Scam Detective

• 

  Keeping scammers at bay

Inspiration

When I visited my grandma back in August, she told me a story about getting a phone call from her grandson saying he was in trouble. She was confused, he didn't sound like one of her grandsons. She didn't believe him and hung up. But she wasn't sure she made the right decision. She lives alone and had nobody to check with. So she called her two grandsons, just to make sure.

She was a target of the grandkid scam, a popular scam targeting seniors, where the scammer pretends to be a grandkid in trouble and asks for money. The grandkid scam is one many scams seniors deal with daily.

The National Council on Aging reported that in 2021, 92,371 seniors experienced fraud resulting in $1.7 billion in losses. Some adults reported getting up to 3.5 scam calls a day.

These scammers prey on seniors by pretending to be government officials, offering fake prizes, and more. Five years ago, my grandpa gave his bank account numbers to a scammer pretending to be from the IRS. The work to get the money back was difficult, but he did get it back.

My grandma also talks to Alexa daily (she loves the rain sounds!). So her situation made me curious.

Could Alexa help her understand what to do if she's been scammed? Could my grandma turn to Alexa to reassure her when she hung up on a scammer? Could Alexa educate her on future scams?

I had questions. So I set out to find out.

What it does

Scam Detective teaches seniors how to identify scams and take action if they've been scammed. If a senior receives a suspicious phone call or text message, they can ask Alexa about it. The skill takes them through the process of identifying a scam. Scams range from government imposters, to bill pay scams, to charity scams. Scam Detective covers up to 15 different scams.

Then it goes further and helps users decide what to do if they have given money or personal information to a scammer. Whether it's calling the bank to reverse a transaction or making a report to the FTC, seniors can take action if they've been scammed. It also encourages seniors to talk to the family members or trusted community when they are in doubt.

If a user successfully avoided being scammed, the skill celebrates them.

Alexa also offers a random scam tip (connected by an API) at the end of the conversation to keep educating seniors about how to protect themselves and their community from scams. This skill was intentionally built for every day use, so the tips are randomized and vary.

How we built it

First, I talked with seniors. After talking in depth with my grandma and her friends about their experiences with scammers, I observed how she interacted with Alexa over the course of a few days.

I dove deep into the documentation on scams. The FTC Pass it On! campaign to educate seniors on scams was particularly helpful. I also mined Reddit for posts from family members talking about how their elderly loved ones had been scammed.

Once I had a sense of the problem space, I started building a prototype on VoiceFlow.

I iterated on VoiceFlow, experimenting with the right way to understand user responses as they described their responses. I wanted to increase the likelihood customers will return to the app and tell their friends about the app, so I also added a scam tip that is offered at the end of the skill. I use an API to pull an Airtable database of tips (I chose an external database to reduce utterance collisions and reduce confusion in my model, and make it easy to update about future scams). I experimented with visual cards for multimodal devices to pair with responses (and in the end chose not to use them).

I added return greetings as this is a skill meant to be used regularly. I also added a "what do you do" intent in case seniors get confused about the purpose of the skill.

Then I uploaded to Alexa for many, many rounds of testing. There are many paths to take, depending on the type of scam, if a user paid money/gave personal information to a scammer, if they want to report the scam, etc.

Challenges I ran into

Collisions and confusion were a big challenge during my first iterations. I noticed from interviews with seniors and Reddit posts that there is overlap in the way that users describe scams. I heard things like:

phone call, a phone text, a message on my phone, messaged my phone,

Those are tough to parse out, especially when users share a short story about their scam. So I worked to a lot to reduce collisions in the initial architecture of the conversations. I experimented with intents versus saving responses to variables and iterating over them with keywords to find the right way to get Alexa to understand the user.

I also had challenges randomizing the data from the Airtable API call as randomizing rows is not a feature. So I assigned a variable to each call, used a short javascript script to randomize the call, and as a result users now get a random tip, so they aren't hearing the same tips over multiple uses. Using Airtable also makes it easier to update scam tips in the future.

Accomplishments that I'm proud of

I'm proud of the fact that this skill helps seniors be more skeptical when it comes to scams. Scammers are incredibly persuasive. And seniors experience some shame around being scammed. This Alexa skill helps them learn about scams and how to prevent them, in the privacy of their own home. It also encourages them to talk to their community more to prevent scams.

I'm also proud of the design decisions I made. While this skill is technically capable of doing more (personalization, multimodal), I chose not to because the initial user data showed I needed more data to understand the population. The 55+ group is not a monolith.

What I learned

The primary learning is that 55+ is a diverse age range. My mother, who is in her 60s, is far more tech savvy than my grandma in her 90s, yet they both fall in this target group. Cognitive issues start to become a challenge as seniors age. I purposely added more reprompts and slowed down some responses in my skill to help with confusion (sometimes Alexa speaks a little too fast for some seniors) for 85+ users. I designed for the group that has the hardest time navigating scammers: 85+

Caregivers can also be a primary Alexa user, depending on the use case. The caregivers I talked to were also curious about using this skill to be better informed about scams so they could talk about it with their patients.

I also learned that some older seniors aren't familiar with where those permissions live in their Alexa app. I had wanted to add personalization but there was some feedback that some groups may not be able to find the permissions in the Alexa app to enable them, which would could derail my skill's experience for a first-time user.

What's next for Scam Detective

I'm one person, so I was limited in my abilities in the time frame. But I'd like to take this skill multi-modal. Most seniors I spoke with weren't using multi-modal devices. My grandma uses an Echo Show, however, she was distracted by the graphics on the screen (which matched what Alexa was saying). So I need more data and testing with user groups before proceeding.

I'd like to great the user by name and text them the website and phone number to report a scam to the FTC. However, I learned in user interviews that sometimes the caregiver - like a family member - controls the Alexa app on phone, which complicates permission handling.

I also think it'd be cool to add voice personalization for the caregivers, and maybe a separate conversation for them.

Built With

• airtable
• voiceflow