With chatbots on the rise and all major tech players investing big in AI we wanted to see how this can work for ING. Most importantly we wanted to witness how our customers will react to a chatbot. AI assistants can make telephone calls for us (crazy right?) but are people ready for making use of something like this in their daily life?
It took us 12 weeks, 3 prototypes, 1 ING design sprint, 20 user interviews & 1 field research to bring Inga to life and find our answers.
Goal
First and foremost, our goal was to provide better service, not replacing people in our call centers. The benefit of a bot is that it’s available 24/7 so the customer experiences no waiting time. Another benefit is that the bot can be a very efficient first contact point. If the bot cannot solve or answer a problem, it will refer to the relevant support agent. The agents can focus more on advice, less on answering repeating queries. Sounds a bit mechanical and boring? Bear with me…
We asked the robots
We talked with many bots during our desk research and immersion phase, almost to the point of speaking as one ourselves. During this stage of the immersion we identified some pitfalls to avoid:
- Low error tolerance. The moment a bot makes an error or reacts in an unexpected way, the user tends to quickly disconnect from it.
- Lack of trust. Users are skeptical of using or relying on Chatbots for their personal data or serious tasks on social networks.
- Expectation mismatch. Bots can not offer natural conversation yet and they can’t do everything. Some bots fail to set the right expectations which results in below par experience.
- Cold experiences. There are thousands of bots available across a variety of platforms, yet very few are memorable. A lot of this has to do with the lack of personality and conversational flow the majority of bots offer.
We asked the people
The bots gave us some of the answers but we were missing the human voice. So in order to get back to reality, we spoke with people who talk and chat with our customers for a living — our support agents!
Triggered by the many fun stories of the interactions that our customers and agents had we started seeing this project as a multiplier to those fun and delightful moments. This insight set us on a path and gave us a guiding statement:
“Conversation can take us beyond the typical banking experience.”
Prototype 1
For our first conversations with customers, we reached out to two groups — teens and seniors. For practical reasons we chose Chatfuel and started handcrafting conversations. We covered the hellos and goodbyes, FAQ’s and few scenario flows that we gave as tasks to our customers. We also slipped few jokes and a loooooot of emojis. We wanted to know how far we as a bank can go. Besides going to the extremes, we also asked the more serious questions — concerns for phishing, trust and experience with the bot, talking about banking on Facebook and much more.
The main takeaways from the first prototype were:
- Continuous onboarding. The first experience is key but getting to know each other will be done over time. Balancing these two aspects is crucial.
- Positive microsurprise. Use positive elements like animations, emojis etc. to make each conversation a memorable branded experience. If you decide to joke, go vanilla so everybody can relate.
- Leverage existing trust and brand. Conversations can start on any chat channel but confidential discussions should always happen in a secure ING environment.
- Design error fallback. Just like a real conversation, allow users to clarify intent before reaching a dead end. After a few tries, offer the help of the agent, don’t go in endless loops.
Prototype 2
With our second prototype, we wanted to test some of our assumptions for the future use of our banking bot — helping with money management and being a financial advisor on any matter concerning money.
One example can be advice for expecting a baby.
The main takeaways from the second prototype were:
- Give advice without sounding condescending. Be neutral and remember that everyone is in a different situation.
- Always go for short and sweet. Some of the scenarios we prototyped were just too long to hold everyone’s attention for a long time. This also increases the chances to go into a dead-end. Give meaningful advice as soon as possible.
Prototype 3
This time we went real, no more phony baloney! We took all our findings, our perfected handcrafted dialogues and asked help from the Superheroes Magic Bullet to bring some AI into it. We used the api.ai platform as the brains and spiced things up with a custom, ING branded UI and animations. She was still not ready for the world, we still relied mostly on handcrafted dialogues but we managed to understand intent better, fill in the gaps with social chit chat and respond better when we got in a dead end.
Example flows:
Onboarding with talking about movies
As we found out, the first experience is crucial. It can serve as an introduction to Inga’s capabilities and can be used to create trust.
We decided to start with an old school, paper introduction with a movie ticket gift as a conversation starter and a QR code. The code can be scanned with the ING app where the conversation can continue on the topic of movie / entertainment spend.
Emergency help for lost credit card
“Sorry, I don’t understand your request” can be the most frustrating thing to hear after loosing your credit card. That’s why we made sure this is one of the flows that works perfect, from understanding to executing.
Inga will respond with empathy, list your cards and block the card right away. She can even arrange a replacement card for the customer.
Money management calculation and advice
What if Inga can also answer your financial Google queries? We started with that entry point in this flow.
We took our findings and made the money management flows much more concrete.
We presented advice straight away and if our customer wanted we could adjust the calculation.
Payment request
Just give the task to Inga and she will get your money back. Like the Soprano’s but much much nicer.
The payment request can be settled with just one tap but there is also an option to spice it up with a sticker, gif or a image.
At the end you can see one of Inga’s gestures signalling job well done. During our usability study we found out that that’s actually an offensive gesture in Spain. Ouch, we need to find a new signature gesture!
Conclusion
During our testing, we saw respondents moving from skepticism beforehand to pure delight at the end. We also saw that for some things people will still just call or visit a branch. So all of this won’t happen in a day…
Following our principle of continuous onboarding, we are introducing Inga slowly. For now, she is presenting our search results and is helping you through your investment calculations.
The team
This project couldn’t be completed without the passion and dedication of this amazing team:
- Gregory Ek
- Pragam Rathore
- Susanne Lourens
- Florine van den Adel
- Stephanie Pelsmakers
- and yours truly 😉