In today’s day and age, Machine Learning (ML) and Artificial Intelligence (AI) are buzzwords you hear almost everywhere you go – especially if you are being sold a new technology product these days. As a salesperson who sells AI/ML technology, I am very self-aware every time those words come out of my mouth. The more and more you use certain phrases, the more you dilute the value of what you really mean to say.
In the same way that we need to be very deliberate with our words, we also must be very deliberate in how we think about applications for AI. When we find a fancy new toy, there can be a temptation to use it whenever we can. But not thinking about the repercussions of our actions can be quite dangerous, and if we pour our efforts into non-practical applications, we are ultimately just wasting our time. So what is the best way to do proper customer discovery if you are trying to solve a problem – practically – with your AI development?
First and foremost, you should try to focus on smaller issues and work your way up rather than looking at generalized issues and working your way down. By narrowing down the scope of the problem, you can ensure that you have the right data to solve it. The more generalized the problem is, the more confounding variables at play that can lead to inaccurate results.
Second, you need to determine if you are focusing on a real problem and whether or not you have the right data to solve the problem. In the case of identifying a real problem, you can typically size this up by considering the opportunity cost to doing nothing, and balance that against the cost of developing or investing in a solution. More importantly, you need to have measurable data attributes that you can utilize to piece together a solution. In some cases, those data attributes are quite intuitive; in other cases, they are not what you expect, or they may not exist at all. In that scenario, you need to try to understand whether or not there is a way to implement practices to help you collect the data you need to build a solution.
All of the above can take a good deal of time, but it is the better way for an AI-driven organization to embrace AI. The end result is that after you do this, you can automate away the stressful work so that you can actively listen and focus on the more important parts of your job. This makes the company more successful.
But how do you go about uncovering these pain points? After all, people become accustomed to the things that are tedious/laborious and/or do not realize that automation can serve certain challenges for them. If you have a routine every day, you might have no awareness that your routine can become more efficient because you just accept it as a status quo. For example, many of us probably never envisioned thirty years ago that we might be able to increase our productivity by riding around in self-driving vehicles. That is because driving a car was such a firm part of our status quo that we never dared to dream big enough to imagine such a world. And yet here we are with autonomous vehicles.
One issue is awareness, and I just touched on that with the example of the self-driving vehicle. In the case of the work that I do helping call centers to embrace automation, it is asking what processes are manual, and then diving into the feasibility of automating those tasks. Some tasks – like speaking – we cannot imagine being automated. But other tasks – like transcribing information or having helpful information surfaced in real-time – can, indeed, be automated. The logical follow-up question then is to understand what is the cost of not automating those functions.
Another problem is mis-attribution. I use an example of this in my book actually, where I reference a hypothetical cable company who has lots of angry customers complaining about their cable bill. The leader of the call center likely perceives in this situation that they have an issue to contend with. The reality is, it is someone on the marketing team who is probably not doing a good enough job providing clarity around the company’s billing practices. The root of the problem is misunderstanding over the cable bill. In this scenario, the company starts automating functions but the problem still continues – and that is because the misattribution of the problem prevents them from realizing the root of the problem.
At Cresta, we are working to use AI for the good of people. We talk often about helping people and not replacing them. Personally, I get very excited about the conversations I have with customers because I know that customer service agents have a difficult job and helping them be successful feels like a noble cause. For every happy agent, there are dozens of happy customers. How could anyone expect agents to listen to every issue when they are in environments with micro-management on metrics? For them to succeed in that environment they really need to be actively listening instead of being stressed out. Automate away the stressful parts of their job and, well, they will be much happier. They actively listen and they get exactly to the right issue with exactly the right answer.
How do we at Cresta use customer feedback to inform what models we build? In my experience thus far, it is through well-intentioned and earnest discovery. We go directly to end-users and thought leaders to figure out what problems they want solved. We do not enter those conversations with a solution – we seek to understand the problems first. And most importantly, we never get complacent about uncovering real issues. We are relentless in our pursuit for excellence.