Ripple

The brief

We had eight hours to make sustainable action part of an everyday habit.

BLOOM, run by UW/UX and UW Blueprint, gave every team the same question and one working day to answer it with a designed product.

My role

I worked on every stage, from the research to the pitch.

In a team of four with eight hours, everyone touched everything. My part of it:

Finding the problem

We went looking for a habit people repeat every day, not a lifestyle change.

We mapped everyday actions with a climate cost and what each could become. We wanted an action where a small change, repeated, adds up, and one that people talk about less.

Finding the problem

We chose AI prompts because they are daily, invisible and rarely discussed.

We held each idea we had brainstormed against the two things we wanted: an action people repeat every day, and one that is not already the subject of every sustainability campaign.

The problem

We found that every AI answer has a cost nobody sees.

LLMs now answer everything from simple facts to questions asked a thousand times before, and the experience is so smooth that the data centre behind each answer disappears. The numbers we worked from:

Competitor review

I looked at where people ask questions today: none reuse an answer or show its cost.

The tools people already use answer questions well. None of them tell you an answer already exists, or what yours cost to make.

The opportunity

So we asked a different question: why not redefine what it means to use an LLM?

AI isn't going anywhere, so instead of fighting it, we wanted to change how it works.

Solution

We designed Ripple to search before it generates.

A question first goes to a library of answers that already exist. If one fits, Ripple shows it with what reusing it saved. Only when nothing fits does it generate, and then with an efficient, low-cost model.

Solution

We kept the answer the same and changed where it comes from.

Side by side, the difference is invisible to the person asking. That is the point.

Design decision

I kept the interface familiar, so nobody has to learn a new tool.

Ripple opens like ChatGPT: a prompt box and nothing else to learn. Type a question and the matching answers that already exist appear first, ready to pick.

Design decision

I designed the answer page so a reused answer earns trust at a glance.

A resurfaced answer has to show which question it answered, how others rated it, and that you can still keep going.

Design decision

I made the saving visible, because people do not care about what they cannot see.

Water runs through the whole design. The profile counts the litres you saved this month, and a river of past questions, with badges at milestones, turns the saving into something you can watch grow.

Solution

We added Discover so people can learn from answers that already exist.

Inspired by Ditto Lists, Discover lets you browse questions others have asked by category, and learn without generating anything new.

Prototype

We built the whole flow as a Figma prototype for the judges.

Ask, pick an existing answer, follow up, see the saving. Open it to try the flow yourself.

Outcome

We placed 3rd at BLOOM.

Out of the teams at UW/UX and UW Blueprint's designathon, in eight hours.

Looking back

I went back to the research: the idea is stronger at scale than per prompt.

Since we worked from the 2023 estimate, Google and OpenAI have published their own per-prompt numbers, far lower. One prompt costs a few drops. Billions a day still add up.

Looking back

I'd make Ripple personal first, because that is where questions repeat.

Studies of semantic caching since have measured how often questions really repeat. Across everyone's public chats, rarely. Within one person's own history, often. A reused answer can also go out of date.

Key learnings

I learned to meet people where they are and to make the invisible visible.