Smarsh EDGE: Banks Pivot From Archiving Communications to Mining Them for Intelligence

Compliance departments at banks have historically viewed employee communications as regulatory obligations—data to be stored and retrieved when auditors came calling. That mindset shifted during the Smarsh EDGE event in New York, where speakers outlined how machine learning can now process months of email, messaging apps and AI-generated prompts to surface business insights. The complication: AI itself has become a communication channel requiring capture and oversight, including interactions between software agents.

The industry is entering a new phase where governance must be embedded from the start rather than added afterward. Many financial institutions are developing their own AI agents outside compliance systems and connecting to them via Model Context Protocol servers. Yet regulations remain technology-neutral, leaving banks responsible for whatever their agents do. A Smarsh Inc. study found that just 26% of enterprises believe their governance structures are keeping up with AI rollouts, even as financial services firms move past experimental phases.

I think that move from system of record to system of engagement to system of intent, it's really won or lost in one battleground, and that is data. Data is a huge moat for companies in 2026.

Gemma Allen, host at theCUBE, SiliconANGLE Media's livestreaming studio

Allen moderated conversations with industry figures at the event, examining how communications intelligence, compliance-focused language models, degrees of agent autonomy and cloud infrastructure partnerships are reshaping Smarsh's offerings.

1. Communications intelligence turns unstructured messages into client context

Smarsh is enabling customers to extract actionable intelligence from communications that were previously too messy to leverage, according to Goutam Nadella, chief strategy officer of Smarsh. Traditional customer relationship management software captures only what was documented about a meeting. Machine learning can now review a full year of conversations spanning email, WhatsApp, Teams and Zoom to recommend preparation strategies for upcoming interactions. Smarsh served as an early partner for compliance application programming interfaces from Anthropic PBC and OpenAI Group PBC, recording prompts so that problematic outputs can be traced to their origins. Nadella stressed that governance should be woven into systems from the beginning rather than treated as occasional checkpoints, citing the billions in fines financial institutions have faced for failing to retain business communications on WhatsApp and similar platforms.

2. Natural-language questions replace Boolean searches for compliance supervisors

Compliance officers who previously constructed intricate Boolean queries to detect market manipulation can now pose questions to an AI agent in everyday language and get back flagged conversations with risk ratings, according to Rohit Khanna, chief customer officer of Smarsh. A context layer that organizes identity structures—which Smarsh has patented—reduces the chance of the AI producing incorrect alerts. The firm also develops specialized compliance language models that consume fewer resources than general-purpose large language models, including a version that handles multiple languages and detects when speakers switch between them. Smarsh's S-Combinator innovation group, made up of 11 volunteers, created a model risk management agent currently in limited release. Khanna reiterated that governance must take priority.

3. Agentic automation scales with the risk of each workflow

Organizations can integrate Smarsh's agents with their own models and agents while the platform handles governance, data permissions and interpretability, explained Kamesh Tumsi, chief product officer of Smarsh. The company rolled out four or five agents over six months alongside its new MCP server. Since 70% to 80% of compliance budgets still fund human labor, the upside from AI deployment is substantial, Tumsi noted. He outlined four tiers of autonomy: assisting, recommending, executing within constraints and fully autonomous. Each process gets matched to a tier by evaluating whether it can be undone, its complexity level and how much visibility it needs.

4. AWS and Caylent help build governance into the infrastructure layer

Concerns about AI that dominated conversations a year ago have shifted to optimism now that supporting infrastructure is in place, Nadella said during a panel with Arym Diamond, head of data and AI go-to-market sales at Amazon Web Services Inc., and Peter Sankauskas, portfolio chief technology officer of Caylent Inc. Under a multiyear strategic collaboration agreement with AWS, Smarsh is embedding governance and security at the infrastructure level, responding to customer demand for visibility into who or what accesses their data. Smarsh has transitioned from a system of record to a system of engagement running on AWS, Diamond explained. Sankauskas added that AI requires equilibrium: authority without context yields poor results, while context without authority forces teams to constantly supervise.