Your customers make thirty commercial decisions a month. You monetise one. Flood is the AI-powered commerce layer that turns the app they already open into the place those decisions happen — and turns the merchants on the other side into a second, higher-margin business you own.
The premise, in two sentences. You already pay for the audience, and you monetise one thing they do with it. Everything else they do — deciding where to shop, what to buy, where to walk, who to pay — happens somewhere else, on somebody else's surface, and is worth more than the thing you charge them for.
Live in South Africa, India and the Maldives. Integrations and agreements progressing across Ghana, Mauritius, Panama, Puerto Rico and Türkiye. Definitions behind every figure available on request — we will not put a number in front of you we cannot defend in diligence.
Roughly 95% of retail in emerging markets still happens in a physical store, and classical e-commerce does not work there — fulfilment costs routinely exceed basket sizes. So the money is not in moving the goods. It is in moving the shopper, and in owning the surface where they decide.
Households shop smaller and more often to control cash flow. Four to five decision moments a month instead of one — which is what makes a discovery layer viable at all.
More shoppers now decide what they need before leaving home. The point of influence relocated to the phone. In most markets nobody owns that surface for the mass consumer.
The cost of travelling to a shopping centre is part of the real cost of the trip. Geo-targeting stops being a convenience feature and becomes a savings feature.
Budget-constrained shoppers notice inconsistent pricing and promotions that fail to deliver. One mismatched price at the till costs more than ten good offers build.
Low-income residents in South Africa's metros were reported to be spending roughly 29% of income on transport, and taxi fares rose again in 2026. Almost no digital surface in emerging markets tells a shopper what that actually means for their basket. Move the sliders.
Illustrative worked example, not a study. It persuades because it is simple — which is exactly why it should not be dressed up as research.
Today you own a customer relationship and monetise a single product against it. The retailer owns the purchase, the card scheme owns the payment, and nobody owns the decision. After: you know what your customer wanted, where they went, what they spent, which merchant won, and why. The offers are the visible part. The funnel is the business.
Stage 4 is the one everyone skips — and the one that decides the outcome. If you cannot prove the walk-in, you have built a banner ad on your own property: nobody funds a second campaign, no brand shifts trade budget to you, and you cannot underwrite a merchant on anything better than a bank statement. If you can prove it, you own the first offline attribution layer in your market — and every revenue line below becomes available. Build the measurement before the merchandising.
A marketplace with one strong side is a directory. Flood runs both sides with the same intelligence layer — which is why the loop closes instead of leaking. Each step below funds the next, and step six returns you to step one with a better merchant than you started with.
Near me, this week, in stock, affordable to me. AI ranks by what this person actually buys and what they can reach on foot — not by who paid for placement.
Recharge confirmation, salary or grant credit, balance check. Placement beats product — a rewards tab requires a habit your app does not have.
Cashback lands where you want the next purchase to start. Points that expire on somebody else's surface build somebody else's habit.
The merchant sees real walk-ins from your audience. Then the AI copilot tells them what to stock, what to price and who to message — so the footfall becomes revenue instead of a wasted visit.
Having seen the effect, the merchant adopts your rails to capture it. You now see real operating behaviour: turnover, inventory movement, repeat rate.
You underwrite on what they do, repaid as a share of takings. A better-funded, better-stocked, better-advised merchant converts more of your consumer traffic. Return to step one.
Why this is structurally hard to copy. A lender who can send its borrowers customers — and then tell them how to serve those customers — is not competing on rate. It is competing on outcome, which changes credit quality and pricing power at the same time. Almost nobody can do this, because it requires owning a consumer audience, a merchant book and an intelligence layer simultaneously. You have the first. We bring the second and third.
This is the most misunderstood point in the category. Every competitor building for informal retail eventually ships a merchant dashboard, and almost none of them get used. Flood's merchants are time-poor, data-poor and cash-constrained. They do not need information. They need decisions — pushed to them on WhatsApp, in their own language, before they have thought to ask.
"What should I stock this week?" · "Why did sales drop yesterday?" · "Can I afford to restock tomorrow?" A CFO, an operations manager and a marketing assistant in one — for a merchant who could never afford any of the three. Lives on WhatsApp, USSD and in-app.
