Airtel AMA: In-app AI Agent
Interaction design and visual identity for an in-app AI Ask Me Anything (AMA) — designed solo from strategic brief to final experience, currently in pilot.
The Brief
This project started with a strategic brief rather than primary research. Product and strategy had a clear ambition: evolve Airtel’s chatbot into an AI agent that could do things, not just answer questions—navigate the app, resolve issues, explain information, and complete transactions through natural conversation.
Underneath, the experience was envisioned as three cooperating agents: one to understand intent and tone, one focused on selling, and one focused on service resolution. The brief also established a set of non-negotiable experience principles: conversational and multilingual; act first, then narrate (“I’ve opened your bill,” rather than “I’ll take you to your bill”); keep responses short; present one next step at a time; and make every interaction dismissible.
My job was to turn that strategic ambition into an actual product experience—to define what an AI agent should look, sound, and behave like inside the Airtel app.
My Role
I was the solo designer on this project. Product and strategy defined the vision; I translated that vision into the actual experience—designing the interaction patterns across takeover, chat, and voice, prototyping how each mode would work within the existing Airtel app, and creating the assistant’s visual identity from scratch.
I owned the work end-to-end, from interaction concepts and flows through to the final prototype. The experience is currently in pilot, with the interaction design running against a live prototype. The underlying AI capability is still being developed—the internal LLMs and agent infrastructure required to power the experience at scale are not yet in place. The interface is only one part of making an AI agent real; the intelligence underneath has to catch up.
The Interface Decisions
Three decisions shaped the interface: where the assistant lives, how it behaves, and how it signals that it is AI.
1. An overlay, not another destination
The assistant lives in an overlay panel anchored to the top of the screen, rather than the bottom. This was deliberate: most Airtel journeys already have important CTAs fixed at the bottom, and adding another persistent element there would compete with the actions customers need to complete.
A fixed icon in the top-right of the navigation bar opens the assistant from anywhere in the app, making it accessible without taking permanent space away from the underlying journey.
2. A copilot, not a chatbot
The assistant was designed to sit on top of the existing app rather than replace it. It doesn't create a separate conversational destination; it uses conversation to help customers navigate and complete the journeys that already exist.
Once it understands a request, it can respond in three ways:
Navigate — take the customer to the page they need.
Redirect — move them to the relevant section within the current page.
Reconfigure — rearrange or surface information already on screen to answer the question directly.
The response depends on what the request actually requires. Crucially, the action happens first. A short confirmation then tells the customer what changed—rather than describing an intended action and making them wait for it to happen.
3. An AI identity without hijacking the app
The assistant also needed to be immediately recognisable as something different from the core Airtel interface, without becoming visually dominant.
I created an AI avatar built around a star-shaped icon with a gradient treatment. It deliberately breaks from Airtel's predominantly flat, numeric-first visual language, giving the assistant its own identity while keeping the surrounding app intact.
Walking Through It
The clearest way to see the concept end-to-end is the flow I prototyped for “buy a SIM for my college-going son”—a near-verbatim version of one of the brief’s own example queries.
In a traditional Airtel journey, this would become a browsing exercise: choose a plan, compare data allowances, select a SIM type, and work through the acquisition flow. The copilot starts somewhere else: with the customer's actual goal.
It interprets the request as a new connection for a dependent, with a likely student use case. It checks what the customer's account already qualifies for, narrows the options to one best-fit recommendation rather than presenting a comparison table, and then pre-fills the acquisition journey. The conversation doesn't hand the customer off to another flow—the conversation and transaction become the same motion.
Key screens from the “buy a SIM for my son” flow
I stress-tested the same interaction model through two other prototypes: buying a postpaid SIM and claiming a Jio Hotstar reward already attached to the account. They represent deliberately different jobs—one is a purchase, the other an entitlement—so the test wasn't whether the original flow worked. It was whether the same overlay, three response types, and visual identity could flex across fundamentally different customer needs without becoming a collection of one-off AI experiences.
Where This Stands
The experience is currently in pilot, not yet live. The interaction model, visual identity, and end-to-end flows are designed and prototyped; productionising them depends on Airtel’s internal LLMs and the underlying agent infrastructure required to execute them reliably.
Pilot: Replacing the existing chatbot with AI led agent
That dependency sits beyond interface and interaction design, but it is an important part of the story: the design is ready to run—the intelligence underneath is still being built. This remains an active product initiative, not a shelved concept.
