Discount Discovery Statistics from Two Retail Field Experiments

Updated 26 source citationsSources checked 12 November 2026

On this page
  1. Quick answer
  2. Key Takeaways
  3. Can presentation change purchasing when prices stay fixed?
  4. Who participated in the first experiment?
  5. How do item outcomes differ from visitor outcomes?
  6. What did the larger experiment test separately?
  7. Which shoppers might respond differently?
  8. What should remain fixed in a merchant experiment?
  9. What does this evidence imply for a blanket discount rule?
  10. Which reported measures support these findings?
  11. Did the interface changes move transaction rate and product mix together?
  12. Methodology and Sources
Download study text

Quick answer

Two experiments at a Philippine fashion retailer found that changing discount-discovery interfaces affected what shoppers bought even though prices and assortment stayed fixed. In the larger experiment, the control recorded 1,351 transactions among 68,343 visitors, while a non-discount-banner treatment recorded 1,713 among 69,825. Those counts imply rates of 1.9768% and 2.4533%, a gap of about 0.4765 percentage points. The result belongs to the studied retailer and its presentation changes (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).

Discount presentation can change which products people inspect even when prices remain fixed. This analysis distinguishes the interface interventions, eligibility rules and outcome denominators in the two experiments. It also examines why a result among purchased items cannot stand in for the effect on every visitor, a distinction that matters when testing a recovery offer or changing a shopping interface.

Key Takeaways

  • The first experiment held prices and assortment fixed The 2015 experiment randomized 104,605 new visitors across a control and three discount-discovery interfaces while keeping product prices and assortment the same (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).
  • The full-price purchased-item share rose 7.71 points in one arm The no-outlet-link arm recorded 67.48% of purchased items at full price, compared with 59.77% in control, a 7.71 percentage-point gap (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).
  • The second experiment included 348,110 visitors The second randomized experiment assigned 348,110 new and returning visitors across five conditions in 2016 (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).
  • Exact transaction counts imply a 0.4765-point rate gap Table 3 reports 1,351 transactions among 68,343 control visitors and 1,713 among 69,825 non-discount-banner visitors, implying rates of 1.9768% and 2.4533% (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).

Can presentation change purchasing when prices stay fixed?

A discounted item can be easy to find through a banner, a dedicated section, a sorting control or a marker on its listing. Each route changes what a shopper encounters first. It can also change which alternatives they consider before deciding to buy.

Discount discovery statistics compare outlet-link removal with non-discount banners: both transaction rates exceed control, but the full-price purchased-item share difference is clearer for banners.
Figure 1: Two higher transaction rates do not establish the same product-mix change. Source: Ngwe, Ferreira and Teixeira, Table 3, source checked Nov 2026. Original analysis of the cited source. Source periods and calculation limits appear in the text.

The experiments by Donald Ngwe, Kris Ferreira and Thales Teixeira varied those routes at a fashion retailer in the Philippines. They held product prices and the available assortment constant across conditions. The research therefore examined how discount discovery and product exploration affected behavior within that setting.

This is a different intervention from giving an abandoned shopper a new coupon. The comparison is useful because it shows why a result about offer size cannot automatically answer a question about offer visibility.

Who participated in the first experiment?

The first experiment ran over nine days in June 2015. It included new visitors arriving through the main landing page on desktop, laptop or tablet devices. Visitors arriving through email coupons, newsletters or third-party links were excluded from this comparison (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).

The control group retained the usual site. The three treatments removed a direct outlet link, removed discount sorting and markers together, or combined those changes. Discounted products remained available through other routes. The experiment changed the effort needed to find them.

Purchased-item outcomes moved substantially. The control’s average discount was 15.74%, compared with 11.54% in the no-outlet-link arm. The share of purchased items sold at full price was 59.77% in control and 67.48% in that treatment, a 7.71 percentage-point difference (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).

How do item outcomes differ from visitor outcomes?

An average discount among purchased items describes the orders that happened. A transaction rate describes the assigned visitors. Those are different denominators, and the first can change partly because a treatment changes who purchases or which items they choose.

The source explicitly notes this selection issue for the purchased-item measures. A merchant should retain that caveat when comparing average selling prices, discounts or the full-price share. A higher price among buyers does not by itself establish a higher expected return from every visitor.

