Digital banking assumes a smartphone, a data connection and the confidence to use both for money. Where any of those is missing, the gap is not closed by a better app. It is closed by a person — a local shopkeeper or agent who handles the transaction on someone's behalf.
What assisted banking covers
The common transactions are cash deposits and withdrawals, balance enquiries, money transfers and bill payments, carried out at a local outlet rather than a branch.
The outlet is usually a business that already exists and already has footfall — a kirana store, a mobile recharge counter, a stationery shop. The banking service is added to it rather than replacing it.
- Cash deposit into an account, which is the transaction a branch is furthest away for.
- Cash withdrawal, often against a biometric or OTP authentication.
- Balance enquiry and a printed or spoken confirmation.
- Domestic money transfer, frequently the reason someone walks in.
- Utility and bill payments, which bring the repeat footfall that makes the counter viable.
Why the agent model works
Distance is the obvious part: a branch twenty kilometres away is effectively a day's travel and lost earnings, while the shop at the end of the road is not.
Trust is the less obvious part and matters more. People hand cash to someone they know and will see again tomorrow. That existing relationship does work no interface can do, which is why agent networks grow through local reputation rather than advertising.
What the agent is actually providing
It is easy to describe the agent as a distribution point, which understates it. They are providing three things a branch provides and an app does not.
- Physical cash handling, which is the entire point for customers whose income arrives in notes.
- Interpretation — explaining what a balance means, why a transfer failed, what to do next, often in the customer's own language.
- Accountability with a face. If something goes wrong, there is a person to go back to, which is a different proposition from a helpline.
What makes a network hold together
- Settlement speed. Agents settle cash against their own float, so how fast they are settled directly caps how much business they can do.
- Authentication that works first time, because a retry in front of a customer is a reputational cost, not an inconvenience.
- Commission worth the counter space, or the service quietly stops being offered in favour of something more profitable.
- Support in the agent's own language, reachable while the customer is still standing there.
- Clear, immediate failure messages — an agent who can say why something did not work keeps the customer's confidence; one who can only say 'it failed' does not.
The float problem, concretely
An agent accepting a cash deposit is lending the network money until settlement. Their capacity is bounded by their working capital, not by demand, and that constraint is the single biggest determinant of how much a counter can transact.
| Settlement speed | Effect on the agent |
|---|---|
| Same day | Float recycles daily; capacity limited only by demand |
| Next day | Workable, but one day's takings are always tied up |
| Two or three days | Capacity roughly halves; agent starts refusing larger deposits |
| Unpredictable | Agent cannot plan, and will prioritise other business |
The last row is the one that quietly kills networks. Agents do not usually resign; they just stop offering the service when a customer asks, because they cannot tell whether their money will be back in time for their own supplier payment.
The inclusion that follows
The transactions are small and routine, and that is precisely the point. Access to cash-in and cash-out near where people live is what makes a bank account useful rather than nominal.
An account nobody can conveniently reach is not inclusion. The last mile is where that distinction gets settled.
What the operating model demands
An agent counter looks simple from the outside and has a set of requirements that decide whether it works. Most networks that struggle are failing on one of these rather than on demand.
| Requirement | Why it binds | What happens when it is missing |
|---|---|---|
| Working capital | Agents fund deposits from their own cash | Larger transactions get refused |
| Reliable connectivity | Authentication happens online | Customers are turned away at the counter |
| Fast authentication | Biometric or OTP, in front of a queue | Customers go elsewhere; the agent loses interest |
| Prompt settlement | Float recycles into capacity | Capacity halves or worse |
| Clear failure messages | The agent has to explain, not just report | The agent carries blame for the network's problem |
| Local-language support | Problems resolve while the customer waits | Unresolved issues become lost trust |
The pattern across the table is that every requirement is really about the agent's standing with their own customers. A network that treats agents as endpoints rather than as businesses with a reputation to protect will keep losing them without understanding why.
Measuring a network honestly
Agent counts are the headline metric and the least informative one. A network of ten thousand registered agents where two thousand transact in a given month is a two-thousand-agent network with an impressive slide.
- Active agents, not registered ones, over a defined window — monthly is usual.
- Transactions per active agent, which tells you whether counters are viable or merely open.
- First-attempt success rate measured at the counter rather than at the server. A retry that eventually worked was still a failure in front of a customer.
- Time to settle, and its variance. The variance matters more than the mean, because unpredictability is what makes agents stop offering the service.
- Agent churn, with reasons. Agents rarely resign, so churn has to be inferred from activity rather than waited for.
Who becomes an agent, and why they stop
Agent recruitment is usually framed as a distribution problem — how many counters can we sign — when it is really a retention problem. Signing an agent is easy and cheap. Keeping one transacting eighteen months later is neither, and the reasons they stop are consistent enough to design against.
The typical agent already runs a shop. Banking is an additional line, competing for the same counter space, the same attention and the same working capital as everything else they sell. It wins that competition only while the arithmetic holds.
What the agent is weighing
- Commission per transaction against the time it takes. A transaction that needs three attempts earns the same as one that works first time and costs three times as much.
- Float tied up against float available for stock. Every rupee waiting on settlement is a rupee not buying inventory that turns faster.
- Footfall gained against footfall lost. Banking brings people in, which is the real prize — but a queue of banking customers blocking the counter costs other sales.
- Risk carried personally. A disputed transaction is argued with a neighbour, not escalated to a call centre.
The usual reasons a counter goes quiet
- Settlement became unpredictableNot slow — unpredictable. An agent can plan around two days. They cannot plan around somewhere between one and five, so they stop accepting the transactions that create the exposure.
- Authentication started failingBiometric capture degrades with worn fingerprints, poor light and cheap scanners. A rising failure rate reads to the customer as the agent being incompetent.
- A dispute was not supportedOne unresolved transaction that the agent had to make good personally is usually enough. The network saved a small sum and lost a counter.
- Commission was cutOften done uniformly in a margin squeeze, which disproportionately removes the marginal agents — exactly the ones in the locations hardest to replace.
- Something easier appearedA competing network with faster settlement, or simply a product that uses the same counter minutes more profitably.
Designing the technology for a counter, not a desk
Software for agents is frequently designed as if the agent were a bank teller with a desk, a reliable connection and time. They are standing behind a counter with a customer waiting, a queue forming, and a connection that drops.
| Desk assumption | Counter reality | What the software should do |
|---|---|---|
| Stable connection | Drops several times a day | Queue locally and reconcile, never lose an entry |
| Large screen | A phone, often an old one | Single-column flows, large targets, no horizontal scroll |
| Time to read errors | A customer waiting | One line, plain language, and the next action |
| Typing accuracy | Hurried entry, often one-handed | Validate as they type; confirm destructive steps |
| English | The agent's own language | Full localisation, including the error strings |
| Training | Learned once, by watching | Flows that are the same every time, with no hidden modes |
The last row matters more than it looks. An agent learns one path and repeats it hundreds of times. An interface that changes its flow conditionally — an extra confirmation only for some amounts, a different screen order after an update — breaks a muscle-memory routine and produces errors in front of customers.
Frequently asked questions
What transactions can a banking agent handle?
Typically cash deposits and withdrawals, balance enquiries, transfers and bill payments — the routine operations that would otherwise require a branch visit.
Why does settlement speed matter so much to agents?
Agents fund transactions from their own working capital. Slow settlement ties up that float and directly limits how many transactions they can handle, which caps both their income and the service's availability.
What actually kills an agent network?
Unpredictable settlement, more than slow settlement. Agents rarely resign — they just stop offering the service when asked, because they cannot tell whether their money will return in time for their own obligations.
Why is a failed transaction worse for an agent than for an app?
Because it happens in front of a customer who knows them and will tell others. The cost is local reputation rather than a retry, which makes reliability a commercial requirement rather than a technical one.
Is assisted banking only for rural areas?
No. It also serves urban customers who prefer cash, lack a smartphone, or find an assisted counter easier than an app — the need is about access and confidence rather than geography alone.
