Today on The Merchant Desk: the agentic commerce frameworks we've been tracking get their first live production transaction in Europe, Visa expands its autonomous shopping infrastructure, and South Africa's informal economy becomes the contested frontier for fintech's next growth chapter.
Macy's deployed Ask Macy's, an AI shopping assistant built on Google Gemini Enterprise for Customer Experience, achieving 4.75x higher revenue per visit during beta testing. The project started as internal development, stalled for six months, then switched to Gemini — launching an initial beta four weeks later and rolling to full production site-wide within weeks. The 4.75x lift comes from agents scanning full product catalogs and asking clarifying questions before surfacing options, deepening basket engagement for ready-to-buy customers rather than reducing abandonment.
Why it matters
The 4.75x revenue-per-visit metric is striking — even outpacing the 3.5x conversion lift we saw from Kate Spade on Amazon's AWS Agentic Shopping Assistant earlier this week — but the more strategically significant data point is the timeline: six months of internal development versus four weeks to beta on a third-party enterprise LLM. That's the build-vs-buy calculus settling in real time at major retail scale. Retailers who prioritise internal AI development for competitive differentiation will need a clear answer to why their proprietary approach justifies the delay — particularly when the model can be swapped and the differentiation increasingly lives in product data quality, catalog schema completeness, and integration depth rather than the underlying LLM. For SA retailers evaluating AI commerce investments, Macy's experience validates the 'configure not build' approach using established enterprise AI platforms, and reinforces that the investment case for AI shopping is now grounded in measurable conversion data, not speculative future value.
Following up on the Mastercard Agent Pay framework we've been tracking, Worldline, ING, and Mastercard completed Europe's first live production agentic payment transaction in the Netherlands — an ING cardholder's AI agent browsed merchant inventory, curated product selections within a defined budget, requested consumer confirmation, and then completed the purchase across existing European acceptance, acquiring, authentication, and issuer infrastructure. Mastercard's framework provided network-level standards; the full issuing bank remained in the loop with explicit consumer consent required before funds moved.
Why it matters
This isn't a demo or sandbox test — it's a production-grade transaction across a live card network with real issuer oversight. That distinction matters enormously, because it proves the current payment stack can support agentic commerce without rebuilding the rails. The harder problems remain structurally unresolved. As we noted last week regarding the agent-initiated dispute gap, American Express moved first on agent purchase protection; Mastercard's Agent Pay gives network-level structure but leaves that exact dispute resolution in a grey zone. Merchants sitting inside these ecosystems need to watch whether their existing chargeback exposure calculations still apply when the counterparty is an AI session, not a human card-present transaction.
Building on the Trusted Agent Protocol integration we tracked earlier this week, Visa unveiled Intelligent Commerce Connect — a full platform enabling AI agents to autonomously browse, select, and complete purchases with configurable per-session spending limits set by cardholders. The system supports both Visa and third-party cards, integrates with Coinbase's x402 protocol via Nevermined (which processed $24M in transactions across 30 days in pilot), and maintains KYC, fraud controls, and network compliance throughout. Broader rollout is planned for late 2026.
Why it matters
Visa entering agentic commerce infrastructure is a structural market signal, not a feature announcement. The network is positioning itself as the trust and compliance backbone for machine-to-machine purchasing — directly competing with decentralised payment protocols like x402 and Stripe's Machine Payments Protocol, while leveraging its regulatory standing and existing fraud infrastructure as the differentiator. The Nevermined x402 integration is notable: Visa isn't rejecting crypto-native rails, it's absorbing them into a compliant wrapper. For merchants and payment operators, the practical implication is that agentic transaction flows will increasingly arrive through existing card network rails with new metadata layers rather than requiring new acquiring integrations — but fraud scoring, chargeback rules, and MDR structures haven't been updated to reflect the new risk topology of agent-initiated purchases. That gap is where merchant tech operators need to focus their compliance and risk architecture now.
Mastercard disclosed this week that 60% of its online transactions in Europe now use tokenised credentials, with a 2030 target for near-total elimination of manual card entry. The tokenisation suite spans network tokens (Secure Card on File, 45 countries), Click to Pay (32 markets, enrollment doubled), and payment passkeys with biometric verification. Separately, Mastercard announced intraday and weekend/holiday settlement capabilities with regulated stablecoin settlement options, integrated into existing global infrastructure and following its BitLicense acquisition from New York regulators.
