A winning interaction model for chatbots to evolve, survive, and thrive.
In China, WeChat users can buy a house and shop for the furniture while they wait for their Peking Duck to be delivered, all within the super app. But WeChat is not yet the global player Tencent hoped it would be (MIT Technology Review).
Elsewhere, WhatsApp, Messenger etc. are unlikely to become the Operating System for our daily needs such as shopping, ride hailing, and bill payments.
Here in Singapore, people still rely on Amazon, RedMart, and Lazada to shop, Uber and Grab to order a ride, Deliveroo to order a pizza, and one of the many local wallets to make small payments. They don’t know or will care if there are chatbots that do the same things.
Most big banks have launched chatbots with limited services exposed. In most cases, those are the services banks have been trying to improve over a decade on the other platforms such as their mobile apps and websites. I doubt customers will spend time striking a conversation to see their account balance. I doubt they would want that information to flow through Facebook’s Messenger in the first place.
Until the messaging behemoths find a business and interaction model that achieves global mass adoption, customers will continue to rely on mobile apps for getting real things done.
People like having multiple ways to engage with businesses, because one way might be more convenient for them at that moment. If a bank calls me for a missed payment, they might as well help me make the payment during the same call. A general insurance company actually calls me every year before the policy expires and the person hangs up only after the renewed policy is in my mailbox.
People however, don’t like being forced to go to a website for getting contact information, an app for making a transaction, and a chatbot for FAQs.
Especially with chatbots, doing one, or a few things really well is not enough to make that a viable engagement channel. For that, it needs to do almost everything that one can do by visiting a physical location, the website, the app, the phone, all combined. Messaging is quite a rich and natural interaction model, so it doesn’t get to expose only a part of content and services.
Human needs are complex, and trying to force fit all those into a conversational paradigm is not the solution. At least during this transition period from conventional apps to fully conversational experiences, we need an interaction model that retains the discovery aspect of apps and introduces the power of conversations.
The most effective business apps of 2018 and beyond need these 4 levels of interactions…
Level 1: Immediate Discovery of Frequent, Recent, Popular, and Important Information
Discovery is one thing chatbots are not great at. They are not good at being found on their respective platform (e.g. Messenger), and they are not good at showcasing their skills and capabilities. Users often don’t know what all the chatbot can do and what is the best way to get started.
I shouldn’t have to ask my bank’s chatbot whats account balance is or if I have any bills due. It should already be displayed within zero to one click. I shouldn’t have to even say Hi or Get Started to see the list of products or services available. I should be able to passively “window shop” without any interaction whatsoever. Websites and apps are great at this, so why change it if it ain’t broke.
The landing screen of the chatbot should be designed for browsing frequent, recent, popular, and important information and categories. This screen is a mashup of a home page and a mega menu and should be easy to get back to anytime during the session.
Level 2: A Search Function Like “Spotlight” in iOS and Mac OS
Browsers will browse, searchers will search. On Apple devices, Spotlight is the fastest way to launch an app, find an email or a message, or calculate the exchange rate. It’s fast to launch, and suggests input after each keystroke.
As you can see, the purpose is not only to index the static content, but also to execute simple requests really quickly.
For a bank, this pattern can be used for retrieval of specific transactions (“Transactions on VISA 1234 in the last 2 weeks”, “Send $12 to Kevin”). With auto-suggest and auto-complete, there is hardly any typing needed.
This level already starts to feel like a chatbot because of natural language support. It’s great for basic tasks that can be translated into simple one way commands (similar to “Alexa, play my daily mix”). These are the tasks for which I am not comfortable starting a conversation (even with a chatbot).
Level 3: A Chatbot That Understands Complex Sentences and Illuminates Unstructured Data
The first two levels are great at many things but this one is where businesses can create maximum customer engagement at scale. Its great for sales, services, transactions, and support.
Here are a few use cases that are best handled at this level:
“Show me the best credit card for travellers, zero annual fee”
“Going to Japan for 2 weeks with family. Insure us.”
“I lost my wallet.”
“I have $10K to invest and my goal is to have $1M in my account when I am 40. Show me a low risk portfolio to achieve that”
“Send $400 to my dad.”
Level 4: An Immersive Video Chat With a Real Person Powered by AI and UI
This level is for those rare and special situations that are either not done well by the previous 3 levels, or the user prefers a more human but slightly less instant way to get that thing done.
Potentially, immersive experiences can be created at this level. A real person can cognitively and emotionally understand everything the customer is saying, and use technology to augment the responses via audio, video, and user interface elements.
The tiniest friction matters to people. Over time, they will learn the perfect way to get a particular thing done. If all levels are done well, they will stop going to L4 for things that are better suited for L2, and so on.