Showing a demo of a flashy, generative AI-powered customer agent is easy. Both newly formed startups and established tech leaders are offering them up—and the market is only trending upwards. The use cases are far ranging: personalized onboarding, delivery management, automating customer support, appointment scheduling, the list goes on. It almost sounds too good to be true —and, sometimes, it is.
Production is much harder to pull off than a canned demo, especially at scale. While it’s easy to dream up all the ways AI agents are going to streamline workflows and personalize engagement, the business world still needs proof of these concepts working in real life.
Twilio’s goal is to power the world’s AI agent builders to create experiences that offer businesses and consumers amazing moments. As such, they wanted to accelerate their firsthand knowledge of what the world’s AI agent builders would need to be successful in the wild. Twilio's Emerging Technology & Innovation team released a developer preview of an AI Assistants builder last year—which offers a framework for building and leveraging conversational AI for complex, customer-facing use cases. This project was testing and prototyping how businesses could connect data, communications channels, and LLMs for scalable, personalized engagement. Twilio took it further and tested their own business with an AI Assistant built on Twilio components.
Isa: The AI Assistant Prototype That Tripled Free-to-Premium Sign-ups
How do you scale customer engagement without any additional headcount or cost?
It sounds like a trick question, but that was the setup for the experiment, which asks, is it possible to create an AI assistant that’s customer aware, able to handle a myriad of different tasks, properly represent a brand and customer focus, and drive growth for the business?
Isa is an autonomous AI agent that was created using the AI Assistants builder, which has access to Twilio’s entire knowledge base. For the pilot, their self-service and growth marketing teams partnered with developers to ensure Isa delivered an exceptional customer experience and robust functionality. It was this combination of the business and technical perspective that Twilio believes made Isa so successful.
To start out, Isa would be responsible for 5 percent of incoming leads. The goal here went beyond screening and scoring; Isa’s ability to have a two-way dialogue meant it could engage with customers in an ongoing conversation, answering questions about billing, code samples, and everything in between. Isa also offered around-the-clock coverage: available 24/7, across 80 countries, and in 12 different languages. This gave Twilio the ability to offer bespoke customer support to every lead.
One new customer used Isa throughout their onboarding, asking follow-up questions that tied back to their specific goals and use cases (their conversation spanned multiple days and 70 emails). People that engaged with Isa also commented on the speed and detail of responses, saying, "I wanted to take a moment to express our appreciation for the exceptional service you have been providing to our company. Your responsiveness and attention to detail have not gone unnoticed, and they significantly enhance our experience working with Twilio."
The returns also spoke for themselves. Customers that engaged with Isa were 3x more likely to switch from a free to paid subscription. Seeing performance like this gave Twilio the green light to expand Isa’s scope; it currently handles 80 percent of inbound marketing leads.
To be clear, the point of Isa isn’t to replace sales. It’s to scale what sales, marketing, and self-service teams are capable of, from driving product-led growth to determining the exact right moment a prospect is ready to speak to a person on the team. Building on top of Twilio’s trusted customer engagement platform, they were able to create an AI agent powered by AI, data, and communications that could adeptly interface with customers and pull in sellers when the time was right.
How to Scale AI Assistants
In piloting Isa, Twilio learned a lot about what makes an AI agent successful and scalable. First, it’s real-time access to contextual data. This should be an aggregate of a person’s behavioral data, historical records, and the insights gleaned from conversations. The latter is especially important: Each interaction with a customer should be a learning opportunity that helps refine and iterate on the AI. This cycle of audits and analytics will sharpen the assistant’s intelligence, its ability to personalize, and its consistency across different interactions.
To leverage this contextual data, you need an interoperable tech stack. This ties into the larger conversation about open platforms or out-of-the-box solutions. Twilio prefers the open platform approach, as it offers the advantage of customization and adaptability (being able to connect with a diverse array of tools and systems). While prepackaged solutions offer convenience, they sometimes run the risk of creating data silos or limiting you to specific capabilities or channels. In short, your agent is only as powerful as the data it has to be effective and personal.
Then there’s the matter of data security and privacy. Twilio follows a privacy-by-design approach, which is crucial for scaling successfully. They created their AI Assistants builder with robust privacy and security controls from the beginning, considering questions like, what data can these systems access, process, and retain, as well as when should conversations be escalated to human representatives? Not only does this help protect sensitive information, but it also establishes customer trust, which is critical for widespread adoption.
What’s Ahead?
Autonomous, AI-powered agents will fundamentally change the way we work and interact with each other. Since their initial developer preview release of the AI Assistants builder, more than 1,000 Twilio customers have expressed interest in leveraging the technology for customer engagement. Through this project and customer input, Twilio has developed a strong strategy and offering to be the platform behind these AI Assistants. By delivering choice to customers—whether through voice and messaging capabilities, speech-to-speech functionalities, or ConversationRelay (available in public beta), which provides businesses with LLM flexibility for building AI-powered voice agents—Twilio is gaining a deep understanding of how to enhance its platform to power AI agent applications for startups, ISVs, and global enterprise customers.
To learn more on how to provide efficient, personalized customer experiences at scale and across channels with conversational AI, visit: https://www.twilio.com/en-us/products/conversational-ai.
By Kat McCormick Sweeney
Kat McCormick Sweeney is a leader on Twilio’s Emerging Technology & Innovation team, focused on exploring the future of customer engagement with bleeding-edge technology. She has previously held roles as an analyst at Tesla, and other roles at Twilio, including the chief of staff to the CEO, business operations & strategy, and enterprise and mid-market sales. Analytically minded and all about people, Kat is passionate about solving customer problems by working across different functions of the business. She believes experiences with brands should be delightful, and the only way to unlock that is with technology in collaboration with people who deeply care about the customer experience.
