Long before the Internet, mobile data plans, and messaging apps, there were simply telephone companies. The mission for these “telcos:” to provide great service for the good ol’ analog phone. 

Times have changed. Now, telecommunications companies offer up a diverse range of analog and digital services. Meanwhile, their customer service leaders are tasked with supporting an increasingly complex and fragmented user base. One tool that can help these companies cope with this challenge: AI-powered virtual agents. 

At Directly, we’ve been working with a Fortune 100 telco titan on a virtual agent-led automation initiative. The goal: to transition customers who dial phone support to AI-powered support via SMS instead. There’s some irony here: one of the original phone companies actually wants its customers off the phone, and there are good reasons for that. 

A diverse customer base needs more intelligent support

The client came to us because it wanted to test Directly’s intelligent automation platform for a segment of customers who use Android mobile phones. The company serves up many types of voice and data plans around the world to this group. However, it was having a hard time effectively supporting the volume of calls regarding the many variations of Android operating systems and hardware devices. Just how challenging is servicing the Android market? Enough that “Android Fragmentation” has become a well-known term in the mobile industry.

Beyond the specific Android challenges, there are a few trends that are driving this client and other businesses toward AI-powered support automation solutions:

–For businesses, the cost of phone support is higher than digital and self-service channels of support. (Report: The Inner Circle Guide to AI, Chatbots & Machine Learning.)

–Increasingly, people today prefer digital and self-service channels over the phone. 

–Interactive Voice Response systems (IVRs), while theoretically helping companies filter support calls to appropriate departments, are by some accounts hurting businesses more than helping — because the user experience of getting stuck in a seemingly endless loop of irrelevant options can be dreadful. 

Better customer experience (CX)

From the beginning, the client was clear that it didn’t want to completely eliminate its IVR system, but instead offer the option of getting digital support — via a virtual agent and SMS messaging — to customers who dialed phone support. So, working with our client, we identified a group of Android-related topics the pilot would support.

If callers utter phrases that match those topics — or, as we refer to them, “intents,” — we immediately send them an SMS message stating, “To connect you with an [client name] expert, go to [URL].”

Those that click on the SMS link are sent an automated response created by subject-matter experts, relating to their specific issue. To ensure the most successful messages for resolving issues are served up more frequently, we constantly test responses and optimize the user experience. 

Along with the content of the automated message, customers are asked if the answer was sufficient. If not, customers that choose to get more help are immediately connected to an expert — NOT the company’s live chat agent.

The expert intelligence layer — our secret sauce

When our client first came to us, the company already had been experimenting with offering Android users automation. But the experience was different — customers who clicked on the original SMS message were taken to a “Frequently Asked Questions” (FAQ) web page and never offered additional support.

Through the Directly platform, we help companies engage with their expert users as a means to improve support. These experts provide content and answers that address the most common problems, and this content becomes the foundation for automated messages that are matched to intent. And when the AI-powered automation doesn’t address a specific question, experts can also serve as live agents to engage with customers in real time. 

That expert intelligence layer is our secret sauce — and it made the big difference for this client, as you’ll see in the metrics below. 


There are four key metrics that were benchmarked in the initial automation program and then measured again throughout the Directly intelligent automation program: 

Customer satisfaction: The CSAT improved from 63% to 95% with Directly (a 50% increase). 

Containment rate: The number of customers who clicked on the original SMS link and had their issue resolved within digital channels increased from 23% to 38%.

Return calls: The number of return calls to customer support within 24 hours dropped by 22%.

Support cost: The cost per resolution of customer inquiry was reduced by 80%, dropping to $1.62. 


Any time we launch with a new customer, there’s a lot of value in what you learn during the process, through success and failure. Here are some of the key takeaways from this pilot.

For companies with legacy support channels, implementing AI and virtual agents (VAs) into your system doesn’t mean a complete reboot. In this case, we were able to plug our AI platform into our client’s existing IVR phone channel — and experience success. 

Narrowly-defined pilots are a great way to get started in AI and automation. Start small by taking a lean startup approach. Create a testable hypothesis (e.g. “Implementing a new solution will decrease support cost and increase CSAT”), then build, measure and learn. And what you learn will guide your next steps. In this case, our client’s hypotheses have been confirmed and the company is significantly expanding the program to cover additional areas of support. 

Words and messaging matter. Naturally, we get excited about testing technology. But one of the key differences between our pilot and the previous attempt at SMS was the wording in the very first message. Instead of asking people if they wanted to read an FAQ, we told them that they would be connected to an “expert.” And that wording alone increased the initial deflection rate from telephone support by 65%. 

Virtual agents are well-suited to be the first point of contact in a company’s omnichannel customer support. With this client, the VA sat within an IVR phone system and moved people to another channel of their choosing. The same approach could be applied to the web, mobile apps, and even home assistants like Alexa. Informed by machine learning and trained by human experts, with data and support layers hidden behind it, these virtual agents can efficiently resolve many questions, and they can gracefully and instantly hand off any complex issues to the right expert and medium. 

Learn more  

Interested in learning more about how you might get started with our AI-powered automation platform, set up a demo and let’s talk.