy it must follow it. Policies such as "no refunds" cannot be used to deny guarantee rights.
The architectural consequence: your return reasons must be structured, not free text. "Faulty", "not as described" and "changed my mind" lead to different legal remedies in Australia and different cost owners everywhere. A dropdown that captures the reason accurately at the start saves an argument at the end.
This is a general overview of public guidance, not legal advice. Check your own obligations with a qualified adviser, particularly for regulated or personalised goods.
The five building blocks of a returns architecture
1. A rules engine at the point of request
The first job is to decide, automatically and consistently, what a customer is offered. Inputs might include the product category, the order date and delivery date, the stated reason, the customer's history, the payment method, and the sales channel. Outputs are an eligibility decision, a resolution menu (refund, exchange, store credit, repair), a return method, and who pays for shipping.
Good rules engines are boring: a table of conditions that a non-developer can read and a retail operations manager can change without a release. Keep rules versioned, so you can answer "what policy applied to this customer on 3 March?" months later.
Rules are also where exchange-first design lives. If UK exchange rates are as low as the Loop data suggests, there is room to test offering an exchange or credit first, for example for size or colour issues, while leaving the refund route clearly available. Do not hide the refund: statutory rights stay, and hiding them invites complaints.
2. An RMA record that follows the parcel
Every return gets an RMA: a single record with an identifier, the order lines involved, the reason code, the approved resolution, the label or drop-off reference, and a status history. The point is that customer service, the warehouse and finance all look at the same object.
Minimum statuses worth modelling: requested, approved, label issued, in transit, received, inspected, resolution issued, closed. Add timestamps to each. The gaps between timestamps are your operational metrics, and they are where the 9.97-day average (or your own equivalent) gets broken down into fixable parts.
If you sell across several channels, the RMA should sit alongside your order management layer rather than inside any one storefront. We wrote about that choice in build, buy or extend for multi-channel order management.
3. A disposition workflow at the warehouse
Disposition is the decision about what happens to the physical item. Typical outcomes are restock as new, restock as open-box, refurbish, return to supplier, liquidate, donate or destroy. The warehouse team needs a handheld or tablet flow that scans the RMA, shows what is expected, captures condition with a photo, and offers only the dispositions that make sense for that product.
Two design points make a large difference. First, make the next action obvious: the operator should never have to decide which system to open. Second, capture the grade at the moment of inspection and feed it back to the refund step, so that, for example, an item received in a condition that does not meet your policy goes to review rather than being refunded automatically. Under UK rules, deductions are limited to used or damaged goods, so the evidence you capture here (photos, timestamps, a condition code) is what protects you in a dispute.
If you run physical stores as well as online, returns often arrive at a till. Unifying that flow with the same RMA record avoids the common split where in-store returns never inform your online stock position. See our overview of retail POS and inventory.
4. A refund and finance link
Refunds should be triggered by events, not by someone typing amounts. When the RMA reaches the "inspected, approved" state, the system calculates the refund (including how delivery is treated under the applicable rules), calls the payment provider's refund API, and posts the result to finance. Failures, such as an expired card or a partial refund, should create a task, not silently vanish.
Finance teams need two further things: a clean ledger entry for each refund and a way to reconcile what was refunded against what the provider says it paid out. If your returns volume is high, manual reconciliation becomes its own cost centre; our article on reconciliation automation for payment platforms explains the patterns.
5. A data loop back to the catalogue and to risk
The cheapest return is the one that never happens. Reason codes aggregated by SKU show where product pages mislead: Riskified's data, reported by Retail Gazette, indicates more than half of consumers have returned an item because it differed from its online images or description, and 45% have bought multiple sizes. Structured returns data lets you fix size guides, photos and descriptions where the problem is concentrated, rather than guessing.
The same data feeds risk scoring. Retail Gazette's reporting on the UK analysis also quotes a survey in which 32% of UK shoppers said they had bought an item to use before returning it, rising to 43% among under-34s. Treat such figures as indicators of behaviour, not as a verdict on your own customers: most returns are legitimate. A sensible approach is tiered friction. Most customers get the fast, free-feeling path; a small number of unusual patterns (for instance, very high return ratios or mismatched items) trigger inspection before refund. Our page on fraud and risk analytics describes the underlying techniques, which transfer well from payments to returns.
Charging for returns: what the data says and what to do with it
Paid returns are spreading. A benchmark by Ingrid, Harper Concierge and Limesharp, reported on Ingrid's blog, found that 23% of the UK's top 100 fashion retailers charged for returns in 2023, rising to 35% by 2026, and that no brand that introduced fees between 2023 and 2026 had subsequently removed them. Loop's Shopify data found around two-thirds (66.1%) of UK merchants in its sample charging return fees. These are different samples, so do not combine them; they point the same way, but they do not prove that fees improve profit for your business.
What the architecture should give you is the ability to run the experiment safely. With a rules engine, you can trial a fee for one category or one customer segment, measure the effect on conversion and return rate, and switch it off without a code release. Without one, a pricing decision becomes an engineering project. A few considerations:
- Legal fit. In the UK, the cancellation rules limit what can be deducted from a statutory refund, so a fee structure needs checking against the guidance above. In Australia, fees cannot be used to avoid guarantee obligations for faulty goods.
