Considering a chatbot for customer service? Think again


Switch Innovation CEO Brett Terespolsky explains why you need human intelligence on your front line.

Over the last couple of years, we’ve seen and heard about the big drive towards artificial intelligence (AI) being a prominent component of businesses’ strategies, but I think it’s important to make sure that we understand AI before we implement it in our businesses. With the correct understanding, we can make informed decisions about what kinds of AI are relevant and which are, well, all media hype.

We’ve seen many articles telling us about the new bot that bank X or Y is releasing that will… answer basic support questions? This doesn’t seem like a great use of AI in business. I say this for two main reasons. 

First, I’m very focused on customer-centricity and so I think that we should have our best people interacting with customers to constantly learn from them and develop our businesses to serve the customer – after all, they’re the reason we’re in business. 

[cfocoza-cta slug=finance-indaba-2019]

Second, and related to the first point, AI is trained to do one thing really well, but in a lot of ways, it’s still a “stupid” machine. You will have seen and heard examples of AI getting things so wrong that the validity of that particular AI comes under question. The problem is that AI is trained on data alone and doesn’t (yet) have the ability to reason and rationalise. (If it could, this would be ‘artificial general intelligence’ – when a computer has the ability to think like a human.) Due to this, when you see an image classifier categorise a giraffe as a skateboard, it seems like it’s useless. 

The reality is that this occurrence is very seldom but when it does happen, it becomes apparent that the AI has no ability to reason. It also shows that the AI is extremely well-practiced in doing one thing so much that it can do it much quicker and better than a human, but it has no idea what it’s actually doing. So, if this is the interface to your customers, they can become very frustrated very quickly – a scenario we need to avoid at all cost.

What exactly is AI?
Before we can make the right choices on what AI to use and when, we need to understand what AI actually is. We keep hearing about AI, machine learning and deep learning, and these things seem to be used interchangeably, but they’re not actually interchangeable. Also, remember that thing … ‘big data’… that we’ve heard about over and over for the last decade? Well, that hype seems to have died down because, well, you know, AI. So, let’s clear a few things up first.

If you were to look at a timeline, you would see that from the mid-1940s to about 1980, AI was a huge focus in the scientific community. From the 1980s to the 2010s, machine learning was the “in” thing. Since then, we are hearing more and more about deep learning. So, what do all of these things mean?

The TL;DR version is that deep learning is a subset of machine learning and machine learning is a subset of AI. To put this another way, deep learning is a technique for realising machine learning and machine learning is a technique for realising AI.

There are lots of articles that you can read to find out more on this but the terms are somewhat self-explanatory: AI refers to machines having some form of human-like “intelligence”, machine learning means that the machine can start to “learn” by themselves, and deep learning is just the next evolution of machine learning and is loosely inspired by the information processing patterns found in the human brain.

I do want to make an important point here. Robotic process automation (RPA) is not AI. This is a set of steps or procedures that need to be taken to achieve something. There is absolutely no intelligence or training required to achieve this. It is nothing more than a flowchart.

Data is everything in AI
In order to train deep learning models, we need large amounts of data. This doesn’t mean you should purchase the best data warehousing system or the best big data processing engine. You might need to but if your data is not in a state to be utilised for deep learning, then this is just a large expense for no benefit. The investment should first be made in data cleansing. Once this is done, the data should be in a form that is usable for a particular use case. Only then will it be time to purchase the big data engine to process these large quantities of data.

It’s also important to be creative with the data available to make sense of it. As an example, the intelligent people over at Splunk had a stream of data showing mouse movements and clicks. This meant that they could have 5,000 to 10,000 events per user per page recorded. They took that data and built images showing mouse position as a line with the velocity of mouse movement depicted using a colour scale on the line. They then used different colour circles for left and right mouse clicks. This turned each session into one image, and they could use that data to train a deep learning model to detect fraud on sites using their system. Basically, they can quite accurately determine if the user is a human or bot in real-time.

AI in financial services
When I think about all the big financial service institutions and the amount of data they have, I truly believe that the possibilities are almost endless from an AI perspective. We could do everything from fraud detection to product recommendations based on app usage, spend, etc. And with all these possibilities available to us, why would we want to create a bot that can be so wrong some of the time, that it could cause us to lose a customer? Instead, we could create things that happen in the background that add enormous value to our customers without them seeing the errors that can happen.

Remember that AI cannot solve problems that we cannot solve by ourselves. We are training the AI and therefore we need to know how the problem is solved in order to “teach” the machine how to solve it. The clarity in the problem to be solved or question being asked is paramount before just building an AI to do something. Until we have artificial general intelligence, make sure that your customers are being serviced by your best people that can feed back into the organisation.

Related articles

Why social impact is a critical issue for CFOs

With South Africa among the bottom 20 percent of countries when it comes to social impact effectiveness, Kearney experts unpack how CFOs can align purpose with profit to improve the “S” in their ESG impact.

CFOs should be Road Runners, not a Wile E. Coyote, says Ray de Villiers

Future of work guru Ray de Villiers says that, as the role of finance teams changes due to generative AI taking over their number-crunching responsibilities, it’s up to CFOs to make sure their people understand what the new future will look like, and the power they have to impact it.