back Back to all

What Kai is learning. What we are learning.


Kai, our banking bot, has been chatting with consumers for just over three months now. Every day, Kai shows us how people use banking bots in ways that are vastly different than a bank’s mobile app. People are doing things that they simply cannot do on mobile or the web. We’re learning how our conversational AI platform enables entirely new experiences.

We created MyKAI with our platform, KAI Banking on Messaging, to help people manage money, track expenses, analyze spending and answer banking questions – all from inside Facebook Messenger, SMS, and Slack. MyKai supports more than 20,000 US banks and credit cards and is available to consumers from our website.

MyKAI constantly improves from a combination of supervised and unsupervised learning. Our AI trainers and machine learning work together, and the combined improvements are available to banks via KAI Banking. Not only does our platform improve with every consumer insight from MyKAI, but so does our development and deployment environment.

When banks integrate KAI Banking on Messaging, it unlocks another layer of features not available on MyKAI, such as the ability to pay bills and transfer money as well as upsell products and services.

Of course, all of these insights and examples are based on aggregate and anonymized data. On that note, we take privacy and security very seriously. MyKAI is as secure and private as a bank’s mobile app. No personal or financial data is ever stored, and we don’t know the identity of any user.

Enough with the context and onto what we’re learning. Here’s how are consumers are engaging with Kai.

People are having conversations in the strict definition of the word – an informal way of talking to one another.

Given Kai is a bot and the user interface is texting, you might expect that, but what we see are exchanges that reflect how we text a friend versus how we complete a task in a search box to find something.

When someone engages in a conversation with Kai, on average they send 5.3 messages during that single session. That means when Kai responds to your first question or request you send another four messages – that first response triggers a longer conversation whether it’s a deeper dive into details or it sparked a question on another topic.

One conversation, or session, on average is 10 messages between Kai and the user. Since Kai is steeped in natural language processing and understanding – there are never prompts, like “Type ‘Recent’ to see your latest transactions.” It’s all plain English – just like we’d text someone we know.

Here’s an anonymous example of multiple subjects in a single conversation with Kai:
User: How much money do I have
Kai: You have $a and owe $b Shall I break that down?
User: What do I owe on my visa cc
Kai: You owe $c on your Credit Card (1111). Want more
User: How much did I spend on entertainment this month
Kai: Let’s see. You spent $d. I searched for:
Category: entertainment
User: How about last month
Kai: ….
User: What about travel
Kai: ….

People are engaged with Kai.

Our analytics show that people come back often for more account and transaction information. Banking with Kai is a sticky experience and on average, consumers engage in a conversation with Kai more than six times per day. We attribute this to the fact that Kai lives where you spend most of your digital time – it’s a text away on your favorite messaging app. This kind of engagement far exceeds a bank’s KPI for their mobile app engagement – which is at best multiple times a month.

People want to know how they are spending their money.

In fact, they want to know something about their transactions almost twice as many times as they want to know their account balance. It makes sense that 47.5% of conversations are related to spending patterns given it’s relatively easy to find out your account balance. Try using your bank app to see how much you spent on Uber between March and August, and compare that with a bot that instantaneously tells you that totals and breakdowns the details in one conversation.

On that note, consumers asked more about their discretionary spending than their bills. In fact, 56.8% of the inquiries were about items such as food or coffee. Kai spends a lot of time focused on our dining and caffeine spending habits.

To understand how we spend our money, Kai understands how to categorize merchants. As you’d imagine, Kai recognizes national brands like Amazon, Macy’s, Starbucks, Uber, and Whole Foods, and knows which ones are food, coffee, or transportation. Kai also categorizes spending at local merchants, like the yogurt shop or Vietnamese restaurant down the street. To understand how we spend our money, Kai understands how to categorize merchants.

People trust Kai.

We’ve been overwhelmed by the conversion we’re seeing – a high percent of the people that sign up for Kai complete the account creation and connect at least one bank account or credit card. On the trust note, Kai is often trusted to pay people via Venmo. It’s seamless for Kai to make Venmo payments and we’re pleased with the growing usage.

Since a conversation is an exchange and two-way street, here are some insights on the other side of the conversation – how Kai engages with consumers.

When Kai starts a conversation, people respond.

A good bot knows how to start a conversation and keep it going. In fact, every smart alert or push notification Kai sent to prompt a conversation led to an average of 150% increase in daily usage.

Example conversations that Kai started:
Kai: Good morning X! Still need to settle up for Saturday night? Ask me to pay a friend anytime – no need to open up Venmo.
Kai: Hi there X, It’s Kai. I don’t get holidays off, so feel free to ask how much you’ve spent on barbecue this month. Happy Fourth!
Kai: Hi X, Kai here. Happy National Junk Food Day! Ask me how much you spent on Junk Food this year.
Kai: Hi, X! Today is payday for many humans, I am told. Good news: Now, I can search your transaction history for income. Ask me how much you made last month or in 2016.

Kai has even prompted structured feedback and sent surveys to users which elicited responses as if a friend asked for it – they gave it to Kai straight.

Kai is keeping up with the unexpected.

Meaning that Kai is parsing all of our precise and idiosyncratic questions that usually we ask only of other humans. Since people are conversing in free-form language, and Kai responds instantaneously, that’s a lot of parsing of our unique personalities.

The range of how people talk to Kai:
User: Show debits on July 11
User: Kai, I feel like I’m spending a lot on restaurants? what was my largest restaurant transaction in 1111?
User: How much did I spend on Uber in April?
User: Has my rent check cleared?
User: Did I buy any X last year?
User: What was my largest hotel transaction this year?
User: Have I ever purchased X?
User: what about on travel
User: what about in june

Kai embodies a persona that resonates.

Creating a bot’s persona is a science and art unto itself, and we put a tremendous amount of time and thought into Kai’s voice and personality. From the conversations and sentiments, we believe most people like Kai – many even refer to Kai by name when chatting. We’ve struck a balance with a dry sense of humor that’s not too cute, and a bot that’s helpful and efficient, yet personable.

Examples of kudos to Kai:
User: Thanks, that was really helpful
User: thanks…thats it
User: Thanks, Kai!
User: cool thanks
User: Already seeing improvements thanks!
User: sweet, thanks
User: Thanks. Talk you soon


From our unique position as a bot maker and platform creator, we’re more convinced than ever that a smart bot has to translate its knowledge into natural language to create compelling experiences. Intelligent and natural conversations are the keys to enabling lifestyle banking.

We will continue to share what we’re learning.