Case study
The business had outgrown the model it was sold through.
How I led Product and Design through the rearchitecture of Purple Carrot’s core digital commerce experience: from plan first to shop first.
Chapter 01
The business had changed.
Purple Carrot began with a focused offer: plant-based meal kits, chosen through a plan.
Over time the offer widened. Ready-to-eat meals. Grocery and add-ons. Premium kits. Limited-time offers and partnerships. Flexible carts. Meals matched to dietary goals.
Every one of them still had to enter through the plan.
The business evolved. The product model didn’t evolve with it.
Chapter 02
The experience still reflected the old model.
A new customer arriving from an ad landed on the homepage, then went to plans: how many meals a week, how many servings.
They could browse the weekly menu. Browsing still led back to plans and sign-up.
Choosing the actual food came after subscribing, inside the logged-in dashboard and shopping page.
Customers had to understand how Purple Carrot worked before they could experience why they wanted it.
Chapter 03
We listened before we redesigned.
The direction didn’t start with a preferred interface. It started with a standing research practice: customer interviews, support and cancellation signals, session replays, funnel data.
Pulled together, the signals kept landing in the same places.
Food came last. Meals looked interchangeable. What was coming next was hard to see. And sign-up, first order and weekly planning didn’t hold together.
The company had organized the product around plans, product lines and weeks ahead. Customers were making a simpler decision: what do I want to eat?
1 in 5
Every square is one in twenty new customers. Before the redesign.
Chapter 04
The problem was the model
The symptoms pointed to one cause: the order of the journey.
The fix wasn’t a better sign-up flow. It was reversing what came first.
The plan didn’t disappear. It became part of the cart: how much you’re getting. What you picked became the point.
Plans stopped being the way in. They kept running underneath.
PlanShopfirst
Chapter 05
We redesigned the platform around it.
The old model had grown separate surfaces for separate stages: homepage, plans, sign-up, menus, dashboard, deliveries, shopping.
The new model needed one place to shop, whether you arrived today or had ordered for years. We called it the Hub.
One cart that persists from the first visit through sign-up, checkout and every week after, so progress is never lost.
Underneath: shared components, a mobile-first layout, homepage content Marketing could change without engineering, and interaction signals captured from day one.
Look closerWhat changed in the system, layer by layer
| Layer | Plan first | Shop first | Why it mattered |
|---|---|---|---|
| Entry | Plans, then a sign-up funnel | Shop the food, subscribe when ready | Lead with food, not forms |
| Shopping | A public menu to look at, a separate logged-in shopping page | One Hub for new customers and subscribers | Continuity from acquisition to retention |
| Cart | Separate steps, progress easily lost | Persistent from first visit to weekly management | Never lose progress |
| Time | Many weeks ahead in view | One week at a time | Lower cognitive load, a weekly habit |
| Product card | Square container that overpowered the food, symbolic icons, mixed servings and quantity buttons | Food-forward image, at most two text labels, one selector for servings, price and quantity | The food carries the decision |
| Filters | Deprioritized, alphabetical sorting | Up to six filters, two visible at once | Relevance without clutter |
| Homepage | Changes routed through engineering | Modules Marketing can edit directly | Campaigns move without a sprint |
| Data | Little behavioral signal | Cart, skip, quiz and browsing interactions tracked | A foundation for personalization |
| Layout | Patterns inherited from desktop | Mobile-first spec tuned to real device data | Most sessions came from phones |
Chapter 06
The hard part wasn’t just designing the future.
The platform couldn’t stop. Existing subscribers still needed to order food every week.
Growth was acquiring customers through the holiday peak, the most expensive time to disrupt a funnel.
The same teams carried other launches at the same time: holiday and partnership offers, premium kits, personalization data, a component library.
So the question was never whether a better experience was imaginable.
Can we imagine a better experience?
How do we change the core commerce model without destabilizing the business that already depends on it?
Chapter 07
Consequential decisions.
Each one had customer experience, technical reality, organizational pressure and business risk pulling on it at once.
