BEFORE
A broad introduction
- Meal preferences
- Dietary preferences
- Personal goals
- Meals of interest
- Cooking frequency
- People at the table
- Meal plan overview
SELECTED WORK / 01 · ACQUISITION & PERSONALIZATION
500 recipes.
Too much choice.
So we got personal.
More choice only works when people can find what they actually want. I helped rethink signup as the start of personalization — learning enough about each customer to make a larger menu feel smaller, more relevant, and distinctly theirs.
Explore the new experience ↓
Net activations · HelloFresh US mobile
Statistically significantProjected 52-week Customer Value Added
HelloFresh US mobile · Financial impact estimateMY CONTRIBUTION
Design direction · UX and content execution · Shared funnel framework
ROLLOUT
HelloFresh US · Mobile web & desktop
Primary success metric: net activations
01 / THE CHALLENGE
More recipes.
Without more overwhelm.
Project 500 was the internal name for expanding our recipe database to 500 recipes in early 2026. To make that assortment useful, we needed a much more detailed understanding of what each customer wanted to eat.
The old funnel collected broad preferences. We needed to learn enough about people’s tastes to recommend meals they would actually want.
The UX challenge was clear: how could we ask more questions without losing people’s appetite for the experience?
02 / REFRAME SIGNUP
Learn their tastes.
Make the menu feel like theirs.
Our bet: asking more specific questions would help us recommend a better first box—and turn more prospects into customers. The trade-off was a longer quiz. Every question had to earn its place.
THE JOURNEY, BEFORE AND AFTER
From broad preferences
to a detailed taste profile.
We chose to collect more specific taste signals before activation. The trade-off was a longer quiz; the intended payoff was a first box that already felt relevant. Goals came forward, followed by reassurance and more detailed preferences.
Compare the old and new signup steps
ZIP CODE CHECK In network → Continue to quizNot in network → Separate eligibility path
NOW
A richer understanding
- Personal goals
- Moment of assurance
- Meals of interest
- Dietary habits
- Ingredients to avoid
- Nutrition preferences
- Favourite cuisines
- Favourite flavours
- Protein preferences
- Dinner preferences
- Cooking frequency
- People at the table
The journey changed from a broad introduction to the inputs needed for personalized meal recommendations.
03 / MAKE IT COMPELLING
A little more about you.
A lot more to look forward to.
Goals, dietary needs, cuisines, flavours, proteins and dish types build a picture of the customer’s tastes. Reassurance connects those answers to what HelloFresh can offer, while household and frequency choices turn preferences into a practical plan.
Look closer at the details ↓THE CRAFT WITHIN THE JOURNEY
From filling in a form
to picturing dinner.
Each moment needed to earn its place in a longer quiz. Real food builds anticipation, reassurance makes the value tangible, and responsive imagery helps customers picture meals in their own lives.
01 / GOALS
Give the first question
some appetite.
The original screen led with a long list of motivations. The redesign puts colourful, plated food into the first impression, alongside a compact set of choices and a clear limit: pick up to three.
02 / A MOMENT OF ASSURANCE
Don’t just save the answer.
Make the benefit felt.
“We’re saving your goals” acknowledged an action. The redesigned moment makes a promise more tangible: big flavour, better value, less prep and more variety.
Close-up food motion brings texture and movement into the experience, while specific benefits give customers a reason to keep going.
03 / HOUSEHOLD
Bring everyone
to the table.
A static couple illustration suggested one kind of household. The new imagery centres on hands and meals: as the number changes, another plate joins the table.
The interaction makes portions visible without prescribing who a household should be. “Just tell us who’s hungry” keeps the tone human and connected to food.
04 / THE ROLLOUT DELIVERED
A more personal journey.
More activated customers.
+5.0%
HelloFresh US mobile
Net activations · statistically significant
+5.2%
HelloFresh US desktop
Net activations
+4.1%
Factor US
Net activations
Fewer people reached registration, but more became activated customers. For HelloFresh US mobile, the projected 52-week financial impact was +€7.5M in Customer Value Added (CVA), our North Star metric for the value of product changes. This is a projection, not realized revenue.
See the supporting rollout metrics
| Platform | Net activations | Gross activations | Early cancellations |
|---|---|---|---|
| Mobile web2026 rollout | +5.0%Statistically significant | +3.5%Statistically significant | Substantial decrease in 24h cancellations |
| DesktopMarch 5, 2026 | +5.2% | +4.1% | −18.7% 24h cancellations |
Mobile traffic reaching registration fell 15%, while net activations increased. We viewed this as a positive trade-off: fewer prospects reached that step, but more became activated customers.
Meal-choice rates for boxes 1 and 2 also decreased—by 2–10% on desktop—consistent with customers keeping more of their recommended meals. We interpreted that as a sign of better fit, rather than direct proof of trust.
FACTOR US / P500 ROLLOUT
+4.1% net activations · −11% in 24-hour cancellations.
05 / WHAT THE FRAMEWORK MADE POSSIBLE
A more personal way in.
The next step was to personalize the journey itself. An intelligent funnel adapts the signup experience to what matters to each customer, instead of sending everyone through the same flow. We created versions centred on Convenience, Quality, Health and Variety, alongside the original P500 funnel, so the model could learn which journey worked best for whom.
On HelloFresh US mobile, a machine-learning model chooses among those five journeys based on the customer’s context. It learns which routes lead to activation and adjusts over time.
CUSTOMER CONTEXTDifferent people.
Different motivations.
Different motivations.
CONTEXTUAL BANDIT MODELRoute → Observe → Learn
Selects among five funnel experiences
Live and learning. An ongoing comparison with the original funnel measures what ML routing adds. No ML conversion lift has been established yet; the results above belong to the P500 rollout.
MAKING INTELLIGENT FUNNELS REUSABLE
Share the behavior.
Leave room for local relevance.
As part of Intelligent Funnels, I built a component library so teams could create different journeys without rebuilding each one. Shared interactions stayed consistent, while imagery, tone, ingredients and claims could adapt to the market. The New Zealand Quality funnel shows how that works.
THE BOUNDARIES, ACROSS THE JOURNEY
Clear on what stays shared.
Flexible on what feels local.
We mapped out where shared interactions needed to stay consistent, and where imagery, tone, ingredients and claims could adapt to the market. This New Zealand Quality flow makes those choices visible.
Inside the component library · 9 funnel variations, 16 market views
My role spans 18 markets; this library snapshot covers 16 market views.
The library documents how to edit a market variant, compose a funnel and review a country’s content. Live and planned rollouts are shown separately.