Password protected

A small password to read on.

This project contains work from Marshmallow, so I've kept it behind a password. If you're here from an application or email from me, it'll be included there.

← Back to all work
All work

Shipping home insurance in five months, as the solo designer.

Marshmallow's first expansion beyond car insurance — a 0→1 home insurance product that launched in January 2026.

Role
Solo designer, end-to-end
Timeline
Jul 2025 – Jan 2026
Team
PM, engineering, underwriting, pricing, compliance

Context

Over a million people move to the UK every year. Mainstream insurers weren't built with them in mind: no UK credit history, no local track record, judged as a blank slate regardless of experience elsewhere. Marshmallow was built to close that gap, starting with car insurance.

Home insurance was the natural next step. Marshmallow had tried once before, in 2021–22, and it hadn't shipped. By mid-2025, the business decided to try again — this time with a clearer plan, a capacity partner in place, and a real delivery commitment. This was the company's first product beyond car insurance.

Speed mattered as much as the product itself. Marshmallow wanted to go live fast to test real appetite — whether people new to the UK actually wanted home insurance — before committing serious investment. That meant a deliberately lean setup: a small team building as scrappily as possible, with resources concentrated on getting the quote experience right.

The problem

A lot had changed since the first attempt in 2021. Four separate pressures landed on the same journey at the same time.

  • A capacity partner to satisfy

    Marshmallow now had a partner underwriting the risk, which meant meeting their requirements before going live: a defined list of data points we had to capture, plus required fraud checks. Approval came through a series of check-ins, presenting the work back at each stage.

  • Parity with the comparison sites

    The pricing model had to match the rest of the market to go live on price comparison websites. That meant asking the same questions competitors ask — around 25 new data points on top of the flow we already had.

  • Consumer Duty

    The FCA's Consumer Duty had come into effect, raising the bar for how clearly cover and price had to be communicated. That mattered most for the genuinely complex parts of the product: explaining what's included, and how customers set their own cover amounts.

  • A design system built for car

    Home has a different shape entirely — different data, different decisions, different customer concerns. Reuse was essential for speed, but only where it genuinely fit.

Together, this meant rebuilding the whole end-to-end customer experience from zero.

The business constraint

Before we could start building the quote journey, we needed to consolidate complex data requirements from underwriting, pricing and the partner — then define and build each question: the answer options, the validation rules, and the logic behind it.

All of it came with real constraints: what we actually needed to capture, what answers we needed to get, and how to show and explain genuinely complex concepts to customers — like how someone sets the total value of their own valuables.

The requirements spreadsheet used to consolidate underwriting, pricing and partner data points.

I worked with underwriting and the head of product to turn that raw data into questions that made sense to a customer, not just to the business. After that translation work, meeting the new requirements alone put the flow at around 57 questions — and capturing all of it was non-negotiable for the partner to approve the product for launch.

The process

1. Designing for conversion

A study of 300 ecommerce retailers found that each additional field in a form can drop conversion by anywhere from 8 to 50%. I knew asking 57 questions in a quote flow was too many and would hurt conversion.

We ran competitive research with our researchers to understand what others were doing. Monzo runs under 8 questions and takes less than 10 minutes to get a home insurance quote. Hastings uses a platform that pulls property information automatically and asks the customer to confirm it, rather than type it in. I wondered whether we could do the same, or something close to it.

I drove the integration with a third-party property data provider, pulling property information through an API and pre-filling a large chunk of it directly. Instead of asking questions, we could present statements for the customer to confirm: if a property had a history of flooding, for example, we already had that information and could simply ask them to confirm it. Fewer questions also meant less to build and integrate on both the front and back end.

57 → 25 Questions in the quote flow, while still meeting every underwriting and pricing requirement.

2. Designing for understanding

Do you know what your roof is made of? Or how far your home is from the nearest river, stream or sea?

Most people don't. We ran rounds of moderated and unmoderated research throughout the project, watching participants go through the quote journey and noting exactly where they got confused, got stuck, or couldn't answer a question confidently. We'd take the feedback, iterate, and test again.

Certain areas drove the most confusion: complex questions like roof type and distance from water, and the whole cover customisation section.

Roof type. Most people simply don't know what material their roof is made of. I leaned on recognition over recall: images alongside each answer option let people identify their roof by sight rather than recall a term they might not know, plus a "Learn more" explainer with guidance on where to find the information if they were still unsure.

Distance from water. We added as much supporting context as we could around the question, and the "Learn more" content pointed people toward how to check the distance themselves.

Both of these questions, alongside others in the flow, determine whether a customer can be covered at all — so getting them right was crucial.

3. Designing for people's mental models

The flow's overall order was a deliberate call. I structured the journey to move from broad to narrow, following how someone actually thinks about buying home insurance — "where do I live" before "who am I". Grouping questions this way meant the customer wasn't switching mental context back and forth between their house and their life every few questions.

Progressive disclosure ran through the flow from the start. The principle was simple: don't show a question to someone it doesn't apply to. What a customer saw next depended on what they'd just answered, so no one sat through a follow-up that wasn't relevant to them.

Cover customisation went through the most significant rethink. A version of this page had existed since the 2021 attempt, but it carried a logic we couldn't implement. My first redesign still wasn't question-led — closer to selecting items from a list than answering anything. Research surfaced the same confusion again, which led to the real fix: guiding people to a decision through questions. Instead of asking every customer whether they wanted to insure a bike, we first asked whether they owned one — turning a browsing task into a sequence of small, answerable decisions.

Solutions

  • Guided questions with progressive disclosure

    Each step asks one question, and only the questions that apply, so decisions stay simple and the flow keeps moving.

  • Support content for confidence

    Most people can't name the material their roof is made of, but they can recognise it. Answers carry images, and "Learn more" explains where to check, so people can answer confidently instead of guessing.

  • Question-led cover customisation

    Cover options are presented as simple, guided questions rather than a catalogue to browse and configure yourself. Less guessing at the quote stage means fewer surprises at claim time.

  • What you're covered for

    The final quote breaks down what's included and why, so customers can see exactly what they're buying before they commit. Clarity removes the excuse to leave.

Learnings

Testing our assumptions and refining based on what we learned proved essential, again and again, to arriving at something that actually made sense to people. Looking at where we started versus where we ended up, I couldn't have got to the end product without research.

Ownership, ownership, ownership.

So much of this project moved forward because I asked, pushed, and took ownership of problems that weren't strictly mine to solve — driving the integration with the property data platform being the clearest example.

That same instinct is also where I'd do things differently next time. We chose a provider we already had a relationship with through car insurance, because it was faster and easier to negotiate a contract with. Looking back, I suspect there were better platforms out there that could have pulled in more property data and cut the question count further than we managed — but at the time, investigating alternatives wasn't a thought I chased. Pushing a project further than the obvious, fastest path takes real effort: initiative, curiosity, and a willingness to question the easy option rather than settle for it.

More work, shorter write-ups

Back to all work