Is your discount strategy costing you revenue?

When a testing dashboard tells you a pop-up succeeded, it usually just means more people checked out. It doesn't tell you what those orders actually cost the business.

That isn't a measurement problem, you just need to choose the right metric. Here is how to fix your testing framework so you stop giving away free margin:

1. Pick the metric that actually protects your business

To evaluate a test accurately, you need a metric that aligns with long-term value and is incredibly hard to game. Ronny Kohavi calls this the Overall Evaluation Criterion. Conversion rate fails this test completely because it always goes up when you drop your prices.

For most scaling brands, the honest metric to track is contribution per session. This is your revenue after discounts, minus COGS, minus returns, divided by sessions.

If COGS isn't visible in your analytics layer, use net revenue per session and hold your gross margin as a strict, unmoving guardrail.

2. Run the brutal arithmetic on your last discount test

Let's look at the classic 20% off welcome pop-up. Imagine a store doing 100,000 sessions a month at an £80 AOV, with COGS at 40%. Every standard order carries £48 of contribution.

Now, you turn the 20% off pop-up on. The conversion rate climbs, and you pick up a few hundred extra orders. But your AOV drops to £64, and your fixed costs haven't moved. Your contribution per order falls from £48 down to £32. You bought a few more orders, but you gave away a third of the margin on each one, including the orders from customers who were always going to buy at full price anyway.

3. Set strict numeric guardrails before you launch

Guardrails aren't tie-breakers, they're veto conditions with thresholds you set.

Before you launch a test, set clear thresholds for:

• Contribution per session: The metric that would have stopped the disastrous discount test above.
• AOV and units per order: To catch variants that just push people toward your cheapest SKU.
• Discount as a share of gross revenue: To tell you what the conversion lift actually cost you.
• Return rate by exposed cohort: Discounted cohorts naturally return more items. You must measure returns by the specific test cohort, not by the calendar month, otherwise, the signal gets buried in your normal traffic.

The Takeaway

If you are testing anything that touches discounts, urgency, or bundling, you should evaluate a 60-day cohort read. You have permanently altered that customer's reference price, and the dashboard won't warn you about it.

This week, re-read your last three successful tests against net revenue per session. Some of them won't survive the audit.

This week's deep dive was brought to you by Dayo Samuels, Rainy City Agency.

The Winter Experience at Somerset House

10th December 2026 | London | Invite-Only

From premium hospitality at the British Grand Prix to summer celebrations at Royal Ascot, we build environments where the leaders of UK eCommerce can actually connect.

To close out 2026, we have secured exclusive private access to the iconic ice rink at Somerset House. On December 10th, there are no panels and no presentations.

Here is what you can expect:

• Private Access: The entire ice rink to ourselves (with skates for anyone brave enough).
• Premium Hospitality: Exceptional dinner and drinks followed by the rink.
• Unfiltered Networking: Swap a year's worth of stories and strategies with a highly vetted room of founders, operators, and tech leaders.

If you want to celebrate the end of the year with us, register your interest now.

New on the blog

We're publishing weekly now, and two from the last fortnight are worth your time if Q4 planning is on your desk.

DTC brand news this quarter: what actually mattered. Meta now rewrites your ad copy in Advantage+ by default, up to eight variants, and it stays on unless you turn it off. Placement controls are being phased out, creative diversity scoring has appeared in Ads Manager, and Google's AI Mode has gone live in the UK. Worth ten minutes just to go and check your own settings.

Three-quarters into 2026: what to fix before Q4. What founders told us actually worked this year. Creative volume beating clever targeting, discovery moving to TikTok and AI search, and why the second purchase needs designing before Black Friday rather than after it. There's a short pre-peak checklist at the end.

In Other News…

ASOS says the next fight is guidance, not choice: ASOS is actively shifting its strategy away from endless product selection and moving toward curated edits. This pivot is backed by their recent research showing that 69% of shoppers find buying clothes online overwhelming, and a third have completely abandoned a purchase because they couldn't picture how to wear the item.

What this means for you: If your PLP is a wall of options with no point of view, there's an opening in styling the outfit or just telling people which one to buy.

Only one in ten UK shoppers trust AI recommendations: ThoughtSpot's YouGov survey of 2,216 UK adults found 10.6% trust AI product recommendations and 53% actively distrust them, mostly because 66% say AI has misread what they wanted.

What this means for you: the appetite is for accuracy and control. Getting your pricing bands, site navigation, and filtering architecture clear will build more trust.