The Green Sheet Online Edition
October 26, 2026 • 26:10:02
Cash forecasting is becoming a treasury discipline - banks need to catch up
Treasury teams have spent the last several years building AI-driven, real-time forecasting into how they manage liquidity and working capital, treating cash position as something to be continuously modeled rather than periodically reported.
Organizations using AI-assisted forecasting now reach 88 to 92 percent accuracy on a 13-week horizon, compared with 60 percent for manual methods (see thestacc.com/blog/ai-cash-flow-forecasting/).
Physical cash represents a significant, largely untapped opportunity. Across most commercial banks, cash across ATMs, recyclers, teller cash machines, smart safes, kiosks and branches is still managed on historical averages and manual replenishment schedules. Since cash handling can account for up to 10 percent of a bank's total operating costs, even modest gains in forecasting accuracy translate into meaningful savings (see brinks-ams.com/how-cash-forecasting-technology-cuts-cit-costs-for-banks/).
The same predictive discipline treasury applies to digital liquidity now needs to extend to physical cash. This is not about removing people from the process. It is about giving the tellers, branch managers, cash center operators and cash in transit providers who run the cash cycle better information earlier, so their judgment goes further.
From isolated devices to one monitored cash cycle
The cash cycle today functions as a collection of separate machines, locations and stakeholders rather than one connected system. AI-supported forecasting starts by treating it as a single cycle to monitor, one where every device and person along the way works from the same picture of demand.
That cycle spans six cash points: ATMs and co-located ATMs, recyclers, teller-cash machines, smart safes, kiosks and bank branches, along with the people who keep each one running. Each generates its own demand signal: ATM replenishment, recycler and teller-cash machine replenishment and emptying, smart safe pickups, inbound and outbound branch flows, and change cash orders.
The forecasting opportunity is not just knowing how much cash a branch needs for the day, but the shape of that day, down to the denomination. Outbound branch cash tends to move in the morning, inbound cash in the evening, and many banks still work from instinct rather than data at this intraday level. AI does not replace that instinct. It gives it something firmer to stand on.
How AI supported forecasting helps banks balance the trade offs
Banks are already balancing three factors, often by instinct rather than continuous calculation: the opportunity cost of idle cash, the cost of cash in transit runs and availability, making sure cash is there when a customer or branch needs it.
AI supported forecasting gives banks a continuous view of that balancing act, rather than a periodic guess, while keeping final decisions with the treasury and operations teams who know local conditions best. In a proof of concept at a mid-sized bank, an integrated model helped cut annual cash in transit costs by nearly 40 percent and reduced idle cash by 29 percent, without any drop in availability (see tinyurl.com/tv6n3my7).
The same approach extends to peak demand planning, both recurring, such as the holiday season, and one time, such as the Olympics, where historical averages tend to fail the most. A model can flag an unusual spike days in advance, but whether to act on it remains a decision for the people who know the market best.
Reducing transport by matching supply and demand across the cycle
In markets that support recycling, the next step is matching surplus cash to nearby shortfalls before calling for a fresh cash-in-transit run. This works at three levels: between branches and nearby cash centers with surplus, forming a hub-and-spoke model for inter-branch and branch-to-cash-center transport; directly between nearby branches; and between cash centers themselves, enabling interbank cash selling and buying, sometimes called note holding transfer optimization, or notes held to order (NHTO).
The business case is concrete. Banks using advanced forecasting to plan routes and delivery schedules have reported cash transportation cost reductions of 5 to 10 percent, alongside cash inventory reductions of 20 to 40 percent (see www.chooseatm.com/blog/atm-cash-forecasting-tools). Every match still runs through a human check. AI can surface where transport is unnecessary. A person decides whether to cancel it.
Additionally, Cash-in-transit companies can plan routes across various bank and retail customers to increase the number of stops per route and thereby efficiently utilize trucks.
Using forecasting to reshape the cycle itself
Beyond day-to-day operations, the same forecasting approach can inform strategic decisions: where ATMs should actually be located and at what density, whether an ATM network is worth running in house versus outsourcing or pooling with other banks, and where recyclers and branches should sit based on actual deposit demand rather than legacy footprint.
This reframes AI forecasting as a network design tool, not only an operations tool. It is also where governance matters most. A recommendation to close a branch or relocate an ATM touches customers, communities and regulators, not just a spreadsheet. AI can detect the pattern. It cannot, and should not, make that call alone.
Central banks are already exploring this at a national level. A recent report of Bank for International Settlements noted that AI can analyze vast datasets to detect unusual patterns and support decision making, language that applies just as directly to a commercial bank's branch network. The cash center, in this framing, stops being just a processing hub and starts acting more like a sensor network, feeding back the intelligence leadership needs to plan ahead.
Imagine central banks would monitor the inbound cash to detect unusual differences in quality and deliberately distribute higher proportions of newly printed banknotes into these local spots increasing quality and lowering ATM disturbances due to bad banknote quality.
What AI actually delivers
The future of cash management is not a simple transition from physical to automated. It is a rebalancing in which continuous forecasting and human judgment coexist, with AI acting as decision support rather than a decision maker.
Cash operations improve first: demand and supply forecasting that reduces idle cash and cash in transit costs while protecting availability, with treasury and operations teams in the loop on what gets acted on. Network optimization and branch transformation follow close behind: smarter relocation of cash points based on real demand data, with decision makers, not algorithms, keeping the final say.
Banks already trust AI to manage digital liquidity, within a framework of oversight and accountability treasury teams have spent years building. Extending that same trust to physical cash is not a future consideration. It is already overdue. 
Robert Rose is the Strategy and Portfolio Director for Cash & Payments at Giesecke+Devrient, a global SecurityTech company. He focuses on driving new revenue streams, scalable business models, and operational efficiencies for banks, cash-in-transit (CIT) companies, and the retail cash ecosystem. For more information, please visit www.gi-de.com. Reach Robert on LinkedIn at linkedin.com/in/robertrose2.
Notice to readers: These are archived articles. Contact information, links and other details may be out of date. We regret any inconvenience.
