The rural agent still needs cash.

According to the GSMA's 2026 State of the Industry Report on Mobile Money, more than $2 trillion flowed through mobile-money rails worldwide during 2025, the clearest signal yet that the Global South built its own payment infrastructure while the rich world debated. From a central-bank tower or telecoms headquarters in Nairobi, that figure looks like the triumph of a weightless digital economy. Step off the tarmac onto a market street in Kitui, Homa Bay or Turkana on a Tuesday morning, and the weightless economy still hangs on whether the shopkeeper behind the counter has enough paper notes in the drawer and enough electronic float on the handset to honour the next customer's request.

National transaction totals measure velocity across the central switch; they say nothing about whether a rural mother can cash out a KSh15,000 school-bursary transfer when she reaches the kiosk. For the agent, mobile money is a two-sided inventory trade: every cash deposit requires handing over electronic float, every withdrawal drains banknotes from the till. An agent can be solvent on paper and illiquid at the counter within ninety minutes, and when that happens often enough, the digital economy quietly re-teaches a village to keep its savings in cash.

Tracing both sides of the till before the queue forms.

Walk through the morning ledger of an illustrative rural agent who opens at 8.00 am with KSh100,000 in working capital, split between KSh40,000 in banknotes and KSh60,000 in electronic float. At 9.00 am, a local builder withdraws KSh20,000 to pay casual labourers, leaving KSh20,000 in notes and KSh80,000 in e-float. Half an hour later, a produce trader deposits KSh25,000 in cash, shifting the till to KSh45,000 in notes and KSh55,000 in float.

At 10.15 am, a livestock buyer walks in to withdraw KSh50,000. Although the agent has KSh100,000 of total capital sitting in the shop, the transaction fails because she is KSh5,000 short in physical banknotes. Asking the customer to wait two hours in the hope that someone deposits cash pushes the liquidity cost onto him; locking the shop to ride thirty kilometres to the nearest branch pushes transport fares, lost trading hours, and robbery risk onto her. Neither of them failed. The float architecture did.

Counting the unit economics of the trip to the bank.

Agents do not move cash out of charity; they work for tiered commissions on completed transactions. When a round trip to a branch or super-agent rebalancing point costs KSh300 in boda boda fares and shuts the shop for two hours, an agent will rationally refuse a transaction paying KSh110 in commission unless replenishing the float unlocks a full afternoon of market-day trade.

Monthly average dashboards hide these local spikes completely. Rural cash demand swings hard on market days, tea-bonus payouts, Inua Jamii social-protection disbursements and school-opening weeks. An agent running three tills for different networks often has idle e-float trapped inside Network A that cannot fund a cash-out spike on Network B, because retail wallet interoperability is not the same thing as real-time agent float pooling. And when the formal pipes run dry, the market builds its own workarounds.

In a 2023 study based on 72 field interviews across Kenya and Tanzania, researcher Karen Sowon and her colleagues documented shopkeepers advancing informal credit out of grocery stock and customers leaving PINs and national ID details behind so the agent could process a withdrawal once cash arrived later in the day. Those fixes keep village trade moving while exposing both sides to privacy, fraud and credit losses. Banning the workaround by compliance circular without fixing the cash shortage simply kills the trade.

Giving the forecasting algorithm a working-capital partner.

Technology vendors love pitching machine-learning models that forecast rural agent liquidity shortfalls 48 hours ahead. In practice, a wall calendar marking market days, harvest weeks, and school-fee dates, paired with the shopkeeper's own memory, predicts rural cash spikes nearly as well as a neural network at zero software cost.

More importantly, a forecast is not a banknote. A chatbot that sends an agent a polite SMS explaining why she is out of cash is useless when she is staring at an empty drawer. A shortage alert adds value only when it triggers a funded response: a cash-in-transit van dispatched, the agent routed to a nearby wholesale merchant holding surplus notes, or a pre-approved short-term float overdraft unlocked, with strict velocity caps and ledger locks on any automated float-sweeping script.

Ahead of the first-quarter 2027 school-opening and harvest payment peaks, agency banking and mobile-money operations leads should run a 200-outlet rural field trial comparing algorithmic shortage alerts backed by pre-funded local rebalancing hubs against a simple calendar schedule, measuring success by the drop in turned-away customers and the net margin left in the agent's pocket after transport and financing costs. Two trillion dollars of annual flow is riding on whether that drawer has notes in it. Fund the drawer first; forecast it second.