Why it matters
Sixty percent European digital transaction tokenisation is no longer a pilot metric — it's table stakes confirmation that the checkout experience the networks are building toward is one where card numbers are invisible to merchants and consumers alike. The 2030 full-transition timeline means the infrastructure investment decisions made now in merchant tech stacks (payment passkeys, Click to Pay integration, tokenised credentials) have a hard deadline. The weekend/stablecoin settlement announcement is the more forward-looking development: Mastercard is building always-on settlement capability directly into its network, removing the 'wait for Monday morning' problem that creates liquidity headaches for high-velocity merchants and cross-border operators. For SA and African market operators, this signals the settlement infrastructure direction — even if the local implementation timeline runs 3–5 years behind European deployment.
MTN has repositioned MoMo in South Africa away from the P2P transfer model that succeeded in Ghana and Uganda — a model that simply didn't translate into SA's mature banking landscape — and toward building fintech infrastructure for the informal and SME economy. The current playbook: 8 million users, partnerships with Sphazamisa (township retail), Sanlam (insurance), and UNDP's DIME initiative, targeting digitisation of spaza shops and the R170 billion informal FMCG market. A separate MOU with Wakanda signed May 28 aims to equip approximately 150,000 spaza shops serving 11.1 million South Africans with mobile payment tools, digital business capabilities, and connectivity.
Why it matters
Two failed relaunches taught MTN something important: South Africa's informal economy isn't just a volume opportunity — it's a fundamentally different go-to-market problem. Spaza shops are not underserved branches of formal retail; they are trust networks with different supplier relationships, cash management habits, and digital literacy baselines. The third attempt is notable for its institutional backing (UNDP, Sanlam) and ecosystem approach rather than direct-to-consumer product push. NielsenIQ's Q1 2026 data showing traditional trade channels outperforming modern retail in SA reinforces that this segment isn't a transitional phase — it's structurally resilient and growing in share. For fintech operators targeting merchant acquisition in SA, the race to digitise informal merchants is now a multi-player contest: MTN/Wakanda on mobile payments and connectivity, Lesaka on cash-integrated POS, and — as we covered earlier this week — Yoco pushing into SME operations with its Dyner.ai acquisition. The infrastructure question is less 'who gets there first' and more 'whose stack earns the merchant's trust and actually handles cash-digital hybridity.'
South African fintech Yoco appointed Carsten Höltkemeyer — a Berlin-based embedded finance executive with a decade across European banking and card payments infrastructure — as CEO effective June 1, replacing co-founder Katlego Maphai who stepped down in September 2025. The hire followed a global search; founding team members return to functional roles while Maphai stays engaged on strategy.
Why it matters
The timing of this appointment is worth parsing carefully. As we've tracked over the past week, Yoco just acquired Dyner.ai (AI-native restaurant ops) and launched its Stub accounting integration. Höltkemeyer's background in European regulatory navigation and Banking-as-a-Service infrastructure is a deliberate bet: the next phase of Yoco's growth runs through the SARB's impending activity-based licensing framework we recently covered, potentially a banking charter application, and regional expansion — not product-market fit. Professional CEOs at this stage in a SA fintech's lifecycle typically signal one of two things: preparation for a capital event (IPO or strategic acquisition) or a serious push at institutional enterprise segments. Given Yoco's 200,000-merchant base and the new AI commerce positioning, both scenarios are plausible — and not mutually exclusive.
Airwallex launched a unified billing platform with invoicing, subscription management, and native usage-based billing, integrated into existing pricing at no additional cost. The $8 billion cross-border payments company cited AI and SaaS companies' shift to token-based and API-call consumption models as the primary demand driver — these businesses need real-time metering and dynamic pricing infrastructure that flat-rate billing tools cannot support. Transaction processing now represents only 30% of Airwallex's revenue.
Why it matters
The 30% revenue share from transaction processing is the buried lead here. When a cross-border payments company with $8B valuation derives 70% of revenue from non-transaction services, the strategic message to the market is unambiguous: processing fees alone cannot sustain a payments business at scale. Airwallex's move into billing, subscription management, and usage-based infrastructure follows exactly the playbook Shopify used to move from checkout to commