Leaving this course?

Lesson 1 of 8

How the system is actually put together

Mobile money is often described as banking on a phone. It is not, and the differences are where the financial crime risk lives.

What a customer holds is electronic money, not a deposit. A provider issues a balance against value received, and the aggregate of those balances is backed by funds held in trust at a bank. The customer has a claim on the issuer; the money itself sits in a pooled account the customer never sees.

Which means the bank holding the float sees one account, not millions. From the bank's position the whole scheme is a single large customer. Every individual transaction and every individual customer lives inside the provider's own records, and the bank's monitoring reads none of it.

Agents are the system's hands. An agent takes cash and credits a wallet, or debits a wallet and pays out cash. They are small businesses - a shop, a kiosk, a pharmacy - remunerated by commission on volume. The provider holds the licence; the agent performs the transaction and sees the customer.

Cash enters and leaves at the edges. Between those two moments everything is electronic and well recorded. At those two moments the record is whatever the agent entered. This is the single most important structural fact in the subject.

Registration ties the wallet to a subscription. The identity behind a wallet is usually the identity behind the mobile number, captured when the subscription was registered, sometimes years earlier and by a different party under a different rule.

Limits and tiers substitute for documentation. Because the purpose is reach, many regimes allow simplified checks for low-value accounts, with transaction and balance limits in place of full verification. That is a deliberate, documented trade: less identity for less exposure.

Interoperability has joined the systems up. Wallets now send to other providers' wallets, to bank accounts, to merchants and across borders. A single value chain may cross a wallet, a bank and a remittance operator, each seeing one segment.

And the volumes are unlike anything in branch banking. Tens of millions of small transactions a day, most of them entirely ordinary. Any control that requires a person to look at a transaction will not be applied to transactions, only to patterns.

So the shape of the problem is this. Excellent electronic records in the middle, two cash edges where the record is made by a commission-paid third party, identity inherited from a different process, and volumes that make rules the only feasible control. Everything in the next seven lessons follows from that sentence.

Lesson 2 of 8

Where the risk sits

The risks in this sector are not the ones usually assumed. The system is highly traceable; what is weak is who, not what.

The agent is the identity checkpoint, and the weakest link by design. Verification of the person standing at the counter happens there or nowhere. The agent is not a compliance professional, is paid for volume, knows the customer socially in many cases, and has every commercial reason to complete the transaction.

Over-the-counter transactions break the model entirely. Where an agent performs a transfer from their own wallet on behalf of a walk-in customer, the sender of record is the agent and the actual sender is unidentified. The transaction is fully traceable to the wrong person.

Third-party deposits do the same thing more quietly. Cash paid in by one person to another person's wallet. Common, usually innocent, and a route by which money enters a wallet with no record of its source.

Simplified due diligence is a known, accepted exposure. Low-value accounts verified lightly are a policy choice made for inclusion, and the mitigation is the limits. The failure mode is not the policy but the many-accounts strategy: several lightly verified wallets used together exceed any single limit without breaching one.

Identity inherited from subscription registration carries its errors forward. Where a subscription was registered on a document that was borrowed, fake or belonged to somebody else, the wallet inherits that identity permanently, and every subsequent control treats it as verified.

Cash out is the step that completes the laundering. Electronic value becomes notes at a shop counter, in amounts and locations the sender chooses. The traceability of the electronic leg is irrelevant once the money has left in cash at a place nobody has looked at.

Fraud and laundering overlap more here than in most sectors. Proceeds of social engineering, fake investment schemes and extortion move through the same wallets, in the same patterns, often through recruited mule accounts. A team that treats the two as separate workflows investigates the same accounts twice.

And the float account hides the activity from the one regulated balance sheet in the chain. The bank holding the trust funds sees aggregate movements. Unless the provider's own monitoring works, nothing in the formal banking system is looking at the customers at all.

The conclusion is uncomfortable but useful. The electronic record is excellent and the identity behind it is thin. Controls should therefore work on behaviour and on the agent network, not on the assumption that the name on the wallet means anything.

Lesson 3 of 8

The methods

Every method in this lesson exploits the structure in lessons one and two. None requires technical skill.

Many wallets, small amounts. Value is split across a number of lightly verified accounts, each kept within its limits, and consolidated at the far end. This is structuring with an extra step, and it defeats per-account rules while being obvious across accounts.

Mule wallets, recruited cheaply. Accounts opened by real people, verified correctly, and then controlled by somebody else. The identity is genuine and irrelevant. Recruitment is often open, framed as a job or a favour, and concentrated in groups with a shared characteristic such as age or location.

