AI Business Editors Choice

How modern C-suites can deploy and govern AI agents that take action, not just advise

 

By Karli Kalpala, Chief Strategy Officer at Digital Workforce

AI is well underway in proving its value and speed. AI can assist in underwriting policies in insurance, can flag fraud across regulated industries, and investigate thousands of documents – all at a pace that the human workforce cannot match. But that speed can be either an advantage or a liability, depending on the governance beneath it. Modern C-suites who want to see success from their AI and agentic deployments cannot simply press “go” and expect results.

Join The European Business Briefing

New subscribers this quarter are entered into a draw to win a Rolex Submariner. Join 40,000+ founders, investors and executives who read EBM every day.

Subscribe

Financial and regulated industries are increasingly mature sectors in AI adoption, but the parameters within which AI operates must be clearly defined to scale to the next level. It’s vital to determine what the technology has access to, the guardrails it’s working within, the governance and compliance expectations – both for your business and for external regulators- and the escalation process that AI and agentic workforces must follow. One of the best opportunities for AI to prove its value in these regulated industries is through risk governance workflows such as control testing and contract certainty, where the evidence and data are vast. 

Typical risk governance is built around sampling of the underlying asset and, for each case, attaches countless data points and documents. While sampling may be representative of the risk exposure, governing the risk across the whole population is a task too big for any person, team, or organisation. Agentic AI can step into this capacity gap and provide coverage for a business.

Sampling has always been a capacity issue

Imagine you run a bank processing millions of transactions every day. You have a regulatory obligation to identify suspicious activity, but auditors generally accept that not every transaction can be tested because it would be impractical. Monitoring systems are set up to flag suspicious activity, and analysts then review those alerts. The challenge is that suspicious activity does not always trigger an alert. Fraudsters and criminals are quickly adapting their behaviour, and in isolation, an individual transaction can appear completely legitimate. The risk is that fraudulent activity may go undetected among transactions the bank doesn’t have the time or capacity to investigate.

When capacity scales from the server, rather than from a reliance on caffeine-fuelled employees, every document, agreement, or transaction can be tested against evidence, with the same attention as if it were the only sample happening that day. AI Agents can then take action by applying judgement based on a set of rules determined by an organisation, and escalating to a human level if further investigation is required.  Human workforces are freed to focus on higher-value work, and the deployment of Agentic AI in risk governance protects organisations from threats that can creep in, whether through intentional fraud or human error due to capacity limits.

The governance scaffolding exists, but a majority of firms aren’t keeping up

An IBM study surveying executives across EMEA reported that just 37% said their governance was designed to scale with agentic AI deployment, and that 80% had an AI-related incident reach their C-suite in the past year. Despite this, only 40% had updated their AI governance frameworks in response.

AI adoption is scaling, but turning that into impact is where there is still a sizeable gap to fill – and a solid governance framework will be the difference. Starting to deploy AI Agents to risk governance operations not only improves how you manage risk but also teaches the governance organisation firsthand how AI and AI Agents work. This informs them on how to create the right governance structures for all AI Agent initiatives across the enterprise.

AI, as it stands, has matured enough that there is sufficient structure in place to implement it and boost governance and compliance. C-suites that leave their governance to chance or fail to update it when problems emerge put their business in risky territory. A good first step is a single area, such as financial crime or credit monitoring, where there is plenty of evidence to test. Autonomous control testing leverages AI agents to continuously log and execute against internal controls, resulting in full population testing. 

Where AI makes the difference

To move closer to a reality in which AI takes action, rather than simply advising, modern C-suites should introduce guardrails and a clear workflow for the agents.

Take a CRO who wants assurance that the firm’s financial crime checks on transactions are working. The firm can deploy AI agents across three stages:

  • Evidence: An AI agent gathers the evidence, such as  KYC files or transaction records, and tests it against the firm’s internal controls.
  • Judgement: The agents check that the evidence lines up and give each result a green, amber or red rating. Every agreement can be tested against every piece of supporting evidence, with a full, auditable record kept as the agents work.
  • Action: The agent takes action on the results where no human intervention is required, and when a result is inconclusive, for example because of unusual activity that could point to financial crime, the agent analyses it and passes it to a person to make the final decision.

Any deployment has to earn its place in your organisation, and the best way to do that is to ensure it behaves predictably. That judgement sits with the human workforce – they set the parameters of who or what can make which decisions, and agents only interact with information based on those rules. 

Start small and scale

Robust governance should be built in from the start, and in risk governance, that means agreeing up front which evidence an agent can reach and own, which results an agent can close, and which require further intervention. With these decisions confirmed, the agents will produce an audit trail for each decision over time, creating a log that regulators can follow and boards and C-suites can trust.

For modern C-suites that want agents who do more than advise, it ultimately requires clear boundaries that are revisited regularly as technology accelerates. Learning and refining from these initial projects can both demonstrate AI’s ROI for an organisation and guide them as they make the case for a business wide, AI-native operating model.

EBM Newsdesk

The central editorial desk for European Business Magazine, delivering daily aggregated global corporate insights, market analytics, and macro-industrial updates.

Leave a Reply

Your email address will not be published. Required fields are marked *