How mPokket scaled personalized lending for 20 million borrowers with Aerospike

For millions of young users in India, mPokket is the gateway to accessible credit, thanks to its instant loans that can be deposited into a bank account in minutes. 

But behind that seamless experience lies a challenge familiar to every fast-growing fintech: how to keep lending decisions both personalized and instantaneous as growth pushes the limits of infrastructure. With more than 20 million active borrowers, mPokket began encountering the real symptoms of scale: heavy concurrency, data scattered across multiple systems, and latency spreading across hundreds of microservices that each pulled from different stores for customer, loan, and behavioral data. 

At the Aerospike Bangalore Summit, Amitesh Bharti, associate director of engineering at mPokket, shared how the company used Aerospike to re-architect its data systems and overcome those challenges, which ultimately was as much a cultural shift as a technical one.

What real-time lending at scale requires

For mPokket, growth has always meant keeping real-time speed intact as volume and complexity increased. Every loan request, credit check, and repayment must happen in real time across millions of concurrent sessions.

As customer adoption accelerated, the same systems that powered early success began to experience strain. Each event, from a new applicant’s eligibility check to a borrower’s repayment confirmation, triggered multiple lookups across hundreds of microservices. At peak times, that meant hundreds of millions of daily reads and writes, each demanding sub-millisecond response times. Even a few milliseconds of latency could ripple into delays across the platform.

“When millions of customers are using your system at the same time, latency hits, and that’s where we had an opportunity as a tech team,” Bharti said. 

Before standardizing on Aerospike, mPokket evaluated multiple in-memory and NoSQL options, but none delivered the consistent low-latency profile they needed at scale. Aerospike’s Hybrid Memory Architecture gave them sub-millisecond reads and writes even under heavy concurrent loads.

From latency spikes to real-time decisions

The first step toward solving mPokket’s scale problem was architectural: shifting from reactive, database-specific fixes to a unified real-time data layer. The goal was not only to reduce latency but to enable every data-driven function (whether personalization, risk assessment, or fraud detection) to run from the same low-latency foundation.

By replacing multiple single-purpose stores with Aerospike’s single, scalable database, mPokket gave all its microservices a shared low-latency source of truth, allowing the system to sustain millions of concurrent reads and writes while keeping response times low.

Personalization that keeps up with scale

As mPokket expanded, personalization had to move beyond static segmentation. After all, every user touchpoint (such as a repayment reminder, a push notification, or an eligibility check) depends on up-to-the-second behavioral data. 

With Aerospike, mPokket can build and update highly dynamic customer segments in real time, using live behavioral signals like repayment history, spending patterns, and time since last disbursal rather than relying on batch processes. Teams can now adjust segments and campaign logic in real time, aligning offers with live user behavior instead of relying on batch updates.

A single, live view of every borrower

Customer information once lived across dozens of systems and databases, loan details in one place, repayment data in another, behavioral metrics elsewhere. Consolidating all of that into Aerospike created a 360-degree borrower profile accessible to every internal team in real time.

Collections teams can now view repayment histories instantly, customer support can resolve disputes without switching systems, and risk operations can spot early signs of delinquency before they escalate. That visibility has translated into faster resolution times, as reflected in recent user reviews.

Fraud detection that starts before the transaction

When it comes to preventing fraud, every millisecond counts. mPokket rebuilt its fraud detection pipeline to run directly on Aerospike’s real-time fabric. “Our idea was to have fraud detection as early as possible to safeguard customer assets and enhance trust,” Bharti said.

The Aerospike-backed rule engine now scores thousands of signals per second and flags suspicious activity before a transaction completes. This “shift-left” approach of moving fraud checks earlier in the transaction flow has caught more fraud early in the flow while protecting genuine customers from unnecessary transaction holds or re-verification loops.

Building operational rigor around a real-time core

Bharti also described how adopting Aerospike pushed mPokket to strengthen the engineering discipline around its broader data systems. Before standardizing on Aerospike, different workloads ran on separate data stores with uneven automation, monitoring, and recovery processes. Moving to a single real-time core gave the team a predictable foundation, and that stability became the catalyst for strengthening everything around it.

With Aerospike in place, the team rebuilt key operational workflows with automation at the center. Provisioning, configuration, and upgrades now run through controlled deployment pipelines instead of ad-hoc scripts, making rollouts safer even during peak lending periods. A unified observability layer tracks read-write behavior, node health, and replication status in real time, allowing engineers to spot anomalies early and tune performance without interruption.

To keep reliability consistent as traffic scales, the team introduced daily validation routines, including load tests, failover exercises, and synthetic transactions. These checks verify that Aerospike can absorb bursts, recover quickly, and maintain the low-latency profile the business depends on.

Bharti described this as a cultural shift rather than a tooling exercise. With Aerospike providing a single real-time foundation and a consistent operational playbook, the team moved from reactive firefighting to predictable, measurable resilience. Scaling new workloads or environments is now systematic instead of risky, and the impact shows in the metrics. 

“It’s transforming our performance. Uptime is up, incident rates are down,” Bharti shared.

Lessons from mPokket’s experience

mPokket’s experience shows how a unified real-time data layer can simplify personalization, fraud detection, and customer insights at scale. By unifying personalization, customer data, and fraud prevention on a single real-time foundation, the company has made its data systems easier to operate, scale, and evolve. It’s a lending platform that grows more intelligent with every transaction.

For fintechs, the takeaway is clear:

• Real-time performance and reliability can coexist: mPokket proved that architectural simplicity (in this case, a unified, low-latency data layer rather than layers of caches and sync jobs) can deliver speed while maintaining consistent performance.

• Operational resilience compounds: Automation, observability, and daily validation are what make growth sustainable. Each new control mPokket built, whether it was health checks, synthetic tests, or self-healing clusters, reduced risk while freeing engineers to focus on innovation.

• A unified data layer creates leverage: When teams such as risk, marketing, and operations work from the same live customer view, coordination accelerates. 

What’s next for mPokket

With Aerospike now serving as the core of mPokket’s data layer, the engineering team is expanding how it applies real-time insights across the platform. One area of active development is pushing fraud detection even earlier in the flow by layering machine learning models on top of Aerospike’s live event streams. The goal is to catch risk signals before they surface in the transaction path while reducing friction for legitimate customers.

The team is also exploring how real-time behavioral patterns can reduce credit approval delays and make lending decisions more responsive to borrower context. These efforts extend the same low-latency foundation that powers today’s lending workflows into new areas of the platform, from personalization to risk monitoring to other decisioning workloads.

“Aerospike is at the center of our database strategy,” Bharti said. “We’re going to use it for many, many things.”

Full session: From data to decisions: Transforming fintech with Aerospike

mPokket delivers instant, intelligent lending for 20 million borrowers with Aerospike at the core of its real-time ecosystem.

Watch the full session in the link below to learn how Aerospike helps mPokket generate thousands of personalized offers per second, consolidate customer data, and prevent fraud before it happens.