Acme Coffee came to us with a familiar problem: strong brand, good product, and a subscription funnel that leaked at every step. In six weeks we lifted their subscribe-to-first-order conversion rate from 2.1% to 2.8% — a 34% relative improvement worth roughly £220k in annualised revenue on their traffic base.
Here is what actually worked, and what did not.
The starting picture
Before we touched anything, the funnel looked like this: 100 visitors hit the homepage, 38 clicked through to the shop, 12 added a bag to cart, 4 reached checkout, and 2.1 completed a subscription. The two biggest drop-offs — homepage-to-shop, and cart-to-checkout — pointed to two very different problems and needed very different fixes.
Experiment 1: single hero CTA (winner)
The homepage had three equally-weighted CTAs: "Shop coffee", "Take the roast quiz", and "Subscribe & save". We killed the quiz and demoted "Shop" to a text link, making "Subscribe & save 20%" the single visual anchor. Homepage-to-shop conversion did not move much, but subscription-attributed sessions rose 21% within a week — visitors who wanted a bag still found the shop, and the subscribers self-selected earlier.
Experiment 2: rating badge above the fold (winner)
The brand had 4.8 stars from 3,200 reviews and buried it in the footer. We surfaced a compact ★★★★★ 4.8 · 3.2k reviews chip directly under the headline. Add-to-cart from homepage improved 11%. This is the cheapest experiment we ran and returned more than any other single change.
Experiment 3: reduce roast-preference friction (winner)
Step two of the subscription form asked new customers to pick their preferred roast, grind size and delivery cadence in one screen. Drop-off at that step was 41%. We deferred roast and grind to a post-purchase email ("we'll dial this in for your first bag") and kept only cadence at checkout. Checkout completion rose from 55% to 71%.
Experiment 4: urgency banner (loser)
We tested a "12 people are looking at this bag right now" banner. Conversion moved by less than 0.2% and the brand hated how it felt. Killed after eight days.
Experiment 5: video hero (loser)
A polished 12-second brand film replaced the static hero image. LCP went from 1.9s to 3.4s on mobile, mobile bounce rose 6%, and add-to-cart fell 4%. Reverted.
The framework we used to decide
Every experiment had to answer three questions before we shipped it: what step of the funnel does this move, what is the minimum lift that would make it worthwhile, and how will we know within two weeks. The banner failed the second question. The video failed the third. Everything else earned its place.