How AI Can Help Retailers Recover More Value from Returned Products

Category

Blog

Author

Wissen Technology Team

Date

November 20, 2026

According to McKinsey, US consumers are spending nearly $1 trillion on merchandise in 2024 - which is more than double what it was just four years ago. On top of that, retailers are spending an estimated $200 billion a year trying to get some value back from their goods. However, the question is what happens to this specific product?

When a product is returned, there are 5 options:

  • Restock
  • Repair
  • Refurbish
  • Resell
  • Remove it from circulation

Manual processes and disjointed systems can create delays when making the correct decision. Using AI can help improve the decisions made by using the technology at 4 stages:

  • Condition assessment
  • Return classification
  • Fraud screening
  • Disposition recommendations

These are covered in detail below, as well as what is required to implement this.

1. Where Returned Value Gets Lost

Disposition Is the Hard Part

The same McKinsey surveyed 30 supply chain executives. Key findings were:

  • Over 50% said disposition was the biggest challenge they faced.
  • Most of the cost of a return is associated with disposition.
  • All but 5 said that their company has very basic or no data to make a decision.

Traditionally, this has been a manual process. Inspect the product and decide if it can be recycled, repaired, or restocked. This can be time-consuming. (Note: Sample size is limited, so these figures should not be seen as a benchmark.)

The Cost Of Delay

This can be seen in the example below. McKinsey looks at a $100 Christmas jumper returned in early December.

Returned by

Path

Recovery

Unknown customer

Central facility, standard inspection and repackaging, then a discount partner in January

About 50%

Trusted customer

Portal predicts no refurbishing, so a nearby store resells it within days

Around 75%

The second option shows how the value can be recovered by getting the product back into the market quickly whilst demand is still high.

2. How AI Can Help

Product Condition Assessment

AI can look at previous defects and match them to what customers are reporting. This can help improve the decision-making around product conditions, allowing products to be prioritised for refurbishment/recycling.

Looking slightly ahead, McKinsey also talks about how the future could be that electronic products have sensor/digital tags to capture how the product has been used and its condition. This would mean returned devices could be automatically assessed and put back onto the market. This is not commonly done yet; however, the benefit will depend on the type of product.

It is important to understand that AI will not completely remove human decision-making.

This article focuses on decisions and design considerations, rather than guaranteed outcomes. The 50% and 75% figures are just examples. The outcome of these will depend on your category, customer base, and network. A good way to start would be to look at one category of product and see if you can increase your recovery.

Return Classification

To get the most benefit from classification, this should be done as early as possible. Often retailers will have all the data needed to classify the product:

  • Margin profile & seasonality
  • Shelf life & resellability
  • Delivery & inspection costs
  • Store locations & stock capacity

Using this in a model can help classify the product and ensure the correct route is taken. The earlier this can be done, the fewer extra steps will be needed.

Fraud Detection

According to a survey by the National Retail Federation, 93% of retailers see returns fraud/exploitation as a major issue. On average, 1 in every 4 transactions will contain a bracketed product. (Product bought in multiple variations, to be returned.)

Using predictive analytics can help identify high-risk behaviours based on previous returns and order patterns. However, this can be too strict for loyal customers. So thresholds will need to be set manually.

Disposition Recommendations

This is the final element of the process. McKinsey talks about having “one decision engine” which will route the product to the best option for value in real time. The model takes into account demand, product condition, and logistics costs. Products can be restocked whilst there is demand. Products to be refurbished will not have to go through default locations. The outcome can be logged and used to train the model.

3. Integration & Controls

The real issue is that many retailers do not have fully integrated systems. Many still use stand-alone disposition systems. There are 3 key things to consider to gain the benefit:

Connect the Data First

  • Integrate customer, product, and supply chain data
  • Create rules to help increase speed, margins, and capabilities
  • Change processes to automate decision-making.

Having an integrated returns data product will help with this. This will have clear owners and governance. This may include variables such as returns history, SKU, reason code, expected value, and cost.

Match Policy to the Customer

Not all customers need the same return policies. According to the survey, 71% of consumers would be less likely to return to a retailer with a policy based on the product and customer. This can help segment customers based on their expected LTV and previous return behaviour. Try experimenting with:

  • Fees
  • Refunds
  • Return periods

Train the models on the results.

Set Controls & Ownership

There should be controls:

  • Limit refunds, fees, and return periods to avoid losing margin.
  • Ensure any experimentation is still within consumer protection, payments, and data protection laws. These will vary depending on the region.
  • Assign an owner. McKinsey suggests assigning a senior leader to be responsible for the returns P&L. They also suggest having regular cross-functional reviews to look at returns data.

Limits Worth Remembering

AI does not remove human decision-making. This article focuses on decisions and design considerations. The 50 and 75% figures are based on an example. The outcome will vary depending on your categories, customers, and network. A good place to start would be to look at 1 category of product and see if you can increase your recovery.

Working With Wissen

We work with retailers to help set up operational models. We provide a quick, simple, and flexible solution that balances automation with a human-centred shipping experience. We focus on:

  • An integrated and customer-centric model. Improving the speed of returns across all retail channels.
  • Real-time data. Using AI to provide insights and automate decision-making.

If you want to move away from manual reviews and improve the speed and knowledge of decisions being made, our technologists can help set up the correct foundations. Contact us to discuss your returns and retail operations.

FAQs

1. How does AI improve condition assessment?

AI compares defect history with reported return reasons to improve condition checks and route items to refurbish/recycle sooner; accuracy depends on data quality (McKinsey gives no figures).

2. How does AI detect returns fraud?

Predictive analytics flag risky return behavior (e.g., bracketing) using return history; flagged cases still need human review to avoid friction for loyal customers.

3. How does AI choose to restock, refurbish, or liquidate?

A decision engine weighs demand, condition, logistics cost, margin, and seasonality to route each return to the most cost-efficient endpoint at return start.

4. What systems are needed?

Connected customer/product/supply-chain data; a governed unified returns data product; rules balancing speed, margin, and capacity; automated early workflows.

5. Can humans override AI decisions?

McKinsey doesn’t detail overrides; use guardrails (refund/fee/window caps), P&L accountability, and human review - especially for fraud flags.

‍