Not takenRework plans, pricing and subscription mechanics along with the experience
ChosenChange how customers enter. Leave subscription mechanics as they are.
Plans kept running underneath. Billing, operations and existing subscribers weren’t put at risk by a change to the experience, and the redesign stayed a product decision instead of becoming a pricing project.
Not takenKeep many weeks of future menus in view
ChosenAnchor the experience on one week at a time
A single weekly cycle lowers cognitive load and builds habit. We showed fewer weeks first to gather signal before committing. The cost was real and showed up after launch: some long-time customers missed planning several weeks ahead.
Not takenPull engineering off QA to launch everything sooner
ChosenSubscribers first. New-customer changes after the holiday peak.
Protect peak performance and acquisition cost, and prove the logged-in experience was stable before asking new customers to trust it. The faster path was on the table, with its risk named plainly: more bugs reaching customers.
Not takenTreat falling conversion as a product problem
ChosenShow leadership the drop started upstream, in traffic quality
Leadership had tied the product team’s conversion target to traffic the acquisition team drove. Diagnosing traffic quality separated what the experience could fix from what it couldn’t, and gave Marketing, Product and Finance one shared view of the number.
Chapter 08
Building the system to ship it.
Every mechanism existed to control a specific risk. Access widened in stages, each one earning the next.
Delivery in slices
Work shipped in defined slices, each with Product, Design, Engineering and partner-team focus spelled out week by week.
Explicit in and out
Release-critical work (navigation, loading states, product pages, UX fixes) separated from what could wait until after release.
A capacity model
Design load mapped against the whole roadmap, with the real gap named: UX architecture and systems thinking, not more visual polish.
Scenario-based QA
Employees tested the flows customers rely on every week (browsing, filtering, cart changes, delivery week changes) instead of open-ended bug hunting. Issues triaged by severity.
A release bar
No critical defects in core shopping or checkout before customers saw it.
Daily release narrative
A shared channel with daily updates from Product and Engineering leads, plus an FAQ for the whole company.
Shared components and a responsive spec
Reusable patterns and breakpoints chosen from actual device data.
Architecture over heroics.
Chapter 09
Before and after.
Drag across each screen to compare. Simplified from the design specs.
Arriving
Before, the first real decision was a plan. After, it’s the food, with a cart that remembers.
Every week
Before, subscribers scrolled through stacked deliveries to work out what was coming. After, one week in focus, in the same Hub new customers use.
The product card
Before, the container overpowered the meal and icons asked to be decoded. After, the food leads, two words of context, one control.
Chapter 10
What changed.
For customers
- One cart that remembers choices across visits and weeks.
- One week in focus instead of many weeks to scan.
- Food-forward product cards and fewer, clearer filters.
For the platform
- One Hub serving both acquisition and retention.
- Campaigns that can land people directly on food and a cart.
- Homepage modules Marketing can change without engineering.
- Interaction signals captured as the foundation for personalization.
- Room to merchandise premium, seasonal and limited-time lines.
For the business
- The model was built to close the gap between signing up and actually ordering, by putting food and a cart before commitment.
What I’d do differently
Bring the team into the ambiguity sooner.
I absorbed a lot of the pressure above the team so they could stay focused. It worked, and it also kept them from building the muscle to handle it themselves. Next time I’d bring them in earlier, with clear guardrails, and still own the call.
Make capacity a condition of the date.
The capacity plan named the gap clearly. I’d make closing it part of agreeing to the timeline from the start, instead of negotiating it while we were already delivering.
Design the habit change, not just the new screens.
The model was right. For customers who had ordered the same way for years, the first weeks were harder than they needed to be. I’d plan their transition as deliberately as we planned the release.
The visible problem looked like a sign-up flow. The real problem was the model.
What I led
The product direction, the Product and Design teams, the sequencing and release governance, and the alignment across functions that let a live commerce model change safely.
Who built it
Engineering, Growth, Data, Culinary, Marketing, CX and Operations, and senior leadership. A platform change this size is never one person’s work.