3 Lessons Learned From Automating Customer Service With AI

Customer service chat automation is one of those areas where already today AI is used to benefit companies. Among large enterprises it is one of the prime targets for automation this year. Virtual customer agents (customer service-focused natural language understanding chatbots or VCAs for short) are already today capable of understanding natural language and help users autonomously i.e. without human customer support. Having built and deployed numerous such solutions over the past 2 years, I thought to share our learnings from working with customers in the finance and telecom sector. These three points come up at every meeting and in addition I’ve added a bonus point to give some guidance where this sector is moving in 2018.

Let’s dive in.

Lesson #1 - understand the real capability of VCAs

One of the first questions that customers want to know is how well the VCAs perform. And it is a fair question. From our experience a natural language based chat VCA is capable of understanding around 70-80% of questions that it gets asked on average. The exact number of course depends on the level of training and setup of the solution.

The VCA is autonomously capable of solving around 15% of user questions. This means that 15% of all the questions the VCA gets asked, it is capable of answering such that when it asks the user in the end “Was that helpful”, users say yes and the case is considered closed. 15% may seem low at first hand but considering the VCA is doing the exact same job as a human customer support representative it is quite a big deal. In addition, applying such a system to work at scale with tens of thousands of chats per month the benefits are obvious.

In this light VCAs should be seen as systems making human support more efficient and augmenting them. Based on today’s NLP capabilities VCAs will not be replacing human customer support anytime soon but rather should be seen as helpful assistants handling inbound requests, solving the simplest and most tedious ones and letting human customer support concentrate on solving more creative and meaningful problems.

Lesson #2 - building a good VCA takes effort.

Ideally, you can train the VCA with thousands of questions (complete with misspellings, slang and  grammatical errors) from actual users of the product/service. But the reality is that most companies do not have existing chat history data readily available for training. In that case, the options are to artificially generate thousands of different questions or to deal with the reality of not having much input data and hope to gather it when the VCA goes live. Neither solution is ideal but it reflects the reality when first deploying a VCA. It is advisable to take a long-term view and understand that building a quality VCA takes time. Even if all the training data is not there at the beginning it is possible to deploy a live solution and over time improve the VCA to a good quality level.

Lesson #3 - there is no magic AI to deploy out of the box.

With all the advances in machine and deep learning, most algorithms rely on largely pattern-based approaches to extract intent from a large corpus of previously seen chat history. Users’ questions to banks differ from questions asked to telecom companies — and there is no off-the-shelf algorithm to fit both cases automatically. An optimal solution is to use a host of different algorithms and fine tune the algorithms used to a specific use case. Such an approach provides users with more accurate answers over the long run.

Prediction - 2018 will see a sharp increase in companies looking to test the capabilities of AI on their own.

2017 was a great year of promise for AI of all the things that it could do. 2018 will be the year when companies will deploy solution on their own to see what AI can do. Even for all the hype, VCAs are capable of demonstrating gains and efficiencies as they work well at scale. And this is something that we see customer service managers and product leaders at large companies wanting to figure out. Newspaper headlines make for good stories but people want to experience the capability of AI. Expect increased interest to pilot a variety of solutions and measure ROI. Those that have a clear mindset from the get go and balanced view, will be able to benefit from what AI can bring to them.

Missed out on the Summits in San Francisco last week? Register now to receive exclusive access to post event presentations and videos.