120 Bank Caregiver Agency Platform

Turned a platform story into a search-first marketplace, so Taiwanese families see real caregiver agents in seconds. Designed in code with AI, no Figma.

Timeline

Apr 7, 2026 to Present

Team

Product Designer: Me Founder & CEO Engineer: 1

MY RoLE

  • Led product and UX design for the public site, employer account, agency dashboard, and admin.

  • Built working prototypes in code with AI and shipped them to engineering as pull requests.

  • Built the design system with tokens, CSS, and React components.

  • Wrote design specs for 16 system emails.

  • Advised the founder on go-to-market for both sides of the platform.

An older woman smiling with her caregiver, next to the redesigned 120 Bank homepage on desktop and phone

Overview

The situation. 120 Bank helps families in Taiwan find a broker agency when they need to hire a foreign caregiver. It is an early-stage, two-sided platform with one founder, one engineer, and me as the product designer.

The key decision. I changed the homepage from a platform story into search, and I added guides that help agents look trustworthy before a family calls.

The result. The redesign is live on 120-bank.com, and 27 verified agencies are listed on the platform as of November 2026. The engineer no longer writes front-end code for design changes. They only review my pull requests and ship them.

I know this problem from my own family. Earlier this year, we hired a caregiver for my grandma. Our biggest question was which agency we could trust with her, and we had no way to know. So we picked one by feeling and luck.

For a new platform, the risk is pretty simple. If families cannot see someone they trust in the first few seconds, they go back to asking relatives. Then the agencies on the platform get no requests, and they have no reason to stay.

Before and after of the 120 Bank homepage

Before and after. The new homepage opens with search, and real agents show right below it.

Challenge

Families did not need a story. They needed proof that real people were there.

Families find agencies in many scattered places. They read the notice board at the hospital, ask relatives, or post questions on Facebook and Threads. Because the information is spread out, it is hard to find an agency, and even harder to compare them. Every agency website also says it is the best, and the official rating only shows the company, not the person who will call you.

So our focus was to bring all of this information together in one place, where families can find and compare real agents. But I found three places where the old design got in the way.

  • The homepage explained the platform instead of helping people search. It followed the SaaS landing page formula. But people come to 120 with one goal, and they already know they need an agency. They only want to know if there is an agent who can serve their family.

  • New families signed up as the wrong role. Signup was one card with a small switch for employer or agency, and many people missed the switch. The login title also said “Welcome back”, so new families felt the page was not for them.

  • Profile photos did not look professional. Agents uploaded daily life photos, business cards, and images that had nothing to do with the service. For a family, the photo is one of the few things they can judge before calling.

Discovery

We could not photograph every agent, so the design had to do the teaching.

My first idea for the photos was simple. We would just ask agents to upload a better photo. But “better” means something different to everyone. And unlike Airbnb in its early days, we could not send a photographer to every agent.

Research also showed how much the photo matters. People form a trust impression from a face in about 0.1 seconds (Willis & Todorov, 2006), and Airbnb guests choose hosts partly by their photos (Ert, Fleischer & Magen, 2016). It is the same as booking a room. If the photos look bad, people do not book, even if the room is clean.

Decision. I put the guide right inside the upload. There is one real example of a good photo and four “don’t” examples, which are selfies, AI photos, business cards or logos, and group photos. The upload also warns agents when the photo is too small.

Trade-off. The “don’t” examples use simple line drawings instead of real photos. They look less real, but they never shame a real person on the platform.

Profile photo guide inside the upload, with one good example and four kinds to avoid

The photo guide sits right inside the upload. There is one good example and four kinds of photos to avoid.

Families had two ways to find an agent, but one was hidden behind login.

120 offers two ways to find an agent. Families can search on their own with the search box, or they can send a request and let 120 match them with an agent. But in the old version, families could only see the request after they logged in. We felt that both are core services of the platform, so both should be easy to see, and the matching service should not be buried.

Decision. I brought both paths out to the search box in the hero, as two tabs. The first tab is “Find an agent myself”, and the second tab is “120 finds for me”. Search stays the default, and families can see both options in the first few seconds.

Trade-off. 120 passes each request to only one agent, and the team assigns it by hand. This is slower to scale. But every family gets a real phone call, and it is the shortest path to the first deal for the business.

Two tabs on the hero search box: Find an agent myself, and 120 finds for me

Two tabs on the same search box. Families can search on their own, or let 120 find an agent for them.

