Inside Airtel’s Converged Data Engine: Real-time intelligence powered by Aerospike
How Airtel built a Converged Data Engine on Aerospike to power Customer 360, real-time decisions, with millisecond latency for 350 million users.
Airtel serves more than 350 million customers in India across mobile, broadband, and digital services, making it one of the largest telecom providers in the world. With digital-native competition moving steadfastly towards innovation and customers expecting instant, personalized experiences, Airtel’s challenge is to evolve beyond its formative telecommunications roots, building the operational intelligence and system agility to serve all real-time interactions.
Here’s the scale Airtel has to handle, according to Vibhu Gupta, Airtel engineering manager, at the Aerospike Bangalore Summit:
Over 7.3 billion call data records processed every day (A range on par with the number of daily Google searches)
5 billion calls and SMS per day
1.2 billion customer communications daily (Which is approximately double the amount of messages shared on X, formerly Twitter)
1.8 million operational reports generated daily
For years, Airtel’s technology environment was split across silos, including Direct to Home (DTH), broadband, telemedia, prepaid, and postpaid. Each had its own data stacks and customer records. This fragmentation led to duplication, slowed down new customer onboarding, and made it nearly impossible to see a unified customer view. The result was limited speed and flexibility in launching new services, and a poor foundation for personalization.
To overcome these hurdles, Airtel redesigned its data platform, which it calls the Converged Data Engine (CDE). The CDE is intended to unify data, support real-time transactions, and deliver intelligence across the business. At the core of this system is Aerospike, the low latency database that helps Airtel sustain 100,000 transactions per second and maintain millisecond-level reads and writes consistently.
The breakthrough: Airtel’s Converged Data Engine
Airtel began with a move away from siloed systems. Instead of managing customers separately across business lines, Airtel created a common framework of “Buy, Pay, Bill, and Serve” to streamline how services are ordered, billed, and supported.
At the foundation of this platform is the CDE, which accepts data from multiple sources, including databases, APIs, files, and streaming platforms; manages quality and governance; and makes intelligence available to every application and channel in real time. The CDE provides the backbone for services such as its Customer 360 unified customer view, which delivers a single view of the customer, captures short-term behavioral context in real time, and gives Airtel a consistent, always-fresh view of customer activity across mobile, broadband, and digital services.
Gupta explained that the CDE doesn’t just unify the data; it “enables the teams to build real-time intelligence,” he said. “The result: more agile decisions, smarter decisions, and tangible business impact.”
Aerospike’s role at Airtel
Aerospike plays the central role in this architecture. Its low latency design supports Customer 360 and the context store. Together, these services help Airtel support everything from personalized recommendations to rapid outage communications. Aerospike provides real-time access to customer profiles that help Airtel make better decisions and support the predictive models Airtel uses for personalization. This combination of a unified data platform with Aerospike’s performance and reliability means Airtel moves quickly, personalizes interactions, and delivers services efficiently across its customer base.
But raw speed isn't enough. Airtel needs consistent performance. When a single customer interaction triggers dozens of parallel database calls (e.g., pulling profile data, checking service status, retrieving context, and feeding AI models), even occasional slowdowns cascade into noticeable delays. A recommendation that should take 50 milliseconds suddenly takes 500. Brief database disruptions amplify even further: a few seconds of downtime means millions of customers can't complete transactions, flooding call centers and eroding trust.
The system is designed for resilience and predictable performance to prevent these scenarios. Airtel runs Aerospike in a cross–data center configuration, with one cluster in Noida for real-time data ingestion and the other in Noida phase 2, about ten minutes away, for bulk data ingestion. It uses Cross Datacenter Replication (XDR) to maintain an active-active setup with just a five-millisecond lag between clusters. Real-time monitoring with Prometheus and Grafana identifies problems before they affect service. The system also uses parallel partition filtering to improve performance and Spring Framework integration to map data from a Java application onto the Aerospike Database and read it back.
Applying real-time intelligence at Airtel
The CDE powers real-world systems that affect millions of customers every day. Two examples highlight how Airtel uses Aerospike at the core of its Customer 360 unified customer view to deliver fast, intelligent responses: outage management and contextual recommendations.
Outage management within minutes
Outage communication is one of the most demanding real-time challenges for a telecom operator. A network element may suddenly fail, affecting thousands of customers in a region, and Airtel has just minutes to notify everyone accurately and consistently. The CDE orchestrates this, from detecting the outage to sending a message, using Aerospike-powered customer information:
It starts with the network stakeholder systems, which capture alerts whenever a network element fails.
These alerts are passed into Apache Spark streaming jobs, which map each affected network element to the associated DSL IDs in the customer master.
