Collecting payments in Kenya and East Africa. 7 chapters and a final quiz.

Chapter 1. Choosing the rail before the provider.

An East African collection project starts with a question of ownership. Who owns the rail your customers already use varies by country: a listed company in Kenya, the central bank in Tanzania, two telecom operators in Uganda. The answer sets your cost, your time to access, and your negotiating room. Answer it before you choose a provider, never after.

One scoping mistake shows up in nearly every regional project. No East African license works as a passport. An e-money issuer licensed by the Central Bank of Kenya has no right to operate in Dar es Salaam or Kampala. You apply country by country, under a separate law, to a separate authority. The only shared infrastructure is two large-value rails, EAPS and REPSS. So the country remains the unit of decision.

• Store amounts in local currency. KES, TZS, and UGX are the only values that hold up with a local bank or regulator. A database that keeps only the euro equivalent leaves you unable to defend any claim.
• The phone number is the account identifier. The MSISDN is the customer key, the authentication channel, and the attack vector. Treating it as ordinary contact data is an architecture mistake.
• Test the USSD flow. The market still runs on basic feature phones with no mobile data. A checkout tested only on smartphones is only half tested.
• Don't sign a long exclusivity deal on a domestic rail. Kenya has not yet settled the choice between PesaLink and a public rail. Build an abstraction layer between “collect” and “which rail to collect on.”

Chapter 2. Paybill, till, Pochi: choosing the reconciliation key.

M-Pesa offers three merchant products. A paybill is a biller number paired with an account reference the customer types in. A Buy Goods till is a checkout number with no reference. Pochi la Biashara attaches business collections to the trader's personal number and keeps those funds separate from the personal balance. Don't choose between them on convenience. Choose on the data the rail will send back to you.

The shortcode is now a tax object, and that changes the design. The Kenya Revenue Authority has tightened the link between merchant collections and e-invoicing. Starting with fiscal year 2026, reported income and expenses must be backed by an invoice issued and transmitted through eTIMS/TIMS (Kenya Revenue Authority, public notice of June 8, 2026). A mobile collection with no matching invoice becomes a tax risk. So from your very first data model, include a field that links the M-Pesa receipt number to the corresponding invoice number.

Chapter 3. Wiring up Daraja: from OAuth token to merchant statement.

Daraja is Safaricom's developer portal, and its pattern has been stable for years. You create an app to get a key/secret pair, exchange it for a token, then call the API that fits your use case. There are two environments, sandbox and production, with separate credentials. Four APIs cover a complete collection, including the outbound leg.

1. TOKEN      GET  /oauth/v1/generate?grant_type=client_credentials
                Basic base64(consumer_key:consumer_secret)
                -> access_token, expires_in 3599   [CACHE IT]

2. REQUEST    POST /mpesa/stkpush/v1/processrequest
                BusinessShortCode  paybill or till shortcode
                Password           base64(shortcode + passkey + timestamp)
                Timestamp          YYYYMMDDhhmmss
                Amount             INTEGER, in KES - no decimals
                PartyA / PhoneNumber  MSISDN 2547XXXXXXXX
                AccountReference   YOUR reference, not the customer's
                CallBackURL        https://.../mpesa/callback
                -> CheckoutRequestID              [PERSIST IMMEDIATELY]

3. WAIT       customer enters their PIN on their phone
                no guaranteed timing, no guaranteed response

4. CALLBACK   POST received on CallBackURL
                Body.stkCallback.ResultCode
                Body.stkCallback.CallbackMetadata.Item[]
                  Amount, MpesaReceiptNumber, TransactionDate, PhoneNumber

5. STATUS     POST /mpesa/stkpushquery/v1/query   [IF NO CALLBACK]
                CheckoutRequestID -> actual state

6. STATEMENT  merchant account reconciliation       [FINAL TRUTH]

One last architecture point concerns the channel. USSD menus remain the foundation of the East African market because they work on basic phones with no mobile data, which is what drove rural coverage. A USSD session is short and closes without warning, so the flow must let customers pick up where they left off. Track success rates separately for the app and for USSD, or the average will hide the problem.

Chapter 4. A callback is not proof: designing reconciliation.

A callback is a signal. It can go missing, arrive twice, or show up after the customer has reloaded the page, because it travels over a network you don't control to a URL you expose publicly. There are three sources of truth, and they don't carry equal weight. The callback informs, the status query decides, and the merchant account statement is authoritative. Sound reconciliation uses all three, in that order.

RECONCILIATION KEY
  good      MpesaReceiptNumber          -> unique, holds up as evidence, on the statement
  useful    CheckoutRequestID           -> links the request to the payment
  weak      AccountReference            -> typed by a human on a paybill
  never     amount + timestamp + name   -> can't tell payments apart

IDEMPOTENCY
  uniqueness key = MpesaReceiptNumber
  callback received twice  -> a single accounting entry
  request resent           -> one active CheckoutRequestID per order

CASES TO HANDLE BEFORE GO-LIVE, NOT AFTER THE FIRST INCIDENT
  callback never received     -> status query at T+90s, then the statement
  late callback               -> order already fulfilled, don't charge twice
  customer pays twice         -> detect, then refund through B2C
  payment with no order       -> exception queue, never a silent reject
  amount differs from amount due -> underpayment or overpayment, two separate processes

One last category of failure escapes technical monitoring entirely, because customers sometimes have to deposit cash before they can pay you. They go to an agent, who may be out of notes or e-float. The payment fails for reasons of place and time, not software. These failures cluster at month-end, on market days, and in rural areas in the evening. No automatic retry will fix them. If your success rates swing in cycles you can't explain, look at cash.

