Ocient co-founder Shantan Kethireddy says slow data is pushing AI agents to fabricate answers
Token costs and silent failures climb together when AI agents pull from fragmented systems, says the Ocient co-founder, and enterprise AI returns depend on fixing the data beneath the model
Shantan Kethireddy’s career began inside the machinery of a database. At IBM, he was a lead developer on the Db2 SQL query engine, became an IBM Master Inventor, and was later named North America Chief Data Officer for IBM Systems Group. He authored technical publications and set the course of study at Db2 conferences, teaching others how a system turns a question into an answer.
In 2018, he joined Ocient, a Chicago company founded two years earlier by the team behind Cleversafe, which IBM had bought in 2015 in a $1.4 billion deal. Ocient now builds AI that works directly on petabyte-scale enterprise data, so that agents, analysts and applications can get answers without moving data across fragmented systems. Eight years on, the questions Kethireddy deals with often come from software agents, and the answers do not always hold up.
“What we’re seeing directly with customers is that when agents have to pull data across fragmented systems, two things spike at once: token cost and silent failures,” said Kethireddy, now Co-Founder and Senior Vice President of Solution Delivery, Data Providers and AdTech at Ocient. “An agent gives you an answer that looks right but isn’t.”
“Speed matters too, and if an agent isn’t getting answers fast enough, it can start fabricating results rather than waiting,” he added.
“Since our founding in 2016, to our exit from stealth mode in 2022, to our introduction of the OcientAIQ ecosystem earlier this year, our growth can be traced back to Ocient’s mission: to help organisations that are doing the most consequential work in the world realise the full value of their data,” Kethireddy said.
Ocient’s telecoms customers apply AI to work such as unifying siloed usage data to prevent revenue leakage and keep billing accurate at scale, and to real-time network configuration and self-healing. Errors in those systems reach ordinary subscribers through the bills they pay and the networks they rely on.
Half of enterprises are still waiting for returns
Ocient’s own research puts numbers on the gap. Its October 2025 Beyond Big Data report surveyed more than 500 IT and data leaders managing workloads of 150 terabytes or more, and found that nearly half had yet to see meaningful revenue growth from AI investments. Respondents named data quality (55%) and the scalability of existing systems (48%) as the main barriers, 84% said AI had significantly increased the complexity of their data processing, and 46% reported difficulty finding skilled staff for AI and machine learning workloads. The study was commissioned by a vendor with a commercial stake in the data layer, and its findings should be read with that in mind.
“The push to prove ROI has brought renewed focus to the data foundation that human analysts and AI agents operate from,” Kethireddy said. “Delivering accurate, trusted answers on petabytes of data without ballooning costs or silent failures requires governance, security, and performance built in.”
“In AdTech, for example, a single agent query can turn into dozens of underlying queries as it samples, reasons, and executes, and audience sampling rates scale dramatically as businesses chase more signal,” he explained. “That volume only compounds the governance and consolidation challenge.”
“The dissonance between how technical teams measure success and how the business defines ROI is why we work with customers from the outset to align financial outcomes like revenue or cost savings, alongside the technical benchmarks that get them there,” Kethireddy added. “Enterprises are moving fast on experimentation, but the gap between that pace and true enterprise-grade governance is exactly where the ROI stalls.”
Costs move faster than the goalposts
“Infrastructure costs are changing by the day,” Kethireddy said. “At the same time, data volumes and duplicate data across heterogeneous platforms and business units keep proliferating. These phenomena are not new, but they are antithetical to a mission of unlocking insight and creating new business value from AI projects.”
Security and sovereignty are reshaping where that infrastructure sits. The share of leaders citing security as a top investment area rose from 55% in 2023 to 75% in 2025, and only 23% now prioritise cloud-only deployments, a drop of 13% from 2024. “Regarding goalposts, they will continue to move as we advance through the AI era, but fundamental enterprise needs remain consistent,” Kethireddy said. “Specifically, these include trusted data, strong governance, predictable performance, cost efficiency, and measurable business outcomes.”
Engineers belong inside the customer’s problem
“In order to help enterprises realise value from AI investments, innovation centres should be organised around customer problems rather than an assortment of technology platforms,” Kethireddy said. “They should use modular architectures and repeatable delivery practices rather than locking customers into a single model or toolset.”
“Multidisciplinary pods of forward-deployed engineers must bring their domain knowledge, experience with the ecosystem of AI solutions, consultative methodologies, data engineering skills, and proven deployment skills to identify high-value use cases, modernise complex legacy environments, consolidate data silos, and build unified semantic and data layers,” he said. “Each engagement must have a tightly defined scope, timeline, success criteria, and expected business outcome to avoid ballooning infrastructure costs and protracted project timelines.”
Ocient sells this approach as OcientAIQ, which combines industry-specific solutions with a Unified Data Platform that brings together data ingest, governance, analytics and AI workflows at petabyte scale. The platform includes native MCP support, giving agents direct access to governed, machine-readable data without moving it. “As costs of infrastructure skyrocket, enterprises should continuously evaluate new technologies, prove those technologies under pilots in controlled customer environments, and operationalise them with a focus on automation and self-service capabilities,” Kethireddy said.
Telecoms and defence are the proving grounds
The company has financed that expansion in stages. A $49.4 million round in March 2024 brought total invested capital to $119 million, and Ocient later announced a $132 million Series B, which concluded in autumn 2024, alongside the appointment of Henry Marshall as Chief Financial Officer. TechNexus puts Ocient’s total funding at $159.4 million. Founder Chris Gladwin is now Executive Chairman, while John Morris serves as President and CEO; Morris previously succeeded Gladwin as CEO of Cleversafe and negotiated its sale to IBM.
That capital remains modest against the largest players in the market. When the 2024 round was announced, Stephen Catanzano of TechTarget’s Enterprise Strategy Group noted that Databricks and Snowflake had each raised billions.
Ocient’s traction has concentrated in two sectors. In January 2026 it reported that its business with communications service providers had more than doubled, added former Ericsson CFO and interim CEO Jan Frykhammar to its advisory board, and signed new partnerships with Minsait Brazil and Clixer. On the defence side, it was assessed “Awardable” in the Pentagon’s Tradewinds marketplace in January 2026, listed in the AWS ICMP for the US federal government in August, and announced partnerships with Siren in July and Shadow Nexus in September.
“Enterprise AI adoption is accelerating, but the conversation has now sharply turned to ROI,” Kethireddy said. “The pattern we keep seeing is that AI success depends less on the sophistication of the model and more on whether the underlying data architecture can give agents accurate, fast answers from a single source of truth rather than assembling them from different systems.”
