Five Chatbot Fails and How to Avoid Them

An effective service strategy covers the many different ways that a customer will interact with their financial services: seamless solutions for phone calls, digital interactions, and more frequently in recent years, automation (ideally, all on the same platform). Through all these touchpoints, it’s more important than ever that banks and credit unions are able to take advantage of these channels to effectively collaborate, meaningfully connect, and create low-effort experiences. A chain is only as strong as its weakest link, so as the systems that manage customer interactions grow increasingly complex the pressure to ensure each channel is serving effectively rises. However, as many institutions introduce AI and chatbots for the first time, some are inadvertently doing more harm than good.

The right chatbot can make a notable difference in a customer’s digital experience while boosting efficiencies for the financial institution. If approached correctly, virtual assistants can solve problems and instantly answer questions, keeping the customer happy and support staff free to focus on high-involvement questions and interactions. If the chatbot isn’t well-designed (or is built on insufficient or unreliable data), however, it can negatively impact the customer experience, damaging customer loyalty and retention.

Below are five errors commonly seen in poorly designed chatbots and how financial institutions can avoid these missteps.

Jake Tyler of Glia on AI chatbots
Jake Tyler of Glia

1. The chatbot does not understand the question being asked

The AI models that power chatbots (and other systems) are powered by data, it is their lifeblood. They need high-quality data, and lots of it. Without this, chatbots will struggle to understand what users are asking. If a chatbot can’t understand what users are asking it, it won’t know which answer to give and when, often resulting in incorrect answers or no answers at all. Financial institutions should invest appropriately to train their chatbots to understand natural language, or leverage pre-built AI models specific to their industry to avoid this potentially costly mistake.

2. The chatbot can’t give a good answer  

Once the chatbot understands what a user is asking it then needs to provide an answer, or a relevant next step in the conversation. Most chatbot systems retrieve an answer from a prebuilt library, but there may be a gap in this library that prevents the bot from providing a meaningful answer. The answer may be too vague or not contain the information the user is asking for, preventing a satisfying conclusion from being reached. Time and effort must be spent in filling out a useful response library that fully answers the many specific questions users may ask of it.

3. The chatbot gives a lengthy, complex answer

In the current digital-first world, customers expect quick, direct answers, especially when the question is relatively simple; no one wants a five-paragraph response to a single-sentence question. Forcing users to mine through useless information to find relevant details indicates that the bot has difficulty delivering concise responses.

This most often occurs when bots are unable to hone in on granularity within wide areas of discussion, which results in long answers to cover a variety of different, smaller questions within a broader topic. Such inordinately long responses make the bot appear clumsy and unintelligent. It’s critical that bots understand user goals to surface the right responses, and this comes from not just recognizing wide topics but specific questions within those topics as well.

4. It can’t recall what was said earlier

Nothing makes a chatbot feel less human than forgetting what it was told a few messages ago. Poorly implemented solutions struggle to retain information shared with them by users, forcing customers to have to re-enter the same information multiple times to receive the help they need. If the customer must constantly repeat themselves, it is unlikely they’ll want to interact with the chatbot again.

5. It can’t escalate to a human when needed

There are many situations where speaking to a live human rep would either be more suited to the issue being had, or entirely necessary to proceed. An intelligent chatbot is able to recognize these situations and can seamlessly hand off a conversation to support staff when needed. Without this ability, transitions between these two support methods can become awkward and require repeated effort from the user. Well-implemented bots screen incoming questions and pass along the ones that benefit most from more high-touch service, saving time and resources spent answering more routine questions.

Overcoming the Pitfalls

Virtual assistants have become a critical tool as part of a wider service strategy. However, they must be well-designed, as a frustrating chatbot experience can be extremely detrimental to both the customer and employee experience.

In order to avoid these many pain points, financial institutions should look to chatbot solutions that are designed specifically for banks and credit unions. They should be trained on hundreds of retail banking workflows and millions of conversations, leverage high-quality data, and have strong domain expertise with proven conversational banking abilities.

By leveraging savvy, intelligent chatbots, designed for financial services and with the right training and priorities in mind, institutions can improve resolution times, lower costs, and provide more seamless customer engagements.

  • Jake Tyler

    Jake Tyler is the GVA Specialist of Glia, the customer interaction leader unifying Digital Customer Service (DCS), phone and automated self-service on a single platform. At Glia, Jake focuses on helping financial institutions leverage intelligent AI automation in their sales and support interactions, to help make banking easier and more convenient.