Solution

Show real people first, and guide both sides on the way.

1. A search-first homepage. The homepage now opens with search. Families pick their city and see real agents right below. The brand story moved to the About page. A row of real agents shows that the platform has real people, and it works better than a line that says we are trustworthy.

Homepage before and after

Before, the homepage opened with platform numbers. Now it opens with search, and real agents show right below.

2. Two clear paths in the hero. The second tab, “120 finds for me”, shows 4 short steps. You fill in the request, 120 passes it to one agent, the agent calls you, and you decide if they are the right fit. It reuses the request form we already had, so there was no new form to build.

Hero before and after

Before, the hero only had search. Now a second tab, “120 finds for me”, shows the 4 steps.

3. Choose your role before login. Families and agents first pick “I am an employer” or “I am an agent”, and then they go to the right page. Many users do not read small details, so the first step has to be impossible to miss. I kept the login method the same as before, because users already know it.

Login before and after

Before, login and signup shared one card with a small switch. Now families and agents pick their role first, so nobody signs up on the wrong side.

4. A simple dashboard for families. The employer dashboard used to look like a back-office system, and that can scare people who do not go online often. Now it looks like a normal web page with just three simple tabs.

Family dashboard before and after

Before, the family dashboard looked like a back-office system. Now it is a plain page with simple tabs and one clear next step.

5. City pages and a map, designed as one flow from Google. Most families start on Google, not on our homepage. So I designed the city pages as the first step of the journey, not only as SEO pages. The first thing is that a family searches something like “Kaohsiung migrant worker agency”, and our Kaohsiung page shows up on Google. The second thing is that when they land, the map of Taiwan shows right away how many agents serve Kaohsiung. The third thing is that they click their city and see every agent in that area. Cities with no agents yet are hidden from Google, so a family never lands on an empty page. The map also helps families who do not type much, because they can just click their city.

Agent list before and after

Before, families scrolled one long list of agents. Now a map of Taiwan shows how many agents serve each city.

City page flow: Google result, Taiwan map with agent counts, and the agent list

One flow from Google. The city page answers the search, the map shows how many agents serve the city, and one click shows the agents.

6. One design system for engineering. When I joined the project, one of the first things I did was list every component the platform used. I grouped the ones that did the same job and wrote rules for them. Each component is built once and reused on as many pages as possible. The design system also covers voice and tone, with UX writing rules for the interface, so every page talks to families in the same plain words they already use. This gives everyday families a clear, familiar interface, and it keeps information comfortable to read as the platform grows. For example, card corners went from 5 sizes to 3. Messages now stay on screen for 4.2 seconds instead of 2.6 seconds, because Chinese messages are longer and people could not finish reading them.

Components from the 120 Bank design system

Components from the design system. Each one is built once and reused across the public site, the family account, and the agency dashboard.

How I worked with AI and engineering

I did not open Figma for this project at all. But I still followed the same design process I have always used. The only thing that changed was the tool. Figma became AI, in this case Claude Code, and I designed in working code.

My five-step design process, with Claude Code in place of Figma

My design process stayed the same. Only the tool changed, from Figma to Claude Code.

  1. Research. I studied the product’s positioning, the market, and our users. For user research, I had AI collect and read public posts, so I could see what families really think. Then I turned the findings into charts, like a positioning map.

  2. Information architecture. I mapped every page of the old site first. Then I compared it with the new plan and decided what to change.

  3. Design system, not wireframes. I skipped wireframes and started with the design system, so the style was set before any screen. Our users range from people who are online every day to people who rarely use the web. So I chose a calm, familiar style over a bold one, because a style that feels too unusual can make some people afraid to try. Research supports this. People judge a website within about 50 milliseconds, and they find simple, familiar-looking sites the most appealing (Tuch et al., 2012).

  4. UI design. I designed every screen from the information architecture.

  5. Test, then ship. I put the screens in my own design environment, where the team clicked through and tested them. After we agreed, the engineer pulled the changes into production.

  • What AI did. I described the screens and the rules to an AI coding tool (Claude Code), and it built the prototype in code. It also helped me compare the prototype with the design system and list every difference.

  • What I decided. The first AI version looked very generic, so I rejected it and set the direction myself. That meant a full-width photo with a search card on top, a white base, and brand pink used only as a small accent. I also reviewed and approved each of the 11 design system rules one by one.

  • What I kept for myself. I wrote the key interface copy myself, like the tab names and the 4 steps, because families need words they already use.