For the same reason, an evaluation of checkout recovery should include both the fraction of eligible shoppers who buy and the contribution from those purchases. The offer that produces the highest order value among a small number of buyers may have a different total effect.

What did the larger experiment test separately?

The second experiment included 348,110 new and returning visitors over two weeks in 2016. It separated the removal of outlet links, discount markers and discount sorting, and added a fourth condition in which discount-oriented banners were replaced with other banners (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).

The published control recorded 1,351 transactions among 68,343 visitors. The non-discount-banner arm recorded 1,713 among 69,825. Using the exact counts produces rates of 1.9768% and 2.4533%, respectively. The difference is about 0.4765 percentage points. The other three arms also had higher reported transaction rates than control in this experiment (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).

Those values are observed results from one particular retailer. They should not be used as a forecast of what would happen if another store removed a discount link. The retailer’s assortment, existing customers and distinctive outlet pricing were part of the setting.

Which shoppers might respond differently?

The researchers examined differences related to price sensitivity and prior customer information. Their account of the mechanism is that some shoppers will undertake extra search to find discounts, while others will buy an acceptable full-priced item they encounter earlier.

For a merchant, the useful question is which shopper problem the presentation is solving. Someone who has already said that the price is beyond their budget may need a relevant lower-priced option. Someone uncertain about product suitability may benefit more from an explanation of the match.

A recovery test can examine those situations without assuming that every abandoned cart needs a stronger discount. Keep actual prices, eligibility and offer conditions clear, and compare explanations that address a documented objection. That is a testable extension of the research, not a measured result of these experiments.

What should remain fixed in a merchant experiment?

Define the eligible shopper group, preserve assignment through return visits and record the exact experience each version changes. If the question concerns how an existing offer is explained, avoid simultaneously changing its value and availability. Otherwise the result will be difficult to attribute.

Measure purchases per eligible shopper alongside discount cost, order contribution and product mix. Record whether the shopper found the option they asked for. A short-term revenue difference is easier to judge when the underlying experience and its consequences are visible.

The source also acknowledges that changing interface elements can affect features beyond search effort. A banner is both information and a navigation route. Even a well-run field experiment needs care when assigning a single psychological explanation to its result.

What does this evidence imply for a blanket discount rule?

The study challenges the idea that making every discounted item immediately prominent must always maximize purchasing. It does not establish the opposite rule. The comparison depends on the retailer, the customer mix and the unchanged underlying prices.

The most useful application is a sharper diagnosis of the abandoned decision. Separate a price objection from a product-selection problem, then test a clear and appropriate response. The published tables provide a concrete example of how different outcomes can move together, while the merchant’s own experiment determines whether that pattern transfers.

Which reported measures support these findings?

The figures below preserve the source’s outcome labels and units. Read comparisons within their stated experiment, population and period.

Reported measureValue and unitSource
First experiment shoppers104605 visitors(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Second experiment shoppers348110 visitors(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment I control average item discount15.74 percent(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment I no-outlet-link average item discount11.54 percent(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment I control full-price share59.77 percent of purchased items(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment I no-outlet-link full-price share67.48 percent of purchased items(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II control visitors68343 visitors(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II control purchases1351 transactions(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-discount-banner visitors69825 visitors(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-discount-banner purchases1713 transactions(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-outlet-link purchases1599 transactions(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-discount-marker purchases1605 transactions(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-discount-sort purchases1605 transactions(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-outlet-link visitors70058 visitors(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-discount-marker visitors70025 visitors(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)
Experiment II no-discount-sort visitors69859 visitors(Ngwe, Ferreira and Teixeira, source checked, Nov 2026)

How are the additional calculations reproduced?

These are arithmetic comparisons of the published inputs. They retain the same sample and limitations as the inputs and do not create an additional experiment.

CalculationExpressionResult and unit
Control reported transaction rate1351 / 68343 * 1001.97679 %
No outlet links reported transaction rate1599 / 70058 * 1002.28239 %
No discount markers reported transaction rate1605 / 70025 * 1002.29204 %
No discount sorting reported transaction rate1605 / 69859 * 1002.29748 %
Non-discount banners reported transaction rate1713 / 69825 * 1002.45328 %
No-banner versus control rate gap1713 / 69825 * 100 - 1351 / 68343 * 1000.47648 percentage points
First-experiment full-price share gap67.48 - 59.777.71 percentage points

Did the interface changes move transaction rate and product mix together?