Our advantage is not data volume — national retailers have more. It is granularity. Block-level, taxi-rank-level prediction of what sells, where, when and at what price. Auto-suggested orders, dynamic pricing, pre-positioning ahead of pension and pay days.
Instead of asking "do you have payslips?", the system states "this shop is stable and creditworthy" — on observed sales patterns, repayment behaviour, foot traffic we generate ourselves, seasonality and repeat rate. This is where the platform becomes a lender's best origination channel.
Segmentation, promotion timing, offer design, copy in the merchant's own voice, and channel selection — done for them. For you, it is what keeps offer inventory fresh without campaign-management headcount, which is the operational reason most offer platforms go stale in month four.
Illustrative of the interaction model. Language, channel and prompts adapt by market.
Photograph the shelf and stock updates — no typing, which is what makes inventory data consistent enough for forecasting to work. The same vision layer catches fake sellers and promotion-gaming at onboarding. Voice and multilingual support expands the addressable merchant base to every merchant regardless of literacy — including the ones every competitor has written off.
Why this matters commercially, not just operationally. A merchant who uses the intelligence layer daily generates the transaction, inventory and repayment data that makes them underwritable. A merchant who ignores a dashboard generates nothing. The AI is not a feature bolted onto the marketplace — it is the mechanism that converts a listing into a data relationship, and the data relationship is what becomes a credit book, a retail media product, and ultimately the moat.
Copilot for daily operations, baseline forecasting, the promotional engine, credit scoring v1. Produces immediate visible merchant value — and the training data everything downstream depends on.
Full lending infrastructure, supplier and marketplace optimisation, computer-vision inventory, risk management. This is the phase where the platform becomes a fintech rather than a marketplace.
Embedded finance, dynamic marketplace intelligence, AI logistics, cross-border. The moat. Attempting Phase 2 without Phase 1 is the most common failure in this category — lending models trained on nothing.
Kaspi.kz started with payments and a wallet, then in 2014 opened a marketplace to any shopkeeper with a phone — no website, no integration project. The bank stopped being a bank and became the mall. Benchmark the ratios, not the absolute scale.
Figures from Kaspi.kz 3Q 2024 investor presentation (9M'24). Now roughly two years old and pre-Hepsiburada; post-2025 figures are not like-for-like. Marketplace and Payments together produced 68% of net income — up from 63% a year earlier, and from zero for Marketplace in 2013.
| Player | What they did | The number that matters |
|---|---|---|
| Nubank Brazil, Mexico, Colombia | Built Nu Shopping — an in-app marketplace with cashback and 200+ partner stores — as part of deliberately rebalancing revenue toward fees and away from interest income. | 255 million Nu Shopping visits in 2023 alone, against a base of 110m+ customers. |
| Capitec South Africa | Live Better rewards across 30+ retail and service partners, plus Capitec Connect (MVNO), on top of a mass-market banking relationship. | 26% of group headline earnings from fintech — VAS and Connect — in the year to February 2026. |
| GCash / GLife Philippines | Layered third-party merchant discovery and commerce onto an existing payments habit, rather than launching a separate destination app. | 70M+ user base reached — discovery anchored to a payments habit, not a standalone app. |
| Pepkor South Africa — the other direction | A retailer buying the fintech layer. Merging Flash with Shop2Shop into a R21.3bn business, with a listing planned. | ~176,000 traders, R200bn+ processed a year. Announced, not yet closed. Retailers are not waiting. |
Read the Pepkor line carefully. The competitive threat to a telco or a bank in this category is not another telco or bank. It is the retailer — who already owns the store, the shopper relationship and the supplier budget, and is now buying the payments and merchant rails to complete the set. Whoever assembles audience + merchants + payments + data first sets the terms for everyone else. In most markets that race has already started, and it is running quietly.
A standalone super app that had to win a daily habit from scratch against WhatsApp. It peaked above 35 million monthly actives and was switched off after seven years, because it was a platform in search of a P&L. MTN's fintech arm meanwhile grew 24.9%.
→ Never build a destination. Embed into the app they already open, and give the layer a named P&L owner.
Six years, 76,000 active users. The Kenyan playbook imported into a market with bank account penetration, card rails and a different informal economy.
