The voltage bug: 18% fewer appliance returns at checkout

An appliance store cut returns by 18% by making one field required. Shoppers were buying discounted low-voltage appliances for homes wired to the other nominal voltage, and the voltage was optional text on the page. The fix wasn’t a better warning. It was removing the option to skip the choice.

The 18% is the drop in the number of appliance returns, measured right after the change.

The report that started it

The data team at Stefanini Brasil didn’t bring a chart. They brought a pattern. Returns were concentrated in two groups, and neither group had anything to do with the quality of the product:

low-voltage appliances shipped to one group of states
high-voltage appliances shipped to another

The problem was in the map, not in the product.

Brazil doesn’t run on one voltage

Inmetro, Brazil’s metrology and standards institute (in Portuguese) and the Brazilian distribution rules describe a country that runs on two nominal voltages, 127 V and 220 V, and the two coexist inside the same region. There is no line that assigns one voltage to a set of states and the other to the rest. Two shoppers on the same street can have different outlets.

The same refrigerator is correct for one outlet and useless in the other. A shopper who orders the wrong model receives a machine that won’t run on the outlet at home. Nothing in the box is defective. The choice was wrong.

What the returned orders had in common

Every returned order carried two traits:

it was a promotion item, bought for the price
the voltage was never chosen, only read

The old page treated voltage as information. It sat under the price, in the same visual weight as the warranty line, and the checkout accepted the order without it. Someone chasing a discount doesn’t stop to check the outlet at home. They click the cheaper model and the wrong one shows up.

The fix: a required choice before checkout

In code, the change was small. In behavior, it was big. Voltage stopped being text and became a required field in the order. No choice, no checkout.

Three rules shaped it:

the choice happens before payment, on the same step as the item
neither voltage comes pre-selected, because a default repeats the same mistake with one extra click
the chosen voltage appears in the order summary, so the person reads it once more before paying

The NN/g article on form design states that a required field must be labeled as required, because the reader can’t tell a required field from an optional one by looking. The same research on product pages says the page has to explain the options the shopper has to choose between. The two support the rule directly: the choice is stated, and its obligation is stated.

Visible information asks for attention. A required field asks for a decision. That distinction is my reading of the flow, not a finding from the research. Where the rest of the purchase decision breaks is covered in ecommerce checkout.

The A/B test that validated it

The change didn’t ship just because the argument sounded good. It was tested: one version with voltage as text, another with voltage as a required field, both live at the same time on the same traffic.

The test measured behavior, not opinion. The question wasn’t whether people found the field annoying. The question was whether the order carried the right voltage and whether the appliance came back. The PicPay case shows the same method, with a single input running against a stepped form.

The result

Appliance returns fell 18%. Refund and logistics costs fell with them, because a returned appliance costs more than the refund. You pay shipping both ways, handling, and end up with a unit you can no longer sell as new.

The number is the small part of the lesson. The useful part is that a return is often a decision taken in the wrong place in the flow. NN/g calls a decision that adds work without adding value sludge, and a return is that work arriving late, after the money moved.

The same pattern applies to size and color

Voltage is one variant among many. Size in clothing, color in furniture, compatibility in electronics: each one is a choice the store can leave as text or turn into a required step.

The rule I use: when the wrong choice creates a return, the choice is required. When the wrong choice only creates a follow-up question, text is enough.

That rule is mine, not a published finding. It links a form decision to a real cost, and that cost is what justifies making the field required.

Where this post disagrees with the research

Two NN/g articles point the other way, and that difference is worth showing.

Pre-selecting a default. The article on defaults recommends pre-populating a field with the most common value as good practice, with the caveat of never defaulting to the most expensive option. This page does the opposite: neither voltage comes marked. The reconciliation is that the default rule assumes one option is genuinely more common. Here the choice is binary and high-impact, so no valid most-common value exists. Any pre-selection would be a coin flip the shopper inherits.

Removing decisions. The article on sludge decisions tells designers to remove decisions that only add friction. This page goes the other way and turns an optional field into an obligatory decision. The reconciliation is that the decision is necessary because the wrong answer produces a physical return. It isn’t friction added to a step that already worked. It is a decision moved to the step where it can still change the outcome.

How to find this in your own funnel

Read the returns report by variant, not by product name. A return rate that follows an outlet rather than a model is a usability problem, not a quality problem.
Open the funnel and find the step where the variant is chosen. The BTG case shows that step measured with Hotjar and Google Analytics.
Test the change as behavior. Two versions live at once, one metric to decide, the same method the PicPay case documents.

The storefront side of the same problem, catalogue and content architecture, is in the Le Biscuit case. The 18% is listed with the rest of the work at Stefanini Brasil in the full profile.

See how the A/B was built in the PicPay case.