As the recent research shows, the vast majority of FinTech companies and banks are ready to implement and take advantage of AI-based chatbots. Gartner predicts that by 2020, most conversations between financial service companies and users will be automated with the help of AI and machine learning technologies. Top FinTech and banking giants including Bank of America, Capital One, City, DBS, and MasterCard have already implemented robust enterprise-level chatbots and AI conversational tools to automate communication with users and relieve departments from the extra burden of handling conversations and processing inquiries.
In 2017, as one of our FinTech client’s loan portfolio grew 10-15% every month, the company found itself in the so-called “growth trap” when they lacked internal resources to process the ever-growing number of incoming user requests and inquiries. To answer them, our Client (whom I can’t name due to NDA) had to inflate the staff proportionally, which affected business profitability and efficiency. After a certain threshold, the service began to lag behind, as call center operators did not have enough time to process all incoming requests with the appropriate quality level. That is, the company was forced to constantly “put out the fire” instead of solving client issues systematically.
As a response to this crisis, they were going to use a traditional approach and hire a dozen of new call center operators and a team of lawyers to work with an ever-growing number of client requests.
However, after careful cost and resources analysis, they made a decision not to invest in new hires but invest in software development instead. As such, they decided to create and implement an AI-based chatbot and hired us to do the development job.
Our business analysts started with a thorough analysis of all incoming requests that were then divided into 3 broad categories:
1. General inquiries
Most clients reach out to get general banking information about how to take or repay the loan, the size of their current debt, etc. 90% of these requests can be processed with just a dozen of typical questions, which makes it extremely easy to automate and no AI is actually needed for this.
2. Legal requests and complaints
That’s a more complicated way of outreach, as such requests are often linked to negativity.
The stats analysis has shown that 80% of such requests have only 5 goals – 1) get details about the debt, 2) recalculate interest rate, 3) apply for payment delays, 4) terminate the contract and 5) withdraw personal data processing consent.
3. Non-standard requests
As a rule, they are related to the human factor on the client’s side. For example, a client has forgotten the password to enter their personal dashboard, or their personal data has just changed. Or a person was sent money through the contact system, after which he realized that there are no terminals nearby to cash out the transfer, and asks to send the amount to the bank card.
Chatbot helped solve most of the client issues with no human involvement
To work with general inquiries, we created a simple basic chatbot and deployed it to WhatsApp, Viber, and Telegram (as an MVP). It took us about 2 months to test and implement the solution.
The most challenging thing was to establish API communication between the chatbot and the messengers’ servers. Whenever the client asks for information from their personal cabinet (for example, about the amount of debt), the chatbot should be able to connect to the Client’s database and retrieve the required data.
At first, the chatbot could only respond to typical requests, check the user balance and connect a call center operator in case the user could not find the necessary item in the menu and began to write their own request. Two months later, we trained the bot to “speak” English with machine learning algorithms, so that it started to recognize keywords in the user’s message and respond to them. The biggest issue was that users would use jargon, slang, and abbreviations when typing their requests, so we used NLP techniques to make sure the chatbot understands them.
The chatbot was able to solve most of the pressing issues. Over six months, it has held more than a million sessions successfully for almost a quarter of a million users. Nearly 80% of the Client’s customers use this bot to communicate with the company.
Thus, today, the chatbot process 40% more calls than all other communication channels combined (including phone calls, email, and messengers). This enabled our Client to unload the call center by at least 1.5 times and increase the speed of incoming calls processing.
Besides, we automated legal requests and complaints processing with the chatbot. Previously, each client wrote a claim at their own discretion, so, despite the general nature, their form was always different.
As a result, our Client responded to more than 3,500 customer requests within the first 30 days thanks to the chatbot, which allowed them to save 450 man hours a month. They also saved a lot on paperless communication.
The measures taken allowed the Client to focus the efforts of its employees entirely on working with non-standard requests, which are most difficult to automate now.
Tips on AI chatbots development and implementation
Chatbots are most suitable for FinTech companies that deal with standard customer requests and provide online services via the web or mobile applications.
Technically advanced customers are well accustomed to chatbots in messengers and actively use them to communicate with brands and service providers. Also, youngsters and millennials are particularly loyal to all new communication channels, and they’d rather talk to a bank chatbot than an outsourced call operator with a strong accent.
Find out right away what kind of chatbot you need
You should immediately determine what type of chatbot you need to implement: a primitive assistant that can answer a number of general and basic questions and still requires human involvement to operate, or a more sophisticated solution that has access to your customers’ personal data and features robust security and encryption functionality. This will help define the size of investment in your custom chatbot development.
In the first case, creating a chatbot will take a couple of weeks and will cost around $2-4K. In the latter case, you’ll have to hire additional programmer staff or build a dedicated development team to deploy all backend integrations, implement machine learning algorithms and embedded security features, and design unique UX. The cost of such a conversational solution development will exceed $20K, but the benefits will be significant, and such an investment will pay off pretty fast!
However, in both cases, pre-built chatbot templates, reusable code, AI frameworks and libraries, and other ready-made tools will be helpful and will allow reducing the total cost of chatbot development.
Choose several chatbot solutions and test them
It’s best to take the time to test several different chatbot solutions internally before choosing the final one. This will allow you to take into account the advantages and disadvantages of the product compared to the competitors, whether UI is easy to use, etc.
Also, in the process of testing, your development team will be able to understand whether or not the solution is customizable enough to meet your most sophisticated chatbot requirements and evaluate technical support.
Launch your chatbot in low season
Even if you can’t wait to launch the chatbot service as soon as possible, wait until the season is low, i.e., when client activity is not so high. Do not rush to downsize your call center staff – you’ll definitely need them as a backup to help avoid technical issues that may pop up at the very beginning, such as a system failure or too many incoming requests that freeze up your bot.
Always ask clients to leave feedback at the end of each session
Program your bot in such a way that the customer can evaluate the quality of service after the end of the conversation. This will allow you to gather feedback, define areas for improvement, and generate valuable user insights to take your chatbot to the next maturity level.