How is Fibe moving beyond standalone AI models toward an interconnected AI ecosystem, and what role does the seven-agent fraud detection architecture play in bringing together signals such as devices, IPs, banking behaviour and customer networks?
AI has supported underwriting, fraud detection and risk assessment at Fibe for years. FibeSense is our new AI platform that brings these capabilities together. It hosts our AI agents on a shared intelligence foundation, allowing insights from one agent to inform other decisions across the lending journey.
Fraud detection illustrates how this works. Seven specialised agents run in parallel on FibeSense, covering identity duplication, KYC, bank statements, company validation, employment, geographic consistency and, for returning customers, credit behaviour. A central orchestrator combines their findings into a single risk assessment, while Graph AI provides a view of the network around the application.
This approach supports faster decisions while retaining human review for flagged cases. Reviewers receive combined intelligence from the fraud and underwriting systems, giving them greater context to assess anomalies.
How is Graph AI helping Fibe uncover hidden relationships and coordinated fraud patterns that traditional fraud detection and bureau-based underwriting may miss?
An application can appear credible when viewed on its own, while its connections to other applications reveal a different picture. Graph AI helps us examine those relationships across Fibe’s customer base, using signals such as device IDs, IP addresses, GPS patterns and banking transaction links.
For example, several applications with different identities may share a device or have connected banking activity. An individual check may not reveal the wider pattern. Graph AI helps surface these connections so that we can identify potential fraud rings, mule networks and coordinated activity earlier in the customer journey.
Bureau information remains valuable for understanding credit history, but it does not provide the same view of these relationships. Graph AI adds that network context to our assessment. A shared signal is a reason to investigate further, rather than proof of wrongdoing on its own. We evaluate it alongside the other FibeSense findings, with human review for flagged cases
As lending decisions increasingly incorporate behavioural and digital signals, how does Fibe’s AI orchestration layer combine outputs from multiple models, and where do you draw the line between automated decision-making and human judgement?
Fibe’s AI orchestration layer brings together findings from specialised fraud checks and our AI Scorecard to form a consolidated assessment of each application. The fraud systems examine identity, banking, employment and network signals, while the scorecard assesses creditworthiness using close to 27,500 variables, with bureau information as one input. Bringing these outputs together gives us a broader view of the applicant and helps surface inconsistencies that individual models may miss.
Around 90% of applications are processed without human review. We draw the line at flagged cases, whether they involve a potential fraud signal or a credit anomaly. These move to a human reviewer with the combined FibeSense findings available, so the reviewer can assess the concern in context. The role of automation is to process information consistently and quickly; human judgement remains essential where an application requires closer evaluation.
What are the biggest engineering and governance challenges of deploying AI at millions of real-time financial decision points, particularly around explainability, data security, model reliability and regulatory compliance?
At this scale, the challenge is keeping AI dependable across millions of decisions made in seconds. This is a key reason we built FibeSense as a single platform, where controls applied at the platform level extend to every agent running on it.
Explainability comes first. A lending decision has to make sense to the customer, the credit team and the regulator, with each requiring a different level of detail. On FibeSense, each agent’s findings are recorded alongside the decision, allowing the key factors behind an outcome to be traced rather than reconstructed after the fact.
Data security is built into FibeSense from the start. Every model and agent operates within a strict data boundary, keeping customer data inside Fibe’s controlled environment, regardless of how many signals we use. Consent, data minimisation, encryption and strict access controls are built into the pipeline, in line with the DPDP Act.
Reliability requires continuous oversight. Every model on FibeSense is continuously monitored against live performance, while challenger scorecards run alongside production models to test whether a better approach exists. A new model is promoted only after it has proven itself, and fallback rules keep decisions consistent throughout.
Regulatory compliance brings these elements together. RBI’s digital lending directions and its FREE-AI framework make accountability explicit. For us, this means audit trails for every decision, bias testing before and after deployment, and human accountability for the policies our agents follow.
As AI becomes embedded in the core architecture of financial services, how is the role of the CTO changing—from building technology infrastructure to overseeing AI strategy, self-healing systems, governance and long-term business transformation?
The CTO’s role is expanding from building reliable infrastructure to shaping how intelligence supports the business. When AI influences fraud prevention, underwriting and customer experience, technology choices directly affect financial decisions. This requires closer collaboration with product, analytics and risk teams to ensure that AI serves a clear business purpose and remains dependable as we grow.
We are building FibeSense as a shared intelligence foundation and extending AI into technology operations. Our self-healing layer monitors system health and resolves low-priority technical issues without engineering intervention, allowing our teams to focus on more complex problems and improvements.
Governance is part of that responsibility. We process customer data and decision signals within Fibe’s internal environment and retain human review for flagged fraud and credit cases. As the CTO, we need to ensure these controls remain effective as our capabilities expand.
Looking ahead, the focus should be on building a foundation that supports new products and changing customer needs. I measure progress through the quality of decisions our technology enables, alongside its speed, security and reliability.