Bhargab Mitra Put a Clock on the Network’s Next Failure

In 2001, India’s mobile market was preparing to reach roughly six million subscribers. Bhargab Mitra sat close to the expansion, managing Ericsson’s relationship with Bharti. The work involved account strategy, technology, partners, and a market consolidating at speed. A network was still something a company built outward: more licenses, more geography, more cable, more users. The operational question waiting behind all that growth was less glamorous. Once the equipment multiplied, how would anyone keep it working?

Mitra stayed with that question as both the networks and his job titles grew. At Ericsson he moved into vice president roles in sales support and technical solutions. At Nokia, his responsibilities included the Vodafone account in India, consulting and systems integration across Asia-Pacific, and global managed IT. He became, in the plainest sense, an operator of operations: the person accountable for what happened when layers of technology, contracts, teams, and customers met in the real world.

By 2015, the next mobile standard was visible on the horizon. Fifth-generation networks promised denser infrastructure, more connected devices, and a corresponding surge in things that could go wrong. Mitra saw a business in moving maintenance earlier. He founded Avanseus in Singapore around a practical idea: use machine learning to identify a likely fault before the fault becomes an outage.

“When I knew this 5G was coming in, I was determined to work in some area which could really address this particular domain.”Bhargab Mitra, 2020

The awkward sale of an event that has not happened

The first product was called Cognitive Assistant for Networks, conveniently shortened to CAN. It analyzed streams of network data and looked for patterns associated with future incidents. Avanseus said it could flag certain faults as much as 15 days ahead. The early warning was the hook, but prediction alone was not enough. An operator still had to decide whether signals were related, locate a likely cause, rank the impact, and choose an action.

This made the sales conversation unusual. Traditional monitoring software could point to a red light that everyone could see. Avanseus asked a customer to buy confidence in a red light that might appear two weeks later. Mitra acknowledged the skepticism directly. Customers, he said, looked suspiciously at a vendor claiming to predict something that had not happened. The only durable answer was evidence gathered inside the customer’s own environment.

The product therefore learned continuously. Mitra said the full cost benefit was expected after at least 12 months of use, when the system had absorbed enough operating history to improve its forecasts. By 2020, the company named Bharti Airtel, Vodafone Europe, and Brazilian operator Oi among its customers. It could deliver deployments remotely, a useful property when travel stopped and network demand rose. Existing clients expanded orders even as new buying decisions slowed.

A career measured by changing bottlenecks

There is a clean line from Mitra’s early account work to Avanseus, but it is not a straight line of technological inevitability. In 2001, the bottleneck was expansion. Operators needed coverage, licenses, partners, capital, and equipment. Two decades later, infrastructure had become abundant and complicated. The bottleneck had moved to comprehension. Operations teams faced torrents of alarms from multiple vendors and technical domains. A prediction without context merely joined the torrent.

Avanseus broadened CAN into what it called Augmented Operations. New releases added anomaly detection, topology discovery, cross-domain correlation, root-cause analysis, and health prediction. In a messy network, topology matters because one failing component can produce many downstream alarms. Grouping those alarms into a fault cluster can turn a wall of noise into one investigation. Rebuilding missing topology can give the investigation a map.

That distinction explains why Mitra’s public language kept returning to action. Avanseus partnered with orchestration platforms and software ecosystems because a diagnosis stranded in its own dashboard could not close the loop. A 2023 collaboration with Aarna Networks combined Avanseus predictions with an edge orchestrator for 5G and enterprise networks. The goal was dynamic, closed-loop work across network, cloud, security, and customer-experience systems.

It also explains the move beyond telecom. Data centers, manufacturing lines, utilities, and networks differ in their machinery, but they share an expensive condition: a critical asset can degrade before a team understands what the signals mean. Mitra described three ambitions for Avanseus: expand into those adjacent industries, automate more workflows, and make the machine-learning system increasingly self-learning and autonomous.

The route was not perfectly smooth

In 2018, Avanseus joined a Microsoft ScaleUp cohort in India, gaining the familiar stamp of an enterprise software company learning to distribute through larger ecosystems. Four years later, it pursued a much larger step. Avanseus agreed to a proposed combination with Fat Projects Acquisition Corp that would have taken the business to public markets. The process produced projections, valuations, diligence calls, and a long paper trail. It did not produce a listing. The agreement was terminated in November 2023.

That episode matters because company stories often compress the years between founding and acquisition into a tidy upward line. Avanseus’s line included a public-market plan that did not close, continued product work, partnerships, and the ordinary pressure of selling technical software to conservative infrastructure buyers. Prediction may be the product, but uncertainty remains the operating environment.

A larger home for the prediction engine

On February 24, 2026, Accenture announced that it had acquired Avanseus’s advanced AI solution. The buyer described models for prediction, anomaly detection, and optimization, designed to help coordinate decisions across network planning and operations. The technology would become a building block inside Accenture’s cognitive network platform, including work on agentic AI and increasingly autonomous networks. Financial terms were not disclosed.

For Mitra, the deal carried a precise logic. A small product company can develop specialized models and prove them in difficult environments. A global services company can place those models inside a wider platform, connect them to transformation programs, and distribute them across more operators. He called the acquisition an important next chapter that would give the technology global reach and resources for the next phase of autonomous network innovation.

The phrasing is revealing. He spoke about the technology rather than declaring the end of a founder’s journey. Avanseus had spent ten years moving a maintenance decision forward in time, then adding the context required to act on it. Accenture bought that accumulated operating logic as much as a set of models.

“Accenture’s acquisition of our AI solution marks an important next chapter for the technology we have built.”Bhargab Mitra, 2026

Mitra’s career now spans two very different telecom anxieties. The first was whether networks could grow fast enough to connect a rising market. The second is whether humans can operate the resulting complexity fast enough to keep it reliable. His answer to the second anxiety was not to remove the operator from the story. It was to give the operator more time.

That is the useful idea inside the Avanseus story. In infrastructure, intelligence has a clock. A correct answer after service fails becomes a report. A credible answer before the failure becomes a decision. Mitra built his company in the distance between those two moments.

What the next operator can steal

Mitra’s playbook begins with sequence. He did not frame the customer’s problem as a general shortage of intelligence. He broke the job into observable decisions: notice the weak signal, predict the incident, connect related alarms, locate the likely cause, estimate the business impact, and trigger the appropriate response. Each capability had a place in work that already existed. That made it possible to judge the software against an operational outcome rather than a model demonstration.

The second lesson is patience with proof. Infrastructure buyers are right to distrust a forecast that can interrupt maintenance schedules or redirect field teams. Avanseus had to learn from customer data, survive a year-long maturity curve, and earn confidence one deployment at a time. The company’s expansion into orchestration followed the same logic: once a prediction is trusted, the next constraint is getting it into the system where action happens. Mitra’s decade at Avanseus reads as a series of those constraints, discovered and removed in order.