Communications API provider Twilio’s product road map is a direct reflection of the growing role AI is playing in digital marketing.
On Wednesday, the company announced an assortment of new marketing-focused capabilities, all of which are built on a new AI-powered product called CustomerAI. Kathryn Murphy, Twilio’s SVP of product, described CustomerAI as “a pervasive technology layer across all of our offerings.”
Conceptually, it functions similarly to Einstein, which is Salesforce’s software engine for integrating AI into its various apps.
The idea is that CustomerAI will underpin all of the different solutions within Twilio’s engagement platform, which includes Twilio Segment (the customer data platform it acquired in 2020), Twilio Engage (its activation platform) and Twilio Flex (its cloud-based software for running contact centers).
“This manifests itself as various new features and capabilities being added to each of these products,” Murphy said, “and that includes both predictive and generative AI.”
Feeding the flywheel
Although predictive and generative intelligence are related – and sometimes conflated under the wide umbrella of AI – they represent two distinct approaches with different goals.
Predictive AI uses algorithms to analyze large data sets and forecast future events, like looking at customer traits to predict future behavior. Generative AI, on the other hand, uses algorithms to create new concepts and content, such as text, images, code – and disturbing fake beer commercials.
Twilio is making a big investment in both, and one feeds the other, Murphy said.
Using predictive AI, Twilio can infer traits about a customer, such as their likelihood to churn or make a purchase. It analyzes their real-time interactions with a company based on structured data, like email opens or clicks, and unstructured data, like conversations with an agent in a call center.
This information then flows into the Segment CDP, where Twilio can apply large language models and natural language processing to generate new insights about a customer’s propensities. This can in turn be used to inform marketing and arm a customer service agent with personalized information about that person.
Lather, rinse, repeat, etcetera.
“You might think of it as a circle,” Murphy said, “but we call it the flywheel effect.”
AI in action
The predictive AI capability is now generally available within Twilio Segment, and there are also other CustomerAI-fueled features still in the works. Twilio is testing a tool called voice intelligence that uses an LLM to process unstructured voice data, automatically transcribe conversations and extract insights from speech and chats.
For example, someone might say the words “Do not call” while speaking with a customer service rep or mention that they did or didn’t like a recent interaction they had with the company. Marketers can add this information to the customer’s profile and flag it to reps for any future outreach.
To avoid pulling in sensitive information from phone calls, Twilio is testing a tool that uses AI to automatically scan for and redact personally identifiable information (PII). The voice intelligence and PII retraction features are both in public beta and should be generally available sometime in 2024.
In addition, Twilio is also working on a handful of generative AI products set to enter public beta next year, including generative email design and a generative “journeys” tool that allows marketers to automatically build customer experiences by telling the system what outcome they want to achieve, the channels they want to use and the audience they’re looking to reach.
But there’s a catch to all of this: security.
If AI models love anything, it’s massive amounts of data. And the more data you have – and the more it moves around – the greater the security risk.
Later this year, through a partnership with cloud-based data platform Databricks, Twilio will launch an open pilot of “zero copy architecture” that allows applications to pass data between systems for analysis – between their data warehouse and Segment, for example – without having to make a copy of it.
All the CustomerAI tools will be available a la carte once they’re released, meaning Twilio customers won’t also need to use Segment, Engage or Flex to play around with them.
“They’re designed to work together, but we also know that marketers need to opt in when they’re comfortable,” Murphy said. “They need transparency and control in how they adopt these things, and we need to honor that.”
