Not every application of AI creates the same level of value.
At the first level, AI improves personal productivity, helping treasury professionals interpret information and complete work more efficiently.
At the second, it contributes to process and control improvement, making recurring activities more consistent and effective.
At the third, AI provides decision support, identifying patterns, challenging assumptions and helping treasury anticipate what may happen next.
At the highest level, these capabilities contribute to strategic impact, supporting decisions affecting liquidity, funding, risk and financial resilience.
Moving up the ladder increases the potential value AI can create. But it also changes what is required to capture it.
Consider a familiar treasury example: cash accounts.
Treasury management systems already perform many fundamentals well: collecting bank balances, consolidating cash positions, monitoring accounts and supporting forecasts. Using AI simply to reproduce existing functionality creates little incremental value. The more useful question is: what can AI add beyond what treasury systems already provide?
At the personal-productivity level, AI could analyse an existing cash-position report, summarise material movements and direct the treasurer towards information requiring attention. The underlying information has not changed, but less expert time is spent finding and interpreting it. The value is primarily time and capacity.
At the process-and-control level, AI could analyse historical cash-account behaviour to identify unusual patterns, recurring exceptions or inconsistencies. For example, an entity may repeatedly forecast customer receipts earlier than they actually arrive. The existing system can show the forecast and actual cash position; AI can help identify the recurring pattern behind the variance. Treasury can then investigate exceptions more systematically, challenge recurring weaknesses and improve process reliability. The value has moved from individual efficiency to better processes and controls.
At the decision-support level, the focus shifts from understanding what happened to anticipating what may happen next. AI could compare historical forecasts with actual outcomes, identify recurring forecast bias and test alternative assumptions. If customer receipts regularly arrive later than forecast, for example, AI could assess the impact of adjusting future receipts to reflect actual behaviour. Would an account become constrained? When could that happen? What changes if a major receipt arrives late or a significant payment occurs earlier than expected? AI is no longer simply making an existing activity faster. It gives treasury an earlier and better-informed basis for action. The value comes from improving the quality and timing of decisions.
At the strategic-impact level, the perspective broadens. Instead of considering whether one account may become constrained, treasury can assess what patterns across accounts, entities and scenarios mean for the organisation's overall liquidity position. AI-supported analysis could show how different assumptions affect liquidity headroom, funding requirements and resilience. This could support decisions about the appropriate liquidity buffer, future funding needs, structural versus temporary liquidity requirements, and the organisation's capacity to absorb adverse cash-flow developments.
And the potential value has progressed from time saved, to stronger processes and controls, to better decisions, and ultimately to strategic impact.
If higher levels create greater potential value, why not simply aim for the top?
Because value is only one side of the decision.
Moving higher generally requires greater investment. Analysing an existing report may require little implementation. Identifying patterns consistently requires reliable historical data and repeatable processes. Forward-looking analysis requires stronger data, assumptions and validation. Strategic applications may require integration across systems and information sources.
Analysing the investment needed:
The last factor is particularly important. An AI application may offer significant theoretical value, but poor data, inconsistent forecasts or fragmented systems can create a substantial gap between potential and realisable value.
This brings the two sides of the framework together. The AI Value Ladder considers the potential level of value:
The investment assessment considers what is required to capture it:
The objective is not automatically to select the application highest on the ladder, nor simply to choose the cheapest or quickest opportunity. A personal-productivity application may create relatively limited value but still be attractive because investment is minimal and benefits can be realised immediately. A decision-support application may require considerably greater investment but deserve priority because it supports materially more important decisions. A strategic application may offer the greatest potential value but not yet be the right investment if the organisation lacks the data, systems or capabilities required to realise it.
The choice depends on the balance between potential value, strategic importance and the investment required to realise it. Value also changes the control requirement
As the potential value and impact of an AI application increase, so can the consequences if its output is unreliable. An inaccurate summary may simply require correction. At process level, an error could cause an exception to be missed. At decision-support level, unreliable analysis could lead treasury to act on inappropriate assumptions. At strategic level, the consequences could extend to liquidity, funding or risk decisions.
The required level of data quality, validation, control and governance should therefore increase alongside the potential impact of the application. This is an important part of value realisation: moving higher up the ladder requires not only greater technical capability, but sufficient confidence in the information and analysis being used.
Treasury may identify dozens of activities where AI could add value. Few teams will have the resources to pursue all of them. The Treasury AI Value–Investment Framework provides a way to make that choice: How much value could this application create? What investment is required to realise it? And is that balance compelling enough to justify allocating scarce resources to it? The most sophisticated AI application is not necessarily the best investment. Neither is the quickest win. The stronger choice is where potential value, strategic importance and the ability to realise it justify the investment in people, cost and time required to get there.
Jing Dong FCCA is a finance and treasury professional based in the Netherlands