Agent-assisted anonymity. The agent performs the transaction on their own account, or registers a wallet for a customer who does not present documents, or splits a large cash deposit into several below-limit entries. Each of these is an agent control failure before it is a laundering typology.

Account takeover through the subscription. Control of the mobile number gives control of the wallet. Fraudulent replacement of a subscriber's module, or social engineering of a provider's customer service, transfers the wallet to the attacker with every identity record intact.

In and straight out. Receipt followed within minutes by onward transfer or cash withdrawal, repeatedly. A wallet used for its intended purpose holds a balance and spends it on things; a conduit does not.

One to many and many to one. A single wallet funding dozens of others, or dozens funding one, within a short window. The first is distribution of proceeds; the second is collection. Both are visible in a single query over the transaction table.

Merchant and till abuse. A business account used to receive payments that are not sales, taking advantage of higher limits granted for commercial purposes. The signal is receipts that do not match the pattern of a trading business.

Cross-border entry through remittance. International transfers delivered directly into wallets, which converts a regulated cross-border payment into local cash at an agent counter within minutes. Where originator information is incomplete, the wallet is the first point at which anybody could have asked.

And loan and float fraud at the agent layer. Agents hold their own float and sometimes receive credit against it. Manipulating float balances, cycling value between agent accounts, or using the float as a transit pool are internal versions of the same arithmetic, and they are found in agent data rather than customer data.

Lesson 4 of 8

Running an agent network

Agent oversight is the largest single control in this sector and the one most often reduced to a contract and a training slide.

Select before you appoint. Identify the agent and its owners, understand the existing business, confirm the premises exist, and check the owners against sanctions and adverse information. An agent is a counterparty, and onboarding one should look like onboarding a business customer.

Train on the specific failures, not on the law. Agents do not need a lecture on international standards. They need to know that they must see the customer, record the identification actually presented, refuse to split a deposit, never transact from their own wallet for a customer, and escalate the particular behaviours they will encounter.

Set limits and make them real. Per-transaction and daily limits for the agent and for customer transactions at that agent, enforced in the system rather than in the agreement. A limit that is a term of a contract is a limit that will be exceeded.

Require records and then inspect them. Transaction logs, identification recorded, and a register of refused or unusual transactions. Then visit, unannounced, and compare what the register says with what the system shows. Inspection that is scheduled and announced measures tidiness.

Mystery shopping is the only honest test. Someone presenting as an ordinary customer, attempting the things an agent should refuse. Nothing else tells you whether the training reached the counter.

Watch commission against compliance. Agents are paid for volume. Every control costs them a transaction. Where the commission structure rewards throughput only, the incentive runs against the control, and the fix is partly commercial rather than entirely regulatory.

Monitor the agents themselves. Transaction volume, average size, cash-in to cash-out ratio, the number of distinct customers, activity outside trading hours, and transactions involving the agent's own wallet. An agent whose profile differs sharply from comparable agents in the same area is doing something different.

One national example, where the liability is explicit. Under the National Payment System Regulations, 2014, a provider may appoint agents and remains liable to customers for acts the agent performs within the agency agreement. Providers must notify the Central Bank before using a new agent, maintain oversight of agents and keep records of their transaction volumes, and the Bank may inspect both the provider and its agents on site. The e-money regulations impose parallel agent requirements on issuers.

And terminate properly when you have to. Removal of system access on the same day, recovery of branding, notification to the regulator where required, and a record of why. An agent dismissed for a control failure who continues to display a provider's brand is still the provider's face.

Lesson 5 of 8

Identity, subscriptions and tiered checks

Identity in mobile money is borrowed. Understanding what it was borrowed from, and what that process was designed to achieve, explains most of the weaknesses in it.

The wallet usually inherits the subscription. Registration of a mobile subscription is a telecommunications requirement, performed to a telecommunications standard, often by a retail dealer paid for activations. The financial identity then rests on that record.

So errors propagate in one direction only. A subscription registered on a borrowed or false document produces a wallet that every later system treats as verified. Nothing downstream re-tests it, and the customer has a complete and consistent history.

Re-registration exercises are the usual remedy and they are blunt. Requiring an entire customer base to present documents again catches some of it, excludes legitimate customers who cannot comply, and produces a new snapshot that begins ageing immediately.

Tiered due diligence is the right idea applied to the wrong unit. Low limits for lightly verified wallets is sound in principle. The weakness is that the limit applies to the account and the customer may hold several, so the control measures the wrong thing.

