Thirty percent in 2020. Six when I picked it up.
Refer a friend, they get a free box and you get credit: the same give-and-get trade every bank and gym and phone company runs. In 2020 it brought HelloFresh 30% of its new customers.
By the time I picked it up it brought 6%. It still cost $12.7M in rewards in 2025, and the number of referrals was falling about 20% a year. The chart below is how we knew. Fewer customers reached the referral page each year, in every region.
Leadership set a Q2 goal to turn the program around and gave it to the Share Squad. I was the squad’s Staff Product Designer, working with my PM. Strategy through to the design of everything a sender and a friend would touch.
We were not starting blind. A study from before this work had already found that people refer for the person, not the reward. The program was still built around the reward.
Visited the referral page
17.83% 10.45% −41%
Share of active customers · North America highlighted
The page was a problem. So was everything after it.
I followed the offer from both ends.
You send “a free box” with no insight as to what’s inside. How many meals? For how many people? Is there a catch? I genuinely had no idea.
Your friend clicks and lands in a twenty-one step sign-up. Plan, household, address, delivery day, and at step twenty-one a card. Only after paying do they see a single meal. The menu, the one thing that might have sold them, is behind the paywall.
I sent a box to my mom. Turns out her box wasn’t free, and she ended up on a weekly subscription. I only found out a week later when she mentioned it to me.
People weren’t holding back because the reward was too small. They were afraid of looking bad in front of someone they liked.
Your friend lands here. Twenty-one steps later. A credit card is needed and they still haven’t seen the menu yet.
Friend lands here
Card required. Menu afterSharing a moment, not a code.
Ten weeks, strategy through to launch in the US and Canada.
We stopped leading with the money. The discount stayed, it just wasn’t the headline any more. What led instead was the person: who sent you this, and what they had given you.
Two fixes. One for the friend who lands, one for the customer who sends.
The friend got a landing page which hadn’t previously existed. I designed the friend’s side end to end and handed it to the Growth Squad, who built it. It opens on who sent you the box and not the offer. The sign-up behind it went to four steps. The free box is applied before checkout, and with Apple Pay you never type a card at all.
Fewer referrals, better ones.
The sender’s fix was about timing. Internally we called them moments of joy. A box on the porch, the first or the fifteenth, and a week of meals is covered. A meal the family loved, just rated. The week’s meals, just picked.
We put the invite in those three moments and let the share happen there, rather than send a customer off to another page. That reversed the old strategy, which had been to add popups to other teams’ surfaces. Those bought impressions and noise. Fewer prompts, better placed: the one that appears right after the week’s meals are picked went from under a hundred invites a week to over six hundred.
The number I watch is a small one. Invites sent to one named person went from 13% to 23% in a single quarter. That is the behaviour the whole argument rests on, and it nearly doubled.
Widening the funnel cost us at first. The share of invites that became customers dipped and has not fully recovered. But by late August the US was adding 1,658 new customers a week against a Q2 average of 1,318, above where we started for the first time, and 69% ahead of the goal leadership had set.
Two of the three moments, across three surfaces. The invite right after the week’s meals are picked, on web and on mobile web. And the invite when a box arrives.
Then we were asked to put the offer back.
It went to a wider review and the answer came back. We had moved too far from the offer. Put the mechanics back in front.
That was a hard note to take. It is also fair. The offer is the part of this the business can actually set a price on and forecast, and a warmer message is a lot harder to hold anyone accountable to.
So the question changed. Not whether to show an offer, but what an offer could be. What went back was not the old banner. It was one component with three states, and the business picks the state for the month.
The plain credit gained a progress track, so a sender is always part-way to something. The reward puts a prize at the end of that track in place of the credit, a late change to keep acquisition cost flat. The raffle turns every referral into an entry for something bigger. Same frame, same button, and the customer who has seen one has seen all three.
All three have run, one at a time. The raffle, run in the US, pushed invites up 10.7% and cost more than it made back. The reward, run in Germany, barely moved volume, but it lifted conversion 15.7% and turned a small profit.
Built for one brand. Inherited by the group.
The component is the part that travels. The landing page, the four-step funnel and the three offer states were built for HelloFresh in the US and Canada. The program has since gone to Germany, with more markets following on the same configuration, and Factor inherits the whole model next.
The next round of mechanics went through the same test before any of it was built. My PM and I took the roadmap’s eight ideas, a leaderboard, a tiered ladder and a public challenge among them, and argued each one against the research. Most of them pay the people who were going to refer anyway, or put the sender’s self-interest next to a friend’s name, which is the one thing the study said stops a send.
Two survived. Sending a box for an occasion, a new baby or a move, where the reason to give already exists. And invites you earn with tenure, so what you hand over is something you were given rather than a coupon. Leadership approved the cut.
What we didn’t fix.
We rebuilt how the program looks and feels. We did not change its economics. A deep-dive by the analytics team, which landed after launch, found that 52% of referred friends were individually value-negative, and that the top tenth carried 57% of the value. The reward still loses money on our most loyal Canadian customers, and that needs a pricing fix rather than a design one.
One reporting flow still cannot tie a new customer back to whoever referred them. Both are on the list.
If I picked this up again tomorrow I would start with the box. It is the only thing we send that has a 100% open rate. There is already a share button on the recipe page and it gets around 3,750 shares a week. The two audiences overlap by 1.7%.