Case study · ecommerce · revenue

Offers after checkout: Post-purchase upsells without the regret

An interactive pattern case study: post-purchase upsells can add revenue or create regret. How to design honest one-click offers and measure them properly.

Pattern analysis

Slide 1 of 10

The pattern: one more click after the order

Post-purchase offers appear after payment and before the confirmation page: "Add X to your order with one click." Because payment details are already entered, accepting is very easy.

That's exactly why they need care. Phil uses a composite shop, Glow Botanics, selling skincare.

Phil says: The easier it is to say yes, the more important it is that people know what they're saying yes to.

Slide 2 of 10

WXM 1 · Wonder (Look)

Wonder: the offer looks like a success

The current offer's dashboard looks great. The team looks further.

  1. A healthy share of customers accept the one-click offer.
  2. But refund requests for the add-on are far higher than for any other product.
  3. Support tickets: "I didn't mean to buy this", "I thought it was free".
Post-purchase offer: accepted vs refunded (composite example) Original wireframe, not a screenshot.
  1. Borrowed revenue. Part of the 'win' comes back as refunds and support cost.

Slide 3 of 10

WXM 1 · Wonder (Look)

Wonder: the offer today

A timer, a big yes, a tiny price and a guilt-trip no.

Control (A)
Post-purchase offer page, current version Original wireframe, not a screenshot.
  1. Fake urgency. The timer restarts for every customer.
  2. Hidden price. The charge is in small grey text under the button.
  3. Confirmshaming. Declining is made to feel silly.

Slide 4 of 10

WXM 1 · Wonder (Look)

Wonder: naming the patterns

Each problem on the page matches a known deceptive pattern. [Source](#src-deceptive-design-types)

  1. Fake urgency: a timer with no real deadline.
  2. Hidden costs / trick wording: price and charge not clear at the moment of choice.
  3. Confirmshaming: the decline link mocks the customer.
  4. Some versions of this pattern also pre-tick extras or sign people up to marketing. Pre-ticked boxes aren't valid consent under EU guidance. [Source](#src-edpb-consent-2020)

Slide 5 of 10

WXM 2 · Architect (Plan)

Architect: an offer that's worth accepting

Because the add-on is refunded far more than other products and support tickets mention accidental purchases, we believe a clear offer (relevant product, price in the button, equal yes and no, no timer) for all customers will increase net add-on revenue after refunds, because customers who accept will actually want it.

Your turn: What is the right primary metric for this test?
Show Phil’s answer

Net add-on revenue per order after refunds. Acceptance can drop and the business still gain: fewer refunds, fewer tickets, happier customers. Net revenue after refunds captures that.

Slide 6 of 10

WXM 2 · Architect (Plan)

Architect: how does it score on PXL?

Very noticeable on a page every buyer sees, removes elements, backed by refund data and support tickets, quick to change.

PXL score: 11 / 14

PXL criteria breakdown
Is the change above the fold?Yes1
Is it noticeable within 5 seconds?Yes2
Does it add or remove an element?Yes2
Is it designed to increase user motivation?No0
Does it run on high-traffic page(s)?Yes1
Does it address an issue found in user testing?No0
Does it address an issue found in qualitative feedback (surveys, polls, interviews)?Yes1
Does it address insights found in digital analytics?Yes1
Is it supported by mouse-tracking heatmaps or eye tracking?No0
Ease of implementationUnder 4 hours3

PXL by CXL. A high score ranks an idea; it does not predict a win.

Slide 7 of 10

WXM 3 · Validate (Test)

Validate: the variation

The offer is chosen from what's often bought with the order, the price is in the button, and both choices look like choices.

Control (A)
A: timer, hidden price, confirmshaming Original wireframe, not a screenshot.
Variation (B)
B: clear, relevant, equal choice Original wireframe, not a screenshot.
  1. Order first. The customer knows their order is safe before seeing any offer.
  2. Price in the button. No one can accept without seeing what they'll pay.

Slide 8 of 10

WXM 3 · Validate (Test)

Validate: the test plan, written before launch

Fixed in advance:

  1. Primary metric: net add-on revenue per order, after refunds within 30 days.
  2. Guardrails: support tickets about the add-on, refund rate on the main order, repeat purchase at 90 days.
  3. Expect: acceptance may fall. That's fine if net revenue and trust improve.
  4. Decision rule: ship if net revenue holds or improves and complaints fall.

Phil says: If the honest version earns less, the old version was earning money it shouldn't have.

Slide 9 of 10

WXM 4 · Expand (Grow)

Expand: what the pattern usually leads to

This is a pattern, not a report of one company's test. Typical next steps:

  1. Honest version wins: test which product to offer (refills vs travel sizes vs complementary items).
  2. Honest version earns less: keep it anyway. Count the refunds, tickets and trust the old one cost.
  3. Apply the same three checks (clear price, equal choice, no fake urgency) to every offer on the site.

Slide 10 of 10

Your turn

Look at every offer in your purchase flow. For each: is the price visible at the moment of choice? Is declining as easy as accepting? Is any urgency real? Then compare the offer's refund rate with your other products. Read ethical CRO.

What this pattern teaches

  • Wonder looked past the acceptance rate to refunds and support tickets.
  • Architect named each deceptive pattern and removed it, keeping a genuine offer.
  • Validate measured net revenue after refunds, accepting that acceptance might fall.
  • Expand applied the same checks to every offer on the site.

Glow Botanics is a composite store used to illustrate a common pattern.

Frequently asked questions

Are post-purchase upsells ethical?

They can be, when the offer is relevant, the price and what happens on click are clear, declining is as easy as accepting, and there's no fake urgency. They become a dark pattern when they rely on confusion or pressure.

Is this based on a real company?

No. It describes a pattern seen across many online shops, illustrated with a composite store. It doesn't attribute results to any company.

Sources (4)

  1. Types of deceptive pattern. Deceptive.design (Harry Brignull). Accessed 4 November 2026.
  2. WXM: the Wave Experimentation Model. Wavesy. Accessed 4 November 2026.
  3. PXL: A Better Way to Prioritize Your A/B Tests. CXL. Accessed 4 November 2026.

Learn the principles behind this case

Search lessons, guides, case studies and templates. Press Esc to close.