Customer service: 3 Conditions For Successful Chatbot Implementations

Chatbots are not the silver bullet for every customer support organisation. While building Solvemate, we’ve discovered that there are three conditions an organization needs to fulfil for a chatbot to have a positive ROI. In this post, I’ll explain these conditions in detail and why they’re so important — so that you can see if your organization fits the bill.

Condition 1: High Inbound Support Volume

For automation to make sense, you need a relatively large support volume. In customer service there is inbound (=reactive, customer to contact you) and outbound (=proactive, you are actively reaching out to customers) support. This post focuses on inbound support, which support-oriented chatbots generally specialize in.

If your organization has less than 10 support agents, the cost savings aren’t significant and a chatbot probably isn’t worth the investment (unless your goal is to improve your response times, rather than saving costs).

It gets a lot more interesting with more than 10 support agents: Let’s assume your organization caters to consumers (i.e. a bank, telco, insurance provider, etc.), you likely have more than 100 customer support agents. In that case, you are likely handling between half a million to 1.5 million support requests every year. Running this service center costs between three and four million euros every year — potential savings are huge.

The best part: Support automation is nothing new to you. You will already have macros for your agents, well-maintained FAQ pages, and scalable processes in place to deal with peak demand. All of these will help you set up a chatbot.

Condition 2: Requests Are Significantly Repetitive

In order to train support agents, a larger organization will have training documentation: They help the new agents to learn the most common customer support requests and how to handle them. Call guidelines exist to help an agent ask the right questions and ensure the same answers are provided for the same requests.

With Solvemate, you train virtual agents based on these guidelines and common requests. Our experience tells us that, on average, there are usually less than 100 different solutions that make up more than 80% of all support requests. The goal of any support chatbot is to find the correct one out of the 100 solutions in order to solve 80% of requests.

If your organization has these kinds of repetitive requests, a chatbot can deflect a significant amount of support tickets by taking care of requests automatically. This creates time for your support agents to deal with the kinds of requests humans are best at.

Some companies have a lot of requests that depend on specific customer cases (e.g. on-premise software or complex business software). Those high-volume inbound support cases are hard to automate as the repetitiveness is missing and the answer depends on the customers’ unique setup or infrastructure.

However, let’s assume you have one million requests per year and roughly 100 solutions make up 80% of the requests. Let’s also assume an even distribution of solutions and requests. Then each solution is repeated 667 times a month. This kind of repetitiveness is perfectly suited for automation and cost savings.

Condition 3: Ability to Categorize Requests

When choosing a chatbot vendor, you need to make sure they can fulfill your requirements.

To answer this question, it’s important to first categorize the requests your customers most frequently contact you with.

First, you need to consider two types of requests: get-it-done vs. troubleshooting.

• A get-it-done request is one where the customer knows exactly what they want. They might not use the right words, but they can explain their wish to a human. For instance, phone companies receive get-it-done requests like “I want to cancel my phone contract” or or “I want to see a copy of my phone bill”. In those cases, the customers’ primary wish is to be served fast.
• A troubleshooting request is one where customers can only describe the symptoms but do not know what the problem is. It’s like a patient talking to a doctor: the agent needs to perform a diagnosis. Examples are requests like “My internet connection is slow and I don’t know why” or “I cannot login anymore.” In those cases, the customers’ primary wish is to get the right solution.

There are also two types of solutions: self-service cases vs. need-agent cases.
I have written about them before, but let me quickly recap it here:

• A self-service solution means that customers can solve requests themselves. A password reset, for example, is usually a 100% self service case. A chatbot can save 100% of human time in these cases, as customers don’t interact with your support team at all, relying only on the chatbot.
• A need-agent solution means that an agent is needed to perform the customers’ desired task, even after the bot has provided the correct solution (e.g. canceling an order). Not all costs are saved — unless your bot is connected to your backend systems and can trigger these actions via an API.

This gives us four categories of support requests:

Not all chatbots will be able to solve all cases. In order to pick the perfect chatbot provider, it’s instrumental that you’ve categorized your requests correctly and find out how the provider addresses them. In a future article I’ll write about Best Practices how to handle these four types.

Summary

Chatbots can seem like a magic solution to the problems most companies with customer support departments face. But success relies on a fit with the organization using them. We’ve found that you need:

1. a high support volume to achieve true ROI;
2. significantly repetitive requests to be able to successfully automate them;
3. a chatbot provider that can handle the specific type of requests your organization deals with.

Should all of this be true for your organization, a chatbot will certainly help you serve customers better and save a lot of money in the progress. Want to see how it works? Head over to this page to request a demo.

Footnote: There is more things to consider when building a support oriented chatbot, such as good CRM integration, detailed bot analytics, serving multiple channels like web, app and FB messenger or that the provider offers SLA’s etc. — but those aspects are nothing new, so I left them out here.