How to Choose Which Process to Automate with AI Virtual Agents

When you decide to implement the first virtual agent in your department or even the whole company, it might be difficult to know where to start. Often you don’t get second chances with stakeholders, and to get their buy-in, your project needs to show ROI on the first try.  The stakes are high, so how do you get results quickly?

When we’re working with our clients, we always recommend the start small, see results first approach, meaning that it’s best to automate one process at a time, see ROI, and then progressively automate more processes. So how do we choose which processes should be outsourced to AI agents first?

Virtual agents will bring benefits everywhere you deal with repetitive customer interactions. This means being part of a particular industry or department isn’t the best indicator of your use cases. Instead, we take into consideration three main criteria to judge potential candidates for automation. Of course, they are used as a filter to spot out best bets, and then we look closer to see which are most suitable in terms of business goals, technical possibilities, and so on. These three parameters don’t need to be all achieved for the automation to be possible and profitable. But the processes that fulfill these criteria are generally the ones you should start with and then explore other options.

Criteria for process automation with AI voice assistants

First, let’s take a look at those three parameters

1. Monthly call volume for a particular call type

Before we dive into technical details, it’s essential to start with the most crucial factor — ROI. We need to check if the investment will return enough profit. A good indicator of that is the monthly call volume currently handled by live agents. A rule of thumb is that if a particular type of process brings at least 5000 calls a month, you will see good results. If it’s between 1000 – 5000 calls, there is still wiggle room, and if the other criteria are met, it can be profitable for you to automate it. If you get below 1000 calls a month, the process will be either really difficult to automate or unprofitable when automated.

2. How agents handle this call type?

Virtual agents are best at handling repetitive tasks. So to assess if a process is ready for automation, it’s important to find out if agents can stick to the script most of the time. And how many exceptions there are to the standard procedure of solving a case. The more agents can stick to the script when handling a particular call type, and the fewer exceptions there are, the better candidate for automation it is.

3. Does the process require empathy, creativity, or critical thinking?

Virtual agents work best when we design clear conversation paths for them, even complex ones.  But they still are far from being able to make out-of-the-script decisions. That’s why they shouldn’t be used for negotiations or areas where complex critical thinking is involved.

Examples of assessing automation-ready processes

Now, let’s look at two examples to see how we use these criteria in real-life:

Example 1

Industry: Insurance company

Process: Sales outreach

An insurance company wanted to use voice assistants to automate its sales outreach process. They have a large database of car owners whose car insurance is ending in a month, and they want to reach out to ask if the car owners wanted to compare offers.

Monthly call volume: More than 5000 calls

How agents handle the case: Mostly script, rarely exceptions

Does handling the process require empathy or negotiating skills? Often

The last parameter was a concern, as processes that require skills belonging only to humans like empathy shouldn’t be automated. But when we looked closely at the processes, it turned out that a big part of the conversation didn’t require those skills, and they were only necessary when customers were interested in the offer and wanted to go deep into details – which was when the call could be transferred to humans. In the end, the process turned out to be a great candidate for automation.

Example 2

Industry: Logistics

Process: Parcel tracking

A logistics company wanted to automate its parcel tracking process: customers provide their parcel tracking number, and the virtual agent returns the current parcel status.

Monthly call volume: Less than 1000 calls

How agents handle the case: Mostly script, rarely exceptions

Does handling the process require empathy or negotiating skills? Rarely

The only concern was the call volume. We had to check if the initial investment will make financial sense with a low monthly call volume. In this case, it did, as the process was straightforward and didn’t need much work in terms of virtual agent design, configuration, and integration with the client’s system.

As you can see, the three conditions don’t need to be met 100%. They are indicators that help you understand where to start and what process should be left to human agents.