  • How we shipped. The engineer set up a design repo on GitHub for me. Each pull request got a preview link, and the engineer reviewed it before it went live. In meetings, the team clicked through a real product instead of looking at static mockups.

AI can build a screen fast, but it cannot decide what the business and the users need. So I will say AI made the building fast, but the design decisions were still mine.

Advice for a two-sided platform

In September, the team was focused on investors and agencies. I suggested working on both sides at the same time. We would keep bringing in agencies, and we would also reach families on Facebook, Instagram, and Threads. This also helps us collect real problems from families. The team agreed, and later, the first agent who found 120 through organic traffic came from my posts on Threads.

Impact

  • The search-first homepage, the two paths in the hero, role-first signup, the photo guide, the new family dashboard, and the city pages are all live on 120-bank.com.

  • As of November 2026, 27 verified agencies are listed on the platform.

  • The first agent from organic traffic came from my posts on Threads.

  • One design system with 11 rules, so engineering builds from one spec instead of guessing.

  • Design specs for 16 system emails, handed to engineering.

Agency claim email, before and after

The claim email before and after. The new version shows agents that families are already visiting their page.

What the team said

“Guo’s designs feel professional and very close to what our users really want. I tried building screens with AI myself, but they never felt right. What Guo designs fits both our business goals and our users’ goals.”

Founder & CEO, 120 Bank

“I don’t need to write front-end code for design anymore. I just review Guo’s pull requests and ship them. It saves me a lot of time, and we skip all the back-and-forth about design details.”

Engineer, 120 Bank

What I would do next. The next thing we are building is a short guide page before the request form. Before a family fills in their needs, the page explains a few basics. For example, the caregiver comes to care for the elderly person, not to do housework for the whole family, and there are a few things the family should know and prepare in advance. It works like a short training before the real start, and it helps in three ways. The first thing is that families learn the know-how they need before the caregiver arrives. The second thing is that agents spend less time teaching every family the same basics. The third thing is that it sets the right expectations early, so families who are not ready can think twice, and the requests that reach agents are more serious. We are also measuring the baseline in November, so I will add numbers when we have them.

So I will say the biggest change is that a family can now see real people within the first few seconds.

The search-first homepage on desktop
The city map on desktop and mobile, with a county highlighted on hover
The city map after a click: the selected county and the agents who serve it
Ten mobile pages, from search and the city map to agent pages, login, guides, and contact
An agent page on mobile
The how-it-works guide: three steps, then a full walkthrough
The care guide on desktop and mobile
The agent claim email, with page views and a weekly chart, as it arrives in an inbox
The contact page on desktop and mobile
The search-first homepage on desktop
The city map on desktop and mobile, with a county highlighted on hover
The city map after a click: the selected county and the agents who serve it
Ten mobile pages, from search and the city map to agent pages, login, guides, and contact
An agent page on mobile
The how-it-works guide: three steps, then a full walkthrough
The care guide on desktop and mobile
The agent claim email, with page views and a weekly chart, as it arrives in an inbox
The contact page on desktop and mobile

I blend product thinking with crisp UI craft to drive adoption and lighten ops. I work async across time zones with clear docs, design systems, and tight feedback loops, shipping fast without the chaos.

guocianyu@gmail.com

@2026 Cian-Yu, Guo All rights reserved.

Interested in working together?

If you have a project in mind, or you just want a second opinion on your product, I’m happy to talk.

I blend product thinking with crisp UI craft to drive adoption and lighten ops. I work async across time zones with clear docs, design systems, and tight feedback loops, shipping fast without the chaos.

guocianyu@gmail.com

@2026 Cian-Yu, Guo All rights reserved.

Interested in working together?

If you have a project in mind, or you just want a second opinion on your product, I’m happy to talk.

I blend product thinking with crisp UI craft to drive adoption and lighten ops. I work async across time zones with clear docs, design systems, and tight feedback loops, shipping fast without the chaos.

guocianyu@gmail.com

@2026 Cian-Yu, Guo All rights reserved.

Interested in working together?

If you have a project in mind, or you just want a second opinion on your product, I’m happy to talk.

I blend product thinking with crisp UI craft to drive adoption and lighten ops. I work async across time zones with clear docs, design systems, and tight feedback loops, shipping fast without the chaos.

guocianyu@gmail.com

@2026 Cian-Yu, Guo All rights reserved.

Interested in working together?

If you have a project in mind, or you just want a second opinion on your product, I’m happy to talk.