Idea Intelligence · b2b2c
AgentFloat: Mobile-Money Liquidity for Kenyan Agents
Working-capital and float rebalancing for the agents who keep M-Pesa cash flowing.
The problem
Mobile-money agents are the physical ATM network of East Africa, yet they operate on razor-thin float. An agent holds two balances: cash in the drawer and e-value in the wallet. When customers cash out heavily on pay-day, the drawer empties; when they deposit, the e-value runs dry. Either way the agent must close, lock capital, or travel to a bank or super-agent to rebalance. Each closure forfeits commission and pushes customers to a competitor down the road. Kenya has more than 300,000 active agents, and aggregators repeatedly cite float depletion as the single biggest cause of lost transaction volume. Global fintechs ignore this layer because it is operationally messy, cash-handling, and invisible from a Nairobi or London dashboard.
The solution
AgentFloat gives each agent a predicted float curve and a same-day credit line sized to their transaction history. By reading settlement and till data through aggregator and Daraja integrations, it forecasts the hour an agent will hit zero cash or zero e-value and pre-positions a short-term advance to bridge the gap. Repayment is swept automatically from the next day of commissions. The product surfaces as a lightweight USSD and Android app: an agent sees today's float health, accepts a top-up, and keeps the kiosk open through peak. Pricing is a flat per-advance fee rather than an opaque APR, which agents understand and trust. The wedge is operational data nobody else aggregates: minute-by-minute float behavior per till.
Why now
Two shifts make this newly buildable. First, Safaricom's Daraja API program and the Central Bank of Kenya's fast-payment standardization are exposing agent and settlement data that was previously locked inside MNO silos. Second, aggregators have professionalized: a handful now manage tens of thousands of agents with structured back offices that can sign data and lending partnerships. A decade ago the float problem existed but the data to underwrite it did not. Smartphone penetration among agents has also crossed the threshold where an Android companion app is realistic rather than aspirational, while USSD covers the long tail. The timing window favors a focused operator before MNOs build this in-house.
The moat
The moat is proprietary float-flow data and aggregator distribution. Once AgentFloat sits between an aggregator and its agent network, it sees repayment behavior, seasonal patterns, and till-level risk that no new entrant can replicate without the same settlement feeds. That data continuously sharpens the credit model and lowers default rates, which lets AgentFloat price below any generalist lender. Aggregator contracts are sticky because rebalancing logistics are deeply embedded in their operations. General digital lenders like Tala underwrite consumers, not the float mechanics of a kiosk, and giant MNOs treat agents as a cost center rather than a credit opportunity. The combination of niche data plus channel lock-in is hard to dislodge once a few aggregators are live.
How it makes money
Revenue is a flat fee per float advance, typically a small percentage of the advanced amount, paid out of the agent's next-day commissions. A second line is a revenue-share with aggregators, who gain higher uptime and transaction volume from agents who never close. As the data set grows, AgentFloat can sell anonymized float-demand analytics to MNOs planning agent expansion and to FMCG distributors who use the same kiosks. Unit economics work because advances are short, self-liquidating from a predictable commission stream, and repaid within days. The model scales with transaction volume rather than headcount, and each new aggregator adds thousands of agents at near-zero marginal acquisition cost.
How you'd build it
Phase one: partner with two or three regional aggregators to ingest historical till and settlement data, then build the float-prediction model offline to validate accuracy against known stockouts. Phase two: ship a USSD plus Android agent app showing float health and a manually approved advance, funded from a small debt facility. Phase three: automate underwriting and same-day disbursement through the aggregator wallet, with auto-sweep repayment. Phase four: extend to Tanzania and Uganda where the agent model is identical. Keep the team lean with one credit-risk lead, two backend engineers for integrations, and a field operations manager who actually rides along with agents. Avoid building a consumer brand early; the customer is the agent and the aggregator, not the end user.
Proof signals
Watch for aggregator pilots where agents using float advances show measurably higher daily uptime and transaction counts versus a control group. A strong early signal is repeat-borrow rate: agents who take a second and third advance within a week are validating both the need and the repayment mechanic. Default rates staying in low single digits would confirm the credit model. On the demand side, waitlists from aggregators wanting to onboard their networks indicate channel pull rather than push. Track whether FMCG distributors begin asking for the float-demand data, which would prove a second revenue line. Finally, regulator engagement that treats agent float lending as legitimate credit, not gray-market activity, de-risks scale across borders.
Cite this. Cancel Atlas Idea Intelligence (2026). “AgentFloat: Mobile-Money Liquidity for Kenyan Agents.” https://www.cancelatlas.com/ideas/agentfloat-kenya (CC BY-SA 4.0). Concept-stage analysis; projections are illustrative, not financial advice.