Catching angry or disappointed customers early and focusing efforts on turning them around can be very effective. You reduce the risk of them going on social media and slandering your company, which can be very harmful. If you manage to satisfy their needs, they can also become strong brand ambassadors. If nothing else, it's always nice to track the number of angry customers. Data is always good!
In this blog post we will set up a simple pipeline intended to alert you when a customer lashes out towards your customer support, enabling you to take actions as early as possible. We will use Outlook to simulate a customer support inbox, Labelf as the AI engine and Zapier to tie it all together.
We will start by training our model. The data consists of actual telecom support tickets. Our goal is to create a model that will label a ticket "angry" or "not angry". We first upload the data and create our model.
Once uploaded, we can start training the model! Our first recommendation seems to be a satisfied customer, we therefore add the label "not angry" and continue!
Next up, we seem to have found an angry customer. Let's add the label "angry" and label the data point.
Now we just continue to label data until we're happy with the performance. There are other blog posts explaining metrics, so we will not go into detail on that subject here. Let's just try our model with some inferences for now.
It seems to be working just fine! Let's proceed to connecting our model with Outlook. We first navigate over to the integrations tab and follow the link to the Zapier page by clicking the "CONNECT ZAPIER TO LABELF" button!
Next, we pick the Outlook+Labelf template.
We simply follow the instructions, i.e. connect our Outlook account, Labelf account and pick the model we've just trained.
We are nearly finished now! We just have to modify the template a bit. That's the cool thing about Zapier, it let's you connect two apps and customize the pipeline to your needs! For this demo, I simply want to be notified once an angry customer contacts our support. I could for example review the case and decide whether we should escalate or just let the customer be angry. There are obviously multiple ways of handling angry customers but this setup will be sufficient for demo purposes, you get the idea! Now, let's modify the template!
We just want to get notified if the model believes a customer is angry. We therefore add a filter after the step where we classify the email.
Next, we want to send an email to notify (in this case me) someone that an angry customer has reached out to our customer support.
That should be it! Let's try out our setup by sending a salty email to the "customer support"!
Oh boy, do I sound salty. Which our AI model clearly also thought! The email was forwarded back to my own inbox.
We can now chose to take actions or simply let it go. I think this guy deserves a refund. Otherwise he might go on a crusade on social media, which we clearly don't want!
This is a very simple case but outlines the possibilities! Maybe you are trying to find very satisfied customers instead. Would you like to find brand ambassadors? Or people that seem to be interested in buying your product? Your much more likely to succeed if your able to take action fast. Customize the solution towards your needs!
