Helfie AI

Outcomes

Increased trust in the app
‍Comparing trust ratings for the current state vs redesign, users ratings increased from 5.8/10 to 8.4/10.

Turned negative feedback positive
“On doing it once, I now have no hesitation of doing it again. And it’d be the sort of thing you’d recommend to other people.”

Exceptional client approval
"You have already made a difference to the company"

BACKGROUND

What is Helfie?

Helfie is an AI-powered app designed to assess skin health. Given that Australians are at a higher risk of developing skin cancer, the app offers quick and affordable diagnoses of spots or moles for $3 per diagnosis, provided by a registered doctor.

At the time, the start-up had an existing iOS and Android app in the MVP stage and was in the process of developing them further.

PROBLEM

Most people weren’t willing to pay $3 for a diagnosis.

The app aimed to provide easy, affordable access to skin cancer checks. It only cost $3 to send a photo a spot/mole to a registered doctor for a quick diagnosis, but not many people were using this service.

As a very early-stage product, there were gaps in the client's knowledge:

They didn’t know their users
The app wasn’t tested with real people

My Role

This was a live client project that was part of a design course. I conducted user interviews, usability testing, rapid prototyping and created high-fidelity wireframes.

I worked with two other students, one of the Co-Founders of Helfie, and received guidance from the course instructor.

The project was remote and spanned 2 weeks in June 2020.

Testing the current state on a user. Fake mole was drawn with a Sharpie.

PROCESS

Research, define, ideate, prototype and test

Understanding the users

Survey. From a survey with 79 responses recruited via Facebook Ads and a health/fitness Facebook group, we found skin checks were more common than we thought:

52% of people see a GP for skin checks
50% go to a skin clinic
58% of people go routinely every 6 or 12 months

User Interviews. We interviews 11 people to learn more about Australians' attitudes towards skin checks. We found:

Generational differences. Older Australians were more vigilant with going to the GP or a clinic, while younger Australians were less concerned.
Motivators. Reasons for getting a skin check were noticing a change in skin, having lots of moles, or having family history of skin cancer

Current state usability testing. Running 5 sessions, we found the main problems with the app were:

People ignored the AI risk profile because they didn't understand it
People didn’t know who the doctors were or if they were qualified, leading to distrust in the app

Synthesising findings from our user interviews.

Creating personas and journey maps representing older and younger demographics.

HMW create more trust in Helfie for Theresa and Ray?

Together with the client, we shared the research and workshopped different ideas to solve the problem. Using an impact-effort matrix, we prioritised:

Showing doctors’ credentials up front
Educating users during key points in the user journey (onboarding, risk profile, connecting with a doctor)

SOLUTION

Help users understand how the app and the AI worked.

Usability

People didn’t know how to use the app.

To improve usability, we redesigned the onboarding experience and provided clear instructions at key points in the user journey. We also redesigned the risk assessment and made sure users understood what it meant.

Trust

People didn’t trust the AI risk assessment or the doctors’ credibility.

To increase trust, we showed what the AI was calculating and doctors’ credentials. We also redesigned the chat experience of connecting with a doctor.

Prototyping and testing

Over 4 days, we worked through the designs from low-fidelity to high-fidelity, carrying out:

3 rounds of design iterations and testing
21 usability tests in total

Testing onboarding in mid-fidelity

Onboarding

When a user opens the app for the first time, they would see how the app works and their partnership with SunDoctors, a well-known skin cancer clinic in Australia.

AI Risk Assessment

Showing an "analysing" screen before the results helped users understand and trust the AI assessment more.

“It’s satisfying to know there’s a criteria, it said what the 3 things it was looking for”

Connect with a doctor

Users are placed in a queue after they submit their results. Showing their place in the queue and an auto-confirmation message was important feedback during the wait.

OUTCOME

Increased trust in the app.

📊 Results

At the end of each usability test, we asked users to rate their trust in the app on a scale of 1 to 10.

In the first round, 5 users rated it an average of 5.8/10.
On the third round, 5 different users rated it an average of 8.4/10.

At the time, we reported a +45% increase in trust.

With what I've learned since, the small sample size makes this result unreliable as a quantitative measure and reporting it this way is misleading.

💬 User Feedback

There was a huge contrast between the first and last rounds of usability testing.

Current state: “Is it a real doctor or a bulls*#t doctor?”
Redesign: “On doing it once, I now have no hesitation of doing it again. And it’d be the sort of thing you’d recommend to other people.”

✅ Client approval

After we presented to the Co-Founder, they said:

”I’m extremely impressed... I can say with pretty high confidence that I’m going to be employing some of these things that you guys have put together.

...you have already made a difference to the company”