Why post-trade operations are ready for agentic AI
The global shift to T+1 settlement has exposed a fundamental gap in post-trade operations: the right data and processes already exist, but executing them fast enough across hundreds of simultaneous exceptions is beyond what manual workflows can deliver. Writing in Global Custodian, Yogesh Shenai, Director of Product Management at Smartstream, argues that this is precisely the problem agentic AI has been built to solve. Rather than flagging exceptions and waiting, agents investigate, communicate, enrich, prioritise, and resolve autonomously – at a speed the compressed settlement window actually requires.
North American markets made the transition to T+1 in May 2024, with the UK, EU, and Switzerland following on 11 October 2027. Shenai notes that firms which have already transitioned found they had essentially gotten faster at the same manual tasks. Exception queues were not prioritised by settlement urgency, break investigation across multiple systems was too slow, and downstream workflows for counterparty notification and penalty tracking remained largely manual. Agentic AI directly addresses each of these failure points.
From reactive investigation to pre-emptive triage
Today, when a human analyst picks up an exception, they start cold – logging into multiple systems, reading email history, and reconstructing what happened. An agentic workflow changes that model entirely. By monitoring live trade data against known resolution patterns in real time, Smart Agents can identify signals that historically precede a break – a counterparty that consistently responds late, an asset class with chronic SSI delays, or a trade type where confirmation mismatches cluster – and trigger an investigation before the break is confirmed. The analyst receives a pre-built case and makes a decision rather than conducting the investigation themselves.
Email communication between operations teams and counterparties is also transformed. An agent can draft and send outreach, monitor for responses, read replies in natural language, assess sentiment and intent, and return that determination to the workflow as structured data. According to Shenai, this approach reduces manual touch rates by 60-80% across in-scope workflows, while an investigation that would take a skilled analyst 30 to 45 minutes can be completed by an agent in under two minutes.
Designing the human role deliberately
Agentic AI is not a replacement for human judgement – it is a redistribution of where that judgement is applied. Shenai is direct on this point: activities requiring human input should be deliberately designed as human-only from the outset, not treated as a fallback when an agent gets stuck. Relationship calls, counterparty escalations carrying reputational weight, and decisions with legal implications are examples where human authority is the correct design choice. Equally, agents are built to recognise barriers they cannot resolve – disputes requiring contractual interpretation, complex multi-leg failures, or third-party infrastructure outages – and escalate accordingly.
Beyond exception handling, the deeper value of agentic workflows lies in codifying institutional knowledge. The judgement experienced analysts carry – which processes to follow, which counterparties need careful handling, which exceptions are routine versus signals of something larger – can now be encoded into workflow design, inspected, and transferred. New team members inherit that knowledge from day one, and it improves over time as workflows are refined. Smartstream’s Smart Agents platform is built around exactly this model: policy-controlled autonomy, full auditability, and continuous learning from analyst behaviour and outcomes.