- Exemptions. Faulty or wrongly sent items should always be free to return. Build that into the reason-code logic.
- Loyalty. Many retailers make returns free for members or exchanges. That is a rule, not a separate system.
- Reputation. NRF's survey found about 71% of US consumers less likely to shop with a retailer again after a poor returns experience. A fee that comes with a clumsy process is worse than either alone.
Build, configure or buy?
There is a spectrum, and the right point on it depends on volume and complexity.
| Approach | Good fit when | Watch out for |
|---|---|---|
| Storefront app or returns SaaS | A single storefront, standard policies, modest volume | Limited warehouse workflow; rules that stop where the app stops; per-return fees that scale with volume |
| Configured platform plus integrations | Several channels or a 3PL, some bespoke rules | Integration upkeep; data split across tools |
| Custom returns hub | Complex products (warranty, repair, serial numbers), multiple warehouses, specialist compliance, or returns as a differentiator | Needs product ownership; higher upfront scoping |
A reasonable rule of thumb: start with the lightest option that handles your reason codes, your legal obligations and your warehouse flow. Move up when you hit a concrete limit, such as being unable to model repair-or-replace for warranty items, or being unable to see the full cost of a return in one place. If you are also deciding on your wider commerce platform, the analysis in SaaS ecommerce platforms versus custom development and the headless commerce guide for UK and Australian retailers are good companions.
Syslabs builds custom software for ecommerce teams, including returns and warranty tooling; the overview is on the Returns & Warranty Hub page. Pricing for any build is scoped on a consultation, because it depends almost entirely on your rules, integrations and volumes.
Integration checklist
Whatever route you take, returns touch many systems. Map these early:
- Storefront and order data: order lines, delivery dates, discounts applied (needed to calculate refunds on partial returns).
- Carrier and drop-off APIs: label generation, tracking events, collection booking. Our guide to API development and integration explains how to structure these connections.
- Warehouse management or 3PL: expected-returns notices in, inspection results out.
- Payment provider: refund calls, partial refunds, failure webhooks.
- Accounting: refund postings, tax treatment (VAT in the UK, GST in Australia) confirmed with your accountant.
- Customer communications: status emails or SMS, triggered by RMA state changes.
- Analytics: reason codes, cycle times, cost per return, resale recovery.
Metrics worth tracking
You do not need a large dashboard. Start with six numbers and review them monthly:
- Return rate by category and by SKU
- Cost per return, split into the four stages above
- Time from request to refund, with the warehouse leg shown separately
- Share of returns resolved as exchange or credit rather than refund
- Resale recovery rate (what proportion of returned items goes back on sale as new or open-box)
- Share of returns flagged for inspection, and the share of those that were truly problematic
Treat the last number carefully. If almost every flagged return turns out to be fine, your rules are creating friction for good customers.
Common mistakes
Treating policy and system as the same thing. A policy is a decision; the system should let you change it quickly. If a policy change needs a developer, the architecture is too rigid.
Free-text return reasons. They are unreadable at scale, and they blur legal remedies.
Refunding on scan, not on inspection. Some retailers refund as soon as the carrier scans the parcel to reduce customer anxiety. That can be a deliberate trade-off for trusted customers, but make it a rule with limits, not the default for everyone.
Ignoring the warehouse user. The people at the returns bench decide your real cycle time. Test the flow with them before launch.
No feedback to merchandising. If product teams never see reason-code data, you are paying twice: once for the return and once for the next customer who gets the same misleading page.
A phased approach for the first 90 days
You do not need to build everything at once. A realistic sequence:
Weeks 1 to 3: measure. Trace 200 returns, define reason codes, and agree who owns the returns P&L. Write your UK and Australian rules down in plain English so they can later become configuration.
Weeks 4 to 8: structure. Introduce the RMA record and a customer-facing return request flow with structured reasons. Connect the carrier labels. Keep the warehouse on its existing process but make it scan the RMA.
Weeks 9 to 12: automate. Add the disposition workflow, event-driven refunds, and the first version of the reason-code report for merchandising. Only then consider experiments such as exchange-first prompts or paid-return trials.
At each step, aim for something you can measure against the baseline you set in the first three weeks.
Conclusion
Returns will not disappear, and a generous policy is a legitimate part of competing online. What separates profitable retailers from the rest is whether they can see the cost of each return, apply their rules consistently, move stock back into sale quickly, and learn from the reasons. That is an architecture problem more than a customer-service one: a rules engine, an RMA record, a warehouse workflow, an event-driven refund step and a data loop.
If you are weighing up whether to configure what you have or build something purpose-fit for your range and markets, a short working session to walk through your last few hundred returns is a good place to start.
Book a consultation: Choose a time that suits you at syslabs.in/book/meeting. The booking page shows availability in your own time zone.
Sources
- National Retail Federation and Happy Returns (15 October 2025). Consumers expected to return nearly $850 billion in merchandise in 2025.
- Retail Times (25 September 2025). UK retailers risk loyalty loss with 78% of returns ending in refunds (Loop Returns 2025 State of Ecommerce Returns Report).
- Retail Gazette (6 October 2026). The returns paradox.
- Ingrid (19 February 2026). Charging for returns? What UK retail data shows in 2026.
- GOV.UK. Accepting returns and giving refunds: the law.
- ACCC. Repair, replace, refund, cancel.



