Subex says three AI POCs passed as HyperSense deployments sped up

Subex said three artificial intelligence (AI) proof-of-concepts (POCs) met all technical exit criteria in the last quarter, while its last two or three HyperSense implementations reduced deployment timelines by about 15% to 20%. The company is negotiating commercial arrangements for the POCs and has not reported conversion into revenue.

What did Subex say about its three AI POCs?

Subex said three customer AI POCs closed successfully on a technical basis in the last quarter after meeting the predefined exit criteria for each exercise. Management said its sales team is negotiating with customers to convert those technically successful POCs into commercial arrangements.

The POCs covered Architecture of Autonomy (AOA), agents, managed-services agents and AI use cases including social engineering, according to management. Subex did not disclose customer names, contract values, conversion dates or whether all three POCs would become paid deployments.

Management described a POC as a test with agreed exit criteria that specify what must be demonstrated. Meeting those criteria establishes technical validation, but commercial negotiations determine whether Subex receives an order and the eventual revenue profile.

How much faster were HyperSense implementations?

Subex said its last two or three HyperSense implementations were completed ahead of customer timelines, with deployment periods reduced by about 15% to 20%. Management attributed the improvement to the HyperSense Library of Controls, which uses reusable preconfigured controls to reduce repeated deployment work.

Management first described the measured improvement as almost 15% and then said it was seeing a direct 15% to 20% reduction in HyperSense deployment time. Subex did not disclose the original project durations, the number of customers involved or a standard deployment-time target.

The improvement is not yet portfolio-wide. Subex said work on a Library of Controls for its ROC stack is at an earlier stage than HyperSense, although that effort has started. HyperSense therefore has a quantified implementation-time result, while ROC does not.

How is AI changing Subex engineering output?

Subex said its AI software development life cycle (AI SDLC) is allowing core engineering teams to deliver more features in the same timeframe. Management illustrated this with a sprint that could deliver 14 or 15 features instead of 10, equivalent to 40% to 50% more features in that example.

Management distinguished AI SDLC from the HyperSense deployment-time improvement. AI SDLC is primarily used to develop product features, while the 15% to 20% HyperSense gain comes mainly from reusable controls and greater implementation repeatability.

Subex said the FraudZap agents announcement and the agent deal it won were delivered using AI SDLC. Management also said the company had not reached its internal benchmark numbers and was continuing engineering work, so the disclosure indicates improved feature capacity rather than completion of the programme.

What could POC conversion mean for Subex revenue?

Subex described its business as backlog-driven, saying 70% of revenue in a given year is recurring or backed by backlog at the start of that year. Any order resulting from the three POCs could add new business, but would not necessarily be recognised fully in the same period.

Management said roughly 40% to 50% of a typical deal's announced value is generally recognised during the first four quarters of implementation. It also said the proportion can vary, at times falling in a 20% to 30% range depending on contract structure, implementation scope and timing.

Subex said licence value is generally recognised during implementation, alongside professional-services work. After go-live in a customer's production environment, the remaining value is usually recognised on a straight-line basis through maintenance or services arrangements over a three-year or five-year contract term in management's illustration.

What is Subex building beyond individual AI agents?

Subex said its agent architecture is designed to support multiple agents and move workflows from detection to investigation. Management said the recently announced deployment was focused primarily on investigation, while manager agents and routers could direct work to appropriate agents.

Management said repeatability at scale is a key design requirement because some outputs can be audited. It used an example of completing 50,000 investigations accurately and consistently, but did not state that Subex has achieved 50,000 production investigations.

Subex said it is working toward a decision-assurance and economic-integrity layer. Management's rationale is that an agent may have proper authorisation and follow standard operating procedures (SOPs), but still reach a decision that is not economically appropriate; it also said governance and zero-trust security are part of that direction.

Conclusion

Subex has reported two distinct AI-related operating developments: three POCs cleared their technical tests in the last quarter, and recent HyperSense deployments were about 15% to 20% faster. The first is evidence of customer-level technical validation, while the second reflects reuse of implementation controls rather than a disclosed revenue or margin outcome.

The next issue to watch is whether Subex converts any of the three POCs into commercial agreements with disclosed scope or value. Management also plans to advance the Library of Controls work for ROC, where progress remains less developed than HyperSense, while continuing to improve AI SDLC output toward its internal benchmarks.