From Cron to Modern Data Stack (MDS): Dataflow Automation and Its Current State
The concept that makes the technological miracles of today possible are defined by data. Enormous amounts of data are collected...
You want to improve your customer service without hiring a small army of employees. That’s when you turn to the power of machine learning. After all, you see a lot of companies out there promising to deliver 24/7 customer service with chatbots and other AI-supported tools. To make smart choices, you need to sort fact from fiction.
Before we explore how to harness the power of machine learning for customer service, you might be wondering if this is worth it. Ocado, a UK company, used machine learning to organize a high volume of inbound customer service emails. In their case, the company started their project with 3 million customer emails. After that was loaded, the machine learning system was better able to classify incoming customer service requests and route them to the right place. That means fewer customers waiting for help.
Asking “can machine learning perform customer service?” is fundamentally the wrong question to ask. Customer service is the name of a department or a whole category of job roles. Instead, let’s get down into the details.
It is far more useful to examine specific customer service tasks. To do that, we need to create an inventory of the most common tasks. As a rule of thumb, only include functions that occur at least twice per month and make sure you develop a list of 10 tasks.
For a home appliance retailer, here are some of the customer service tasks you might expect.
Tip: Validate the customer service tasks you come up with by speaking with your front line customer service representatives.
Using the inventory of tasks you developed in the step above, categorize each task. If the task can be effectively completed with rules and standard operating procedures, classify it as “Rule-Based.” If the task requires significant judgment, empathy or management involvement, categorize it as “non-rule based.”
Here are some examples:
Most machine learning projects tend to be most successful when they have a significant amount of data and rules available. Take the example of checking whether a given product is in stock. That is an entirely rule-based task - execute a query against your database. If the inventory is low (i.e., less than three units), you might add a rule to phone the location to confirm it is in stock.
As you define the critical rules used to complete a given customer service task, you will run into some unusual situations. For example, how do you handle a customer who has submitted multiple refund requests in a month? In that situation, you may want to question the customer further before agreeing to their request. Fortunately, there is no need to have predefined answers to every possible customer service task or question. Instead, focus on the most common questions.
It is best to start small when you first get involved in machine learning. Therefore, we recommend that you choose one to three customer service tasks for your pilot test project. We recommend selecting tasks that tend to occur in high volume and provide value to customers — for example, giving order or delivery status information at a retailer.
You’re probably wondering how all of the thinking and analysis you’ve done so far connects with machine learning. That’s where the final step comes into play.
While machine learning for customer service has come a long way, using an off the shelf solution is still not good enough. You need to use the services of a professional software development company that knows machine learning and Big Data like Blue Orange. If you complete the first three steps outlined in this post before reaching out, you will have a much more productive conversation. We can set up automation for repetitive customer service tasks so your employees can spend more time on high touch customer needs.