Chapter 5. Moving money out of the wallet: PesaLink, KEPSS, agents.

When you collect into a wallet, your cash stays in a wallet. Moving it to a bank account is a separate operation, with its own rail, limit, and timing. Kenya offers three routes, and the amount decides. Below 1 million shillings, the interbank instant payment rail is enough; above that, settlement moves to central bank money. Cash withdrawal remains a third route, and the most expensive one.

PesaLink is the banks' response to M-Pesa, launched in 2017, 10 years after it. More than 80 institutions are connected, counting banks, SACCOs, and providers (IPSL/KBA, 2025–2026). Its volumes grew 78% in 2023 over 2022, according to the Kenya Bankers Association annual report. No 2025 figure could be verified, and that gap tells you something in itself: Kenya's interbank rail discloses far less than the operator it is taking on.

For regional cross-border payments, three rails coexist and serve different needs. EAPS links the RTGS systems of Kenya, Rwanda, Tanzania, and Uganda in local currencies, from 8:30 a.m. to 4:00 p.m. EAT, Monday through Friday. REPSS, the COMESA rail live since October 2012, settles in dollars and euros, and eight countries are actually connected (Central Bank of Kenya, 2026). PAPSS pursues the same goal across the whole continent. Kenyan access to EAPS and REPSS goes through KEPSS, and therefore through your bank.

Chapter 6. Calculating the all-in cost, and who sets it.

Kenyan mobile money pricing is not set freely. On December 17, 2020, the CBK published its Pricing Principles, the framework for setting those prices. Safaricom's and the banks' fee schedules follow it and require the regulator's approval. So the posted price has been negotiated with the central bank. That doesn't make it predictable. It changes, and it carries levies that don't appear on the fee schedule.

ALL-IN COST = collection fee
             + 15% excise duty ON THAT FEE
             + cost of sweeping funds to the bank
             + cost of the exception queue

SAMPLE STRUCTURE - excise duty only
  collection fee charged to the merchant           KES 100.00
  15% excise duty on the fee (Finance Act 2023)     KES 15.00
  ----------------------------------------------------------
  merchant's actual cost                           KES 115.00

  A model built on the pre-duty fee schedule is off by 15%
  from the first line. The gap widens on high-frequency,
  low-value flows.

WHAT THE FEE SCHEDULE DOESN'T TELL YOU
  - the CONSUMER price is not the MERCHANT price
  - cash withdrawals are paid by the CUSTOMER, and they weigh
    on your conversion without showing up in your costs
  - the exception queue is a staffing cost, not a rail cost

The 2020–2023 episode offers a negotiating benchmark. On March 16, 2020, the CBK waived fees between mobile wallets and bank accounts, then reinstated them on January 1, 2023, at levels well below pre-crisis pricing. The announced cuts reach 61% on bank-to-wallet transfers and 50% on collection paybills (CBK, December 6, 2022). Pre-2020 pricing is no longer a defensible benchmark.

Chapter 7. Central bank licenses, foreign exchange, and repatriation.

There are three ways in, and they differ in both time and cost. Signing up with a local acquiring provider as a sub-merchant gets you live in weeks, with no license in your own name. Getting your own license takes months. Holding customer funds puts you in a different legal category. The due diligence question is the same in all three cases: who has the right to hold your balance, and in what legal capacity.

• Customer funds go into a trust, never onto the balance sheet. Regulation 25(3) requires you to set up a trust, keep the fund balance at least equal to what is owed to customers, never transfer those funds to the operating account, and never commingle them with third-party funds.
• Diversification has hard numbers. Under the Fourth Schedule, below KES 100 million, the funds are placed with one highly rated bank. Above that, each bank's share is capped at 25%, across at least two highly rated banks.
• The trust's investment income doesn't go to you. Under Regulation 25(5), it must be used in line with trust law, in consultation with the central bank, or donated to a public-interest charity. A business plan built on float income therefore doesn't work in Kenya.
• E-money limits are set by regulation, not by commercial terms. Regulation 43 sets KES 70,000 per transaction and KES 1,000,000 in monthly top-ups, and the central bank can approve higher limits for certain categories of issuers. The limits published by the major operators rely on that exemption.
• The light-touch regime caps activity. A small e-money issuer must be a company registered in Kenya, cannot exceed KES 20,000 per transaction, and can never hold more than KES 100 million in outstanding e-money liabilities.

Uganda structures its licenses differently. The National Payment Systems Act, 2020 gives the Bank of Uganda the power to grant, amend, and revoke them. There are three license types: payment system operator, payment service provider, and payment instrument issuer. One structural constraint comes before any setup. A provider that was not incorporated to issue e-money must create a separate legal subsidiary for that business, unless it is a financial institution or a deposit-taking microfinance institution.

One rule of method remains. Kenya is the African country with the most published literature, and the most outdated figures. Registered mobile money accounts exceeded 94.2 million in June 2026, and active agents 572,000 (Central Bank of Kenya, June 2026). These numbers shift from one quarter to the next, so date every figure you reuse and name its source. In this market, three-year-old data describes a different country.