Which argues for identity resolution across accounts. The practical control is matching across wallets on the attributes available: identification number, name and date of birth, device, next-of-kin details, the funding agent and the counterparties. Clusters sharing attributes are one customer, whatever the account count says.

Upgrading a tier is a control moment worth using. A customer asking for higher limits is volunteering to be verified, which is the cheapest verification an institution ever gets. Treat it as onboarding rather than as a service request.

Business and merchant accounts need a separate standard. Higher limits are granted for a commercial purpose, and the evidence of that purpose - a registration, a premises, a trading history - is what justifies the limit. Granting merchant status on a form is a known route to a high-limit personal wallet.

Data protection is a live constraint, not an obstacle. Identity, transaction and location data about millions of people is held for a purpose, and using it for monitoring is lawful where that purpose and basis are documented. Your own data protection law sets the framework, and a monitoring programme that has never been assessed against it is exposed in a different direction; in Kenya, for example, that law is the Data Protection Act, 2019.

And the honest position belongs in the risk assessment. Verification for a large part of the customer base rests on a telecommunications registration record. Saying so plainly is what allows proportionate controls to be designed; claiming the base is verified to a documentary standard is simply inaccurate.

Lesson 6 of 8

Monitoring at volume

With tens of millions of transactions a day, monitoring is a modelling exercise. The value lies in a small number of well-chosen comparisons rather than a long list of rules.

Velocity against the account's own history. Transactions per day, value per day, and both compared with this wallet's normal rather than a population average. The population average is meaningless in a base this diverse.

In-and-out within minutes, repeatedly. The strongest single indicator. Build it as a ratio of time held to value received, and the conduits separate from the users.

Fan-out and fan-in counts. Distinct counterparties per wallet per day in each direction. Ordinary customers transact with a handful of people; distribution and collection patterns do not.

Cash-out concentration. Several wallets withdrawing through the same agent, or one wallet withdrawing through many agents in a short period. The first points at a scheme with a trusted exit; the second at somebody avoiding limits.

Device and subscription changes. A wallet accessed from a new device, a replaced subscriber module followed immediately by large transfers, or a sudden change in location pattern. These are fraud signals and laundering signals simultaneously.

Dormancy then volume. An account inactive for months that begins transacting at scale, particularly where the first transactions are receipts. Mule recruitment produces exactly this shape.

Round numbers and repetition. Identical amounts repeated, or amounts sitting just below a reporting or verification threshold. Threshold-adjacent behaviour is measurable: count transactions in the band immediately below each limit and watch the proportion.

Network analysis rather than account analysis. Group wallets by shared device, shared agent, shared counterparties and shared registration attributes. Most schemes are visible as a cluster and invisible as individual accounts, which is why single-account rules under-perform.

Agent-level monitoring is a separate programme. Every signal above also applies to agents, measured against comparable agents by location and size, and an agent appearing in many customer alerts is itself the finding.

And tune against outcomes. Record which alerts led to reports, closures or nothing, and feed that back. A monitoring programme that has never compared its alerts with its reports does not know which of its rules work, and in a base of this size the cost of a useless rule is measured in thousands of reviews.

Lesson 7 of 8

Rails, interoperability and crossing a border

A wallet is no longer the end of the journey. Value moves between wallets, banks, merchants and countries, and each handover is a point where information is lost.

Wallet to bank and back is now ordinary. A customer moves value into a bank account and out again. Each institution sees its own leg: the bank sees a transfer from a provider, the provider sees a transfer to a bank. Neither sees the whole, and the combined path can convert cash at an agent into a bank balance and back to cash elsewhere.

Interoperability between providers has the same effect. A transfer from one scheme to another arrives with whatever the sending scheme chose to pass, and the receiving scheme applies its own limits to a customer it has never verified.

Pass-through and aggregation hide the real customer. Where a payment service provider or aggregator holds an account and moves value for many underlying parties, the wallet or bank at the far end sees the aggregator. Look-through to the underlying party is a contractual and technical arrangement that has to be designed, not assumed.

Cross-border remittance into wallets compresses the chain. An international transfer can become local cash at a kiosk within minutes of arrival. The originator information travelling with that payment is therefore the only identity the receiving side will ever have about the sender.

Which is what the payment transparency standard is for. The international standard requires originator and beneficiary information to accompany transfers, with the chain preserved through intermediaries. Incomplete or meaningless originator data is not a technicality; it is the removal of the only control available at the far end, and the receiving institution is expected to have a policy for what it does about it.

