Recap & Recording: Should Your AI Agent Be Allowed to Invest Your Cash?

Nigel Owen

1. Start with controlled, read-only access

Nigel described how AI can already make live portfolio information easier to use. Treasury teams can ask questions in natural language, review current holdings, check available investment options, analyse historical positions and prepare reports. The AI inherits the user’s existing permissions, so it cannot access more information than the account already allows.

“It can only read the data your account’s already authorized to see.”

This provides a practical starting point for treasury teams that want to test AI without allowing it to move cash or change account settings.

2. Let AI prepare the work, but keep human approval

AI can connect information from treasury systems, ERP platforms, accounting software, bank feeds and investment platforms. It can prepare daily cash positions, identify upcoming maturities, suggest suitable investments and check whether proposals comply with the treasury policy.

Nigel stressed that the final decision should remain with a person, particularly when the transaction has a material financial impact.

“Should it be allowed to input and contribute and suggest to how you invest your cash? Absolutely. But autonomously, I don’t feel we’re there yet.”

Wilfred Veeris

1. Reliable data and forecasting must come first

Wilfred focused on the conditions needed before treasury can safely automate cash investment. Companies need clean data, centralised balances, real-time information and a cash forecast they can trust. Investing based on outdated balances could leave the company without enough liquidity for payments or other obligations.

“Your cash forecast needs to be clear. You need to have oversight on your data. Is your data clean, can you trust your data, and can you trust your cash forecast?”

AI can help improve the data and calculate cash buffers, but it cannot replace the need for a sound forecasting process.

2. Define limits before introducing automation

A gradual approach could allow AI to identify cash above a defined minimum balance and suggest where it could be invested. Treasury would still need to account for payment cut-off times, debt repayments, interest payment dates and the location of cash across different accounts and currencies.

Wilfred also stressed the need for human checks before a transaction enters the market.

“I would definitely say, try to build in checks and controls, like, for yourself.”

Adriaan Robertson

1. The change must fit the company’s governance framework

Adriaan explained that introducing AI into cash investment can change the company’s risk profile. A proposal should cover the implementation plan, available resources, counterparty due diligence, expected returns, stress scenarios and the way the change fits within the existing investment policy.

The required approval will differ between companies, depending on their structure and risk profile.

“Each has different and unique governance requirements that will require a tailored approach and also determine who is the ultimate approver.”

2. Decision support and execution need different controls

Adriaan made a clear distinction between AI that supports decisions and AI that executes transactions. Recommendations for cash investment may require validation and review. Payment proposals, bank instructions or other executed actions require stronger controls, including segregation of duties, a four-eyes principle and a documented audit trail.

“You need to have proper segregation of duties in place between AI and yourself. You need to have four eyes principle.”

His recommendation was to involve auditors early and build confidence through a phased approach before considering autonomous investment.