Currently there is a revolution going on in artificial intelligence. ChatGPT is disrupting many markets including customer contact markets. What will be the impact of these robotic type of activities, is this really going to replace human interaction? This question probably is now in mind of every call center Professional.
ChatGPT can sometimes give better customer support than live agents! Thus ChatGPT will, especially once it is easily available through the search engine Bing, replace part of our customer contacts. How much, and how fast this will happen, nobody knows, but when you experience an unexpected drop in volume, you know in which direction to think. On the other hand, ChatGPT composes its answers using texts available on the internet. It cannot give answers to questions for which the answer is not publicly available, thus there is a limit to what chatGPT can do.
Will ChatGPT also change our internal processes? And, more specifically, can it simplify or improve WFM? Let’s put it to the test. The first question we asked is: “what is the best algorithm for call center forecasting?” We get a long answer that makes sense, it even suggests using linear regression including factors such as marketing campaigns, which is indeed a fairly good method. Its answer is not completely correct, it suggests also using logistic regression which is not appropriate for this kind of situations. This is a known issue with ChatGPT: it produces what is commonly said in similar situations, but there is no concept of true or false in the algorithm, there is no guarantee that the answer is correct. When we get more concrete and give some simple data containing seasonality and trend it simply does not recognize the components and produces wrong results. Also, when asked for staffing models it comes with an elaborate text, but Erlang is not mentioned. We conclude that ChatGPT is worth trying if we are looking for an answer to a typical customer support question, in a way it is a knowledge system that obtains its information from scraping enormous amounts of information on the internet. But if we want to be sure about the correctness of an answer, or when answers to closely related questions are not yet answered on the internet, ChatGPT is less useful. Thus for automating and simplifying WFM we have to consider different avenues. These avenues do exist: AI is more than ChatGPT. (What if doing nothing?)
Forecasting tools that do not even require the push of a button, scheduling tools that adapt themselves to agent preferences, automatic real-time management taking expected service levels into account… it exists already or will soon arrive. It is a combination of AI and more traditional method that makes this possible, and it might disrupt WFM just as ChatGPT is disrupting customer contact. What is needed is thinking in possibilities instead of constraints. It will bring us benefits for all stakeholders!
It’s all about the prompts, the strong point of ChatGPT is that it’s really good in coding, if you have some knowledge on the subject and ask your questions (prompts) very specific it will produce incredible coding results. It does know Erlang and a whole lot of forecasting techniques. I’ve created as a example a solid working prophet model with chatgpt which produces excellent results (still far from CCforecast’s awesome accuracy though). But that was not my goal. 🙂 But it’s an example, if you spend some days on the same code with chatgpt you can create almost everything.