By Nicholas Drayson
There’s been a lot of chatbot chit-chat in the financial inclusion space recently, and for good reason. In addition to driving down customer support costs, chatbots are uniquely positioned to improve the accessibility of financial products, deliver tailored financial education programs, and extract new streams of consumer data.
That’s why in May we published our first blog on chatbots, titled “Can Chatbots Promote Financial Inclusion?” Since then, we at Accion Venture Lab have seen a number of models hoping to use chatbots to reach the underbanked. Here’s what we found.
Beware big bots for big banks
The most technologically advanced chatbot providers are partnering with large financial institutions. This makes sense. Large financial services firms can provide, among other benefits: (i) access to large consumer segments and data assets; (ii) strategic, operational, and regulatory experience and support; (iii) relationships with large potential customers and investors; and (iv) brand recognition. They’re also often the only customers that can afford to hire the very best chatbot developers.
However, despite the hype, the brightest financial services chatbots, such as Kasisto, Abe, Clinc, and others, are still developing or piloting early-stage products, and there are few widely-deployed machine-learning chatbots in financial services.
Additionally, two of the most recognizable banking chatbots — Capital One’s Eno and Bank of America’s Erica — are being developed in-house. It’s unclear whether a third-party provider can match that level of integration.
Each of the above businesses is arguably well-placed to grow. However, we think it’s likely they will all continue focusing on long-term projects targeting existing clients of large institutional customers in the medium term, limiting their financial inclusion impact for now. So where should we look for chatbots that support financial inclusion today?
The three things we look for in financial services chatbots
Our experience has led us to seek the following features in financial services chatbot providers:
1. Narrow applications
Chatbots are particularly well-equipped to guide users through well-defined modular processes, such as transaction funnels or educational programs. These functions can preserve low chatbot development costs and failure rates (the rate at which a chatbot fails to fulfill or understand a query), while improving customer engagement and capturing meaningful data assets.
A great example of where this can drive financial inclusion is in insurance brokerage. Insurance access is critical among lower income groups, for whom small shocks can generate outsized consequences. Unfortunately, selling insurance to this demographic has long-been a pain-point for insurers due to a combination of small customer lifetime values, and relatively extensive consumer education requirements.
Companies like FinChatBot and ToGarantido are betting that chatbots will help make the purchasing process more consumer-friendly while lowering brokerage costs. FinChatBot, a startup in South Africa, is partnering with several insurance carriers to create insurance brokerage platforms designed around a machine learning chatbot. ToGarantido, an online microinsurance broker in Brazil, is working with industry-agnostic chatbot provider Fred to create a rules-based chatbot that will sell small life insurance policies.
We think there are many similarly modular applications, such as underwriting or KYC verification, where chatbots can have the same effect.
2. Cohesive business models
Much of the excitement around chatbots stems from progress in natural language processing and machine learning capabilities. When looking at individual early-stage companies, however, these innovations can be hard to quantify, are rarely defensible by themselves, and often belie a more complicated reality.
Fortunately, cutting-edge businesses are not always dependent on cutting-edge technology, especially for some of our target markets. That’s why, when it comes to chatbots, we look for more than technological excellence. We look for innovation throughout the business model.
Arifu is partnering with Equity Bank in Kenya to create customized financial literacy courses distributed via SMS. Through its interactions with learners, Arifu can promote Equity Bank products while capturing valuable customer insights –income, savings rates, age.
To date, Arifu has engaged over half a million learners. The company’s strength isn’t necessarily in its chatbot technology, which is rules-based and comprises no natural language processing capabilities whatsoever, but in Arifu’s unique ability to engage elusive demographics.
3. Robust data and technology strategies
We’ve found that even chatbots performing the simplest functions need to fulfill certain strategic technological requirements.
First, a chatbot provider needs a clear data acquisition and monetization strategy. Data access is the key long-term determinant of machine learning-fueled technology quality, and ultimately scale is what will differentiate the most powerful chatbots from the rest. To that end, every viable chatbot startup should have a clear path to data acquisition, and any chatbot with genuine network effects is worth looking at closely. At the very least, chatbot providers need to understand how the data extracted from their engagements with consumers will create value for themselves and for their customers. This is particularly relevant for financial services, where information gleaned from customer interactions could lead to improvements in areas as diverse as underwriting quality, customer conversion rates, and product design.
Second, a chatbot company’s team should have highly-credentialed leadership, particularly with regards to data science. This is far from the only leadership requirement, but it is the one that cuts across the best chatbot providers in the financial services space. A top data science team will help chatbot providers differentiate themselves early on from a pack in which very few benchmarks for success are available. This can help startups gain a foot in the door with large customers who tend to be risk-averse and technologically restrained.
Third, a chatbot builder should not be outsourcing crucial processes. Startups that are relying on third-parties for key chatbot functionalities (such as natural language processing) are limiting their flexibility, their ability to learn and improve, and their level of differentiation. We would generally recommend not investing in these companies without extremely strong innovation and defensibility elsewhere.
Chatbot or chat-not?
Chatbots are only beginning to make their way into the financial services space, and the landscape is shifting fast. While the above can serve as a preliminary guide, we have compiled ten questions we think that every investor should ask themselves before choosing whether to invest in a chatbot to promote financial inclusion:
1. What problem is the chatbot solving?
2. Is a chatbot the best medium for this outcome?
3. Do the end-customers want to use chatbots for this purpose?
4. How will the chatbot go to market?
5. Is the chatbot reducing cost or increasing sales (or both)?
6. What operations does the chatbot need to perform to be effective?
7. What level of conversational sophistication does the chatbot need to serve your customers?
8. What data does the chatbot need access to?
9. How will the company measure successful interactions?
10. Is the chatbot replacing tasks that people love doing?
This is far from a comprehensive list of diligence questions. Its aim is to establish a starting point for investors to refine their own chatbot theses. In a space with so much talk but so few results, investors should be wary when choosing chatbots with which to chat.