→ The friction you remove has to be one people actually feel in that market. Find the local friction first.
A tab requires a habit the app does not have. Users do not navigate to a discount section; they respond to an offer at a moment when money is already on their mind.
→ Anchor to an existing high-attention moment. Placement beats product, every time.
Being honest about this is the difference between a deployment that works and one that gets quietly shelved eighteen months in. Pick your business.
Rates are set deal by deal and market by market, so none are stated here — but this is the structure to model, and the order in which the lines actually switch on. Anyone who tells you all six start on day one has not run this.
We do not ask for a national launch, a rebuild or a budget. We ask for a placement and a defined geography. Every phase has a gate — and if a gate is not met, the correct action is to diagnose, not to scale.
One geographic cluster. 40–60 merchants in weekly-cadence categories only — grocery, pharmacy, airtime-adjacent. Randomised exposed and holdout cohorts. Offers at the chosen high-attention moment, nowhere else. Full attribution from day one. Copilot live.
300–500 merchants across two or three clusters. First supplier-funded campaigns. Copilot adoption tracked as a first-class metric. Build the first attributable case study.
Widen categories. Launch merchant services tiers. Turn on your partner-specific revenue line. Begin retail media conversations with brands. Credit scoring v1 running in shadow mode against real behaviour.
Financial services attach to the merchant base. Consumer BNPL where licensed. Computer-vision inventory rolled out. Demand-signal data product designed and governance-approved. Design the next market.
The two numbers to put on the wall. Consumer side: verified in-store transactions per active user per month — the only metric that proves the layer changed real-world behaviour. Merchant side: the share of merchants who act on an AI recommendation in a given week — the only metric that proves the intelligence layer is an operating system rather than a feature, and the leading indicator for every credit and data revenue line that follows.
The software is not the moat and we would not claim otherwise. In-house builds stall because within two quarters the organisation discovers it has accidentally become a field sales operation for small businesses — a completely different cost structure, hiring profile and management rhythm to a telco or a bank. We carry that. You keep the customer.
The app is yours, the brand is yours, the data governance is yours. Flood does not build a consumer brand that competes with our partners. We are infrastructure.
Where the basket cannot carry a delivery cost, the store is the fulfilment centre. Building warehouses is how the classical model dies in these markets.
The layer goes into the app you already have, at a placement you already control. SDK in, front end untouched.
Critical. Hard launch gate at 98% integrity. Merchants who cannot guarantee price parity are excluded, not persuaded. One-tap "price was wrong" reporting treated as a priority-one incident.
Critical. Contractual revenue share with a named P&L owner and a quarterly reporting rhythm from month one. This is the specific failure mode that killed Ayoba at 35 million monthly actives.
Critical. Delivered on WhatsApp, USSD and voice in the merchant's own language, pushing decisions rather than waiting to be queried. Adoption is tracked as a primary metric, not a product vanity number.
On exclusivity. We will agree category and term-limited exclusivity in a defined channel — a partner investing placement and brand deserves protection while the model proves out. We will not agree blanket national exclusivity across all sectors. That is not a partnership; it is a free option on a whole market, and it is bad for you too, because it removes the pressure that makes the layer improve.
Independent validation, not self-praise.
Thirty minutes is enough to scope a cluster and agree how we will measure it. Bring whoever carries the revenue number — we will bring the numbers.
How long does money stay with your customer before it leaves your rails? For a bank: how many hours between the credit landing and the cash withdrawal. For a telco: how much of your base has usable data at any given time. For a retailer: how much of the shopper's wallet you already see. The answer determines the design, the sequencing, and in some cases whether this is the right first market at all — and you can answer it from your own data in an afternoon.
Sources for third-party figures on this page: Kaspi.kz 3Q 2024 investor presentation; Nubank company release (2023); Capitec annual results to February 2026; GCash/GLife enterprise materials; Moneyweb on the Pepkor–Flash–Shop2Shop transaction (announced July 2026, not yet closed); TechCabal on the Ayoba shutdown (March 2026). Transport and basket figures are an illustrative worked example based on reported South African commuter cost data, not a study. Flood platform figures are first-party; definitions available on request.