Raast as a primitive: payment rails and where AI helps merchant ops

Three ways to see it

1. 1

   Raast is the State Bank's instant payment system, launched in three phases starting 2021: bulk payments first, then person-to-person (P2P), then person-to-merchant (P2M) in 2024. It is operated by 1Link as a settlement engine connected to every bank in Pakistan. From the user's view, Raast moves money from any account to any account in under five seconds, twenty-four hours a day, free for retail. From the engineering view, it is an ISO 20022 messaging stack on top of a real-time gross settlement system at the central bank.

2. 2

   Raast is a primitive in the same sense that TCP is a primitive for the internet. It is not a product. It is a free, neutral, low-level capability on top of which products are built. The merchant pays no interchange. The bank cannot lock the merchant in. The customer types or scans an alias (a phone number, an account number, or a QR-encoded ID) and the rail does the rest. This is why every Pakistani fintech of the last three years has Raast at the bottom of its stack, the way every African mobile-money story has M-Pesa.

3. 3

   Six places where AI now sits productively on top of Raast for merchants. One: real-time fraud scoring on incoming payments, flagging mule patterns before reconciliation. Two: automatic classification of incoming Raast messages into income, refunds, and personal transfers, so the SME owner gets a clean P&L without manual coding. Three: Urdu voice notification of received payments for shopkeepers who cannot read English screens. Four: receipt generation that includes GST line items pulled from the order context. Five: cash-flow forecasting based on Raast trail, telling a kiryana store owner whether next week's stock order is safe. Six: dispute drafting in plain language when a customer claims a payment did not arrive.

The why-tree

Try this with Claude

AI-edge prompt: 'I am building a merchant-ops product on top of Raast P2M for unbanked-yesterday small shopkeepers in Pakistan. List ten realistic AI-assisted features ranked by both customer value and engineering effort, then pick the one I should build first and write the first-week prototype scope.' Then ask the model to predict three reasons your scope will slip; pre-mitigate them.