Customer service is moving beyond chatbots that respond to a fixed set of prompts. Companies are now looking at AI solutions that can remember previous interactions, understand customer context and adapt their responses in real time.
Twilio sees this leading to a more personalised way for businesses to interact with customers. Twilio is a customer engagement platform that helps businesses communicate with customers across channels such as voice, SMS, WhatsApp and chat.
As businesses look to make these interactions more personalised, AI agents are increasingly using customer profiles, conversation history and real-time sentiment to determine how they respond, says Seth Familian, Global Head of Enterprise Strategy at Twilio.
Speaking to AIM at Twilio SIGNAL World Tour in Singapore, Familian says Twilio’s enterprise customers are already experimenting with AI agents that adapt conversations based on the individual customer rather than relying on a uniform experience.
“Our most innovative customers are varying it for every single one of the callers,” Familian says.
He explains that before an AI agent answers, Twilio can retrieve the customer’s profile and attributes, along with observations from previous conversations, and provide that information to the large language model.
“In that way, every conversation is different, and the agent is actually always personalised based on the customer context,” he says.
Nearly seven in 10 Indian consumers stopped using an AI agent because it failed to remember basic information about them or their previous conversations, according to new data from Twilio.
The company’s 2026 Customer Insights Series found that 68% of Indian consumers abandoned an AI interaction for this reason. Of those consumers, 5% said the experience led them to cut ties with the brand altogether.
Analysing Customer Interactions
Twilio is also using AI to analyse conversations as they happen. Its Conversation Intelligence product can perform real-time sentiment analysis for voice interactions and post-interaction analysis depending on the channel.
Familian says Twilio uses “operators”, which are pre-written prompts that define what the system should look for, including sentiment. Customers can customise those prompts to tune the type of sentiment they want to measure.
For voice, the analysis needs to happen almost instantly. For text-based interactions, there is more time to process the conversation.
The same approach can work across channels such as voice, SMS and WhatsApp. This allows companies to track how a customer’s sentiment changes across multiple interactions rather than treating every interaction as an isolated event.
He also sees the possibility of AI agents changing synthetic voices based on customer context. In an experiment involving Twilio’s partnerships with Deepgram and ElevenLabs, Familian explored whether the voice used by an AI agent could vary based on factors such as geography, age and the nature of the interaction.
Conversations Across Channels
Familian mentions four capabilities Twilio introduced earlier this year around AI-powered conversations, including Conversation Orchestrator, Conversation Memory, Conversation Intelligence and Twilio Agent Connect.
Conversation Orchestrator brings voice and text interactions into a single conversation thread associated with a customer. It can also activate another channel when required, such as sending a WhatsApp or SMS message after a voice interaction.
Conversation Memory works on top of Orchestrator by turning lengthy conversations into smaller observations that can be stored and retrieved later.
Rather than sending an entire conversation transcript to an LLM every time, the system distils it into smaller pieces of information. Familian says this reduces the amount of context that needs to be passed to the model.
“Conversation memory distils that discussion or that transcript into tiny little observations that are very size efficient and therefore token efficient,” he says.
Conversation Intelligence works during the interaction itself. It can detect sentiment, monitor governance and script adherence, and trigger automation when specific conditions are met.
For example, if sentiment drops below a defined threshold, a company could trigger another action such as sending a discount or coupon.
Twilio Agent Connect takes a different role. It provides the connectivity between the communication channel and the customer’s chosen LLM, allowing companies to bring their own model and voice technology.
“We are not providing the LLM. We are providing the connectivity between the channel and your LLM,” Familian says.
Multi-Modal Strategy
This modular approach also means that Twilio does not control the quality of the underlying LLM response. Familian says Twilio’s responsibility is to provide access to communication channels and move information between those channels and the customer’s chosen model.
“When it comes to the value of the LLM’s response, that’s entirely up to our customers,” he says.
He expects enterprises to increasingly use multiple models rather than relying on a single LLM for every interaction.
A simple customer query could be routed to a cheaper model, while a complex request requiring more reasoning could be sent to a more capable and expensive model.
“I actually think the future of AI and using these LLMs is to route these inquiries and these queries to different models based on the nature of the query,” Familian says.
The approach could allow companies to balance model capability with inference costs rather than sending every request to a frontier model.
Start with Narrow Use Cases
The question of trust remains a major challenge for AI agents in customer service. Familian says companies should not try to make AI agents capable of handling every possible customer request from the outset.
He mentions Philippine Airlines as an example. The company used Twilio’s technology to reduce support costs by 30% and cut customer wait times from minutes to seconds.
For him, the lesson is that companies should first identify a limited set of problems that AI can reliably solve.
Australian fintech Lendi Group is another example Familian mentioned. The company reduced loan processing from weeks to hours after building AI infrastructure around its operations.
Familian also pushes back against the idea that companies should try to achieve 100% automation in customer service.
He sees AI handling routine interactions while human agents focus on complex problems that require more time and judgement.
Identity Could Become a Bigger Challenge for AI agents
Familian identifies identity as an area that will become increasingly important as AI agents begin interacting with other AI agents.
Twilio already has identity and verification capabilities through products such as Verify and Lookup. The company also owns Stytch, an identity platform for AI agents that Twilio acquired last year.
The challenge, Seth says, will increasingly be determining whether an AI agent calling on behalf of a person is authentic and whether it has permission to perform the requested action.
Twilio expects identity to become a larger strategic focus in 2027.
What About ROI?
Enterprise AI budgets are growing, but Familian says companies now want to see clear returns from that spending.
Familian says enterprises initially went through a phase of simply experimenting with AI. They are now looking more closely at the costs associated with those experiments, including token consumption.
The same trend is affecting the build-versus-buy decision. Companies are increasingly weighing whether to build AI infrastructure themselves or buy ready-made software.
Familian says some companies that initially buy packaged software later realise they need more customisation, while others that build their own systems eventually find that maintaining the infrastructure takes more effort than expected.
Twilio positions itself between those approaches, providing infrastructure that companies can build on rather than requiring them to buy a fully packaged system.
Familian said Twilio’s platform remains usage-based, with customers paying based on how much of the platform they use.
India is a Key Market
Familian says Twilio is working with AI-native companies globally and expects to provide a list of Indian AI startups it is already working with.
He also points to Twilio’s large R&D presence in India and says the company is working with Indian brands as they expand beyond the country’s borders.
Familian cites Reliance as an Indian company using Twilio’s technology.
He says the company uses Twilio Segment’s Customer Data Platform to understand customer behaviour, identify patterns such as customers with a high propensity to return products and distinguish customers who could receive loyalty benefits or a better customer experience.
For Twilio, the broader opportunity in India is therefore not limited to deploying AI agents. It involves using customer data, communication channels and AI together to create more personalised interactions while keeping a close watch on the cost of running those systems.
Twilio says AI customer service agents save Indian consumers an average of 101 seconds per interaction, compared with 84 seconds across APAC. About 67% of Indian consumers said they are now more likely to choose an AI chatbot over a human than they were a year ago, including 25% who strongly prefer a chatbot.
Your reaction