AI!!
The buzzword is too hot in the market nowadays. Do you have a technical product or an idea? If it doesn't have AI in it, then chances are it's not going to be sold like hot cakes. That is how things have changed lately.
It is no wonder why me and my colleagues at Gelato want to see what AI can do in a niche department like customer support and service. And that is exactly what we did.
For a company like Gelato, which is in the market for production-on-demand, there are a lot of customer questions you need to answer. It can be related to products, queries about shipping and pricing, and so on and so forth. Thus, our customer support team is always happy to help with these recurring questions.
Let's take one example. A customer asked us, "Do you ship to Norway?" Now that is an easy question to answer if you have the knowledge written somewhere where you could refer to it and say, "Yes! As a matter of fact, we do." Following the same thread, the next question could be, "What is the shipping charge or production charge to Norway?" Now we are in the realm of tricky questions. You see, it depends on what kind of product you are asking for. For example, apparel can cost you less when compared to a picture frame. But still, a human looks into the knowledge base and sees what the rough estimate is for production in Norway after asking the customer a follow-up question, "Hey, which product are you specifically looking to ship?" After answering this question, the same human then needs to answer (maybe from a different customer this time): "Hey, do you ship to Sweden?"
Now you see the problem as well as the potential here, right?
There are much more complex queries a human needs to answer at Gelato. So questions like those above could be quite trivial at times. And an FAQ section does not solve the problem. Since keeping tags on each question based on countries and then keeping tags on all the product prices, etc. (After all, we have local production in 32 countries for fast and more sustainable delivery!).
This is exactly where AI could shine. And that is exactly what we have done at Gelato.
We were actually researching solving this problem, and voila, we hit the jackpot in GPTs introduced by OpenAI.
Now all you have to do is create a GPT. Put the right prompts and train it on a knowledge base (This is exactly why you need to keep documentation. sigh!). We asked the GPT to answer the customer questions based on the knowledge base we have. We used the new GPT 4 turbo version to answer the questions from customers, which, in our experience, was far superior to 3.5.
The implications were mind-blowing. There were instances where the customer asked for a human directly, but then GPT politely asked for the problem he or she was facing and was able to answer the question by taking a look at the knowledge base, saving the agent's and the customer's valuable time.
Though I must say not everything is a walk in the park, you need to prompt engineer the GPT since it is actually talking to a potential customer who is having questions or problems, so you need to be careful.
In the early stages of testing, we have seen that the GPT was not willing to transfer the customer to a live agent. Also, the cost of the 4 Turbo is quite high when you compare it with the 3.5 model.
The impact of the above change was astonishing. We were looking for a deflection of around 20–30% from our previous stats but got a whopping 50–55% when it comes to overall deflection when using GPTs.
AI was a term that was here for some time. But with the advancement in neural networks and big data, it seems like “AGI” is here, and OpenAI is playing a big part in it. And the value products like GPT is bringing into the table is remarkable. And it was a wonderful opportunity to work with it to bring in immense value to what we are doing at Gelato.
The takeaway here is that, yes, AI or “AGI” is here; there is no doubt about it. But rather than seeing it as a potential threat, see it as a tool that you can use. After all, it is said that learning to use fire as a tool made humans, well, humans.
Comments
Post a Comment