Once the mapping is complete, the data moves through a messaging layer to retrieve the registered telephone numbers linked to each account. These numbers become the delivery targets for communication.
From there, events are published into queues, typically Solace or Apache Kafka, where they fan out to downstream systems. At this point, two flows occur simultaneously:
Customer 360 updates: Outage states are written into Customer 360, which runs on Aerospike. This makes the outage information part of each customer’s real-time profile, accessible to the app, call center, or retail store within milliseconds.
Outbound communications: The same events flow into Airtel’s communication layer, which sends SMS or WhatsApp updates to customers.
“Customer communication is sent within five minutes when a network element fails,” Gupta said. “Customer 360 is being used both to push the outages as well as pull the customer profile or the registered telephone number.”
Customer 360 also drives detailed personalization. High-value subscribers may receive more apologetic, carefully worded messages, while other customers get a more straightforward update, all from the same Aerospike-based platform.
The design also handles dynamic outage management. If a planned outage is scheduled to end at 12:30 but extends to 1:00, the same pipeline re-triggers, updating Customer 360 with the new expected time and sending revised messages. By maintaining outage data in Aerospike, Airtel’s communications reflect the latest information across all channels.
The result is a system that supports millions of customers, notifies them within five minutes, and provides contextually aware messaging, all anchored by Aerospike’s ability to serve updated customer profile data in real time.
Contextual recommendations, not static rules
Airtel also changed how it recommends products and services. “Three years back, we used to have a static recommendation engine,” Gupta explained. “The channels used to take the customer profile, the customer demographics, or the customer history from Customer 360, and it had rules. ‘There are three KPIs, this is the value, pitch this.’ But this was not the way we were looking at the future.”
The new flow is built on the CDE. When a customer interacts through the Airtel Thanks app or a call center, the product recommendation service calls Customer 360 to retrieve demographics, history, and segment data, along with the context store for short-term behavioral signals such as recharge drop-offs. This information is written to a feature store in Aerospike and provided to AI models that generate the next recommendation. “This is the kind of intelligence I was talking about,” said Gupta. “Which helps in this particular model. Traditionally, this was not possible through a rules-based model. Now, it has become possible through this predictive AI or gen AI model, and it’s being powered by Aerospike.”
Previously, a customer reporting a network problem might get pitched broadband services, Gupta said. The customer relays something akin to: “I'm not able to connect to the calls, I don't have that basic functionality that telecom provides, and I come to the app or I go to the call center or I go to the store, register a complaint, and the person there tells me, ‘Sir, please take the broadband.’ What customer experience is that?”
That kind of contextually askew pitch can put off a customer because that wasn’t what they were looking for. Now, the app or call center tells the customer it knows about the service issue and when it will be rectified, and the customer might then be more amenable to more appropriate offers, Gupta explained.
By combining Customer 360 and the feature store, all built on Aerospike, Airtel aligns recommendations with the customer’s actual state. The result is service-aware interactions for more satisfied customers.
Results Airtel counts on
Gupta described striking Customer 360 performance numbers thanks to Aerospike:
100,000 transactions per second (TPS) or 50,000 TPS per cluster
1 millisecond read latency
10 millisecond write latency
500 million records loaded into Aerospike in 20 minutes
Nearly six years of uninterrupted uptime, with no downtime since deployment
Together, these results show how Aerospike helps Airtel keep pace with both the volume and velocity of today’s telecom, while providing the real-time intelligence needed to keep its customers happy.
Lessons from Airtel’s experience
Airtel’s experience shows the limits of static, siloed systems in today’s telecom environment. Without a unified platform and a real-time data layer, it was impossible to quickly personalize interactions or maintain the agility needed in a highly competitive market.
By designing its CDE and putting Aerospike at its core, Airtel moved from duplication and delay to real-time intelligence and satisfied customers.
The lesson is clear: Unifying data is only the first step. To compete, enterprises must deliver intelligence in real time, which requires a data platform that supports high throughput with predictable low latency and reliability.
Airtel’s next frontier: Advancing operational intelligence with AI and Aerospike
The CDE gave Airtel the foundation to unify data, support real-time transactions, and provide contextual recommendations. The next phase reinforces AI and generative models into this platform, with the goal of moving from simple real-time responses to operational intelligence.
Operational intelligence means using live data not just to communicate with customers, but to guide decisions across the business, from anticipating outages to personalizing service. With Aerospike at the core of Customer 360 and the feature store, Airtel is positioned to take this step.
“Traditionally, what was data engineering?” Gupta said. “Data engineering was about bringing all the data together and making it accessible. Now, what is the future? The future looks like turning data into intelligence, consuming it efficiently, and driving real business outcomes. Until and unless you drive the real business outcome, organizations will not be able to grow.”