The second experiment's Table 3 gives a more specific answer than a ranking of conversion rates. Replacing discount banners raised the full-price share of purchased items from 48.83% to 52.58%, a 3.75-point difference, and lowered the mean purchased-item discount from 18.25% to 16.80%. Both displayed comparisons are marked significant at the 1% level. Removing outlet links raised the full-price share only to 50.15%, with no significance mark for that share; its mean discount of 17.32% is marked only at the 10% level. Both arms nevertheless had higher displayed transaction rates than control. (Ngwe, Ferreira and Teixeira, Table 3, source checked Nov 2026).

Second-experiment armTransactions per 100 assigned visitorsFull-price share of purchased itemsMean purchased-item discount
Control1.976848.83%18.25%
No outlet links2.282450.15%17.32%
Non-discount banners2.453352.58%16.80%

This distinction prevents a broad mechanism claim. More transactions under two interfaces do not imply that the interfaces changed discount discovery or product mix in the same way. The source itself describes the outlet-link treatment as an exception in its second-experiment discussion. Purchased-item averages also remain selected outcomes, so the table does not show that every otherwise identical shopper shifted to a full-price product.

What economic change could offset the higher transaction rate?

With all other per-visitor costs held fixed, the banner arm's transaction-rate ratio is about 1.2410 relative to control. A break-even identity therefore requires average contribution per transaction to remain above about 80.58% of control. A decline exceeding about 19.42% could erase that transaction-rate advantage. This is an original accounting threshold from the source counts, not a measured margin result or a claim that costs actually changed.

The same calculation does not license multiplying the purchased-item full-price share by a headline margin. Products can differ in cost, items per transaction and return behavior. Prices remained fixed across the experimental conditions, but the mix of purchased products could still change economics. A merchant needs realized contribution on the original visitor denominator.

Which local comparison would test the actual decision?

If the shopper needs a lower-priced suitable item, preserve an accessible path to it. If the issue is that discount banners distract from selecting an appropriate product, compare a different presentation while holding genuine prices and eligibility fixed. The paper explicitly notes that banners are both information and navigation, so removing one changes more than an abstract search-friction parameter.

The transaction and mix accounting and calculation retain each denominator and source significance label. The source table labels transactions as conversion; without person-level records, this analysis adds no new independent-binomial test or between-treatment winner claim. The largest displayed rate is a candidate for another comparison, not a universal instruction to hide offers.

Methodology and Sources

The comparison uses the original HBS working paper’s two randomized field experiments. Exact transaction rates are recalculated from Table 3 arm counts, while item-level discount and full-price-share outcomes retain their purchased-item denominators. Source treatment definitions, visitor eligibility and historical context are kept visible. No changes in prices, assortment or a current merchant’s conversion are inferred (Ngwe, Ferreira and Teixeira, source checked, Nov 2026).

Ngwe, Ferreira and Teixeira: 2019 working paper; experiments in 2015 and 2016. Citation months identify the cited publication or version. Where a citation explicitly says “source checked”, November 2026 is the verification month, and the original study period remains stated in the text. Source observations, calculations and the proposed merchant applications have distinct meanings.

One dominant Philippine fashion retailer with distinctive outlet pricing and no equivalent discounts from competing sellers.

Different arms have slightly different sample counts; rates use each arm’s own denominator.

Glossary

  • Search friction: Effort required to find or compare available products.
  • Assignment: The experimental condition a visitor was allocated to.
  • Treatment: The tested change in the interface.
  • Control: The comparison version of the interface.
  • Selection: Which items or shoppers appear in a measured outcome.

Sources

Cite this page

Use the citation below when referring to this analysis. The original source and calculation scope accompany each finding.

Said Maybe. "Discount Discovery Statistics from Two Retail Field Experiments." Said Maybe, November 2026. https://saidmaybe.com/guides/discount-discovery-statistics-from-two-retail-field-experiments/
<a href="https://saidmaybe.com/guides/discount-discovery-statistics-from-two-retail-field-experiments/">Discount Discovery Statistics from Two Retail Field Experiments</a> (Said Maybe, November 2026)