Who is the originator in an agent-assisted deposit is a real question. If cash is paid in by one person to another's wallet, and then sent onward, the originator of record may be the wallet holder while the source of funds is the depositor. Systems that record only the account holder lose the person who actually brought the money.

Merchant payments and bill payments are a different risk profile. They are mostly low risk and they are also a laundering route where a merchant is complicit or fictitious. Merchant onboarding, and monitoring of merchant receipts against the trading profile, belongs in the same programme.

And reconciliation between legs is a control, not just an accounting task. Where value entering a scheme cannot be matched with value leaving it, the difference is where the unexplained activity sits. Treating reconciliation breaks as a finance problem rather than a financial crime signal misses one of the few checks that covers the whole chain.

Lesson 8 of 8

One network, followed through

A mobile money provider with several million active wallets and tens of thousands of agents. Four analysts. The case begins with a fraud complaint, which is how most of them begin.

The complaint. A customer reported losing a sum to a caller who had persuaded her to confirm a code. The money had left her wallet in three transfers within four minutes.

The first hop. All three transfers went to wallets registered within the previous month, each lightly verified and within tier limits. Each forwarded the value within minutes to a fourth wallet, which withdrew cash at a single agent.

The second finding. Querying for other wallets that had withdrawn at that agent in the same period returned forty-one, of which nineteen had been registered in the same three-week window and had received their first credit from a wallet that had itself been credited by an unrelated customer. The shape was identical to the complaint in every case.

The cluster. Grouping the nineteen on shared attributes - two devices, a common next-of-kin number, sequential registration times at the same dealer - produced one group rather than nineteen customers. None of the individual accounts had breached a limit.

The agent. Its cash-out volume was four times the median for its area, its ratio of cash out to cash in was heavily one-sided, and a disproportionate share of its withdrawals were from accounts less than sixty days old. No single figure had triggered a rule because no rule compared an agent with its peers.

What the provider did. Suspended the nineteen wallets and the agent's access the same day. Reported to the financial intelligence unit with the cluster analysis. Preserved the device and registration data. Referred the dealer who had registered the subscriptions for investigation, since the registration pattern pointed there rather than at the agent alone.

What the money did. Most of it was gone. A proportion was still held in two wallets and was frozen on instruction. The complainant recovered part of her loss, which is a better outcome than usual and not a good one.

What changed. Agent peer comparison became a monthly report on cash-out ratio, account age mix and volume against area median. Clustering on shared registration attributes ran weekly rather than on request. New wallets receiving their first credit from a wallet funded by an unrelated third party within minutes became a rule of its own. Dealer-level registration quality became a metric with an owner.

The point. Every individual transaction was within limits, properly recorded and fully traceable. The scheme was invisible at the account level and plain at the cluster level, and what found it was a question about an agent that nobody had been asked to ask.

Knowledge check

Thirty questions

Answer all thirty questions, then check your answers. You need 27 out of 30 to pass and receive a certificate. If you score less, you will see which answers were right and wrong, and then go through the course again before you retake the check. Your answers, progress and times are kept only in this browser.

Standards and frameworks covered

The published frameworks this course teaches toward. Saolix is not affiliated with, endorsed by or accredited by any of these bodies; naming a standard is a citation, not a claim about its author.

Sources

The official documents this course relies on. Laws and guidance change, so check the current version.

  1. Guidance for a Risk-Based Approach for Money or Value Transfer Services · Financial Action Task Force (FATF)
  2. Guidance on Anti-Money Laundering and Terrorist Financing Measures and Financial Inclusion · Financial Action Task Force (FATF)
  3. The Role of Hawala and Other Similar Service Providers in Money Laundering and Terrorist Financing · Financial Action Task Force (FATF)
  4. Revision of Recommendation 16 on payment transparency · Financial Action Task Force (FATF)
  5. National Payment System Act, 2011 (Cap. 491A) · National Council for Law Reporting, Kenya
  6. National Payment System Regulations, 2014 (Legal Notice 109 of 2014), including agent appointment, oversight and liability · National Council for Law Reporting, Kenya
  7. National Payment System Regulations, 2014: Kenya Gazette Supplement No. 119 · Central Bank of Kenya
  8. E-Money Regulations, 2013 · Central Bank of Kenya
  9. National Payments System Vision and Strategy 2021-2025 · Central Bank of Kenya
  10. Proceeds of Crime and Anti-Money Laundering Act, 2009 (No. 9 of 2009) · National Council for Law Reporting, Kenya
  11. Data Protection Act, 2019 (No. 24 of 2019) · Office of the Data Protection Commissioner, Kenya