A three-person Kenyan startup put an AI analyst inside M-Pesa, and 6,000 small businesses signed up
Fastagger’s Auni reads mobile money statements, runs partly offline on cheap phones and tells barbers when to staff up. Agents that act on its advice are next.
DevIsDead Newsroom
Most AI products assume a laptop, a fast connection and a credit card. Fastagger, a Nairobi startup, built for a barbershop owner with a basic phone and a paper sales book.
Its assistant, Auni, launched in 2024 and took off after M-Pesa, the country's biggest mobile money platform, built it into its app for merchants, Lucila Pinto reports for IEEE Spectrum. About 6,000 businesses have signed up. It analyzes their transaction data and suggests marketing moves for barbershops, restaurants and small manufacturers.
Built around the constraints
Co-founder Mutembei Kariuki says the idea came at a 2019 AI conference in Tokyo where, as he tells it, nobody mentioned Africa at all. The obstacles were obvious: slow and uneven internet, little compute, scarce structured data, and users on old, low-memory phones.
So Fastagger started with data people already have. More than 80% of Kenyan adults use mobile money apps. The team adapted an open-source OCR model to read the messy PDF statements those apps produce, with odd layouts, stamps and errors. Key steps, like turning extracted text into tables, run on the phone. Standard OCR pulls the text and a small neural network works out the document's structure. Reading statements, analyzing data and drawing charts can happen offline. Deeper analysis and questions about the data need a connection.
What it does for a barber
Peter Chege, who runs a salon and barbershop in Nairobi, used to keep sales on paper. Auni showed him his busiest days and hours, and that mornings were slow, so he now runs errands then and keeps fewer barbers in until about 1 p.m. It also flags regulars who stop coming so he can message them. A Nairobi fish restaurant that used to go through SMS receipts one by one to find loyal customers now gets the list automatically.
Fastagger also built a tool for Kenya's security companies. An agent talks to guards over SMS, tracks their routes and flags anomalies for operations teams.
The reality check
It's tiny. Despite the M-Pesa deal and a place in accelerator programs run by Nvidia, Microsoft and Google, the company is still just its three founders. Columbia economist Daniel Björkegren also warns of a split, with small models for poorer markets and powerful ones for rich markets.
But the lesson for developers runs both ways. On-device, offline-first AI that does one job on cheap hardware is reaching real users while plenty of cloud copilots look for a use case. Next, Fastagger is adding agents that carry out the marketing actions instead of just recommending them.
Scoreboard update: AI +1. An offline AI analyst on basic phones does the bookkeeping for 6,000 small businesses.
Stay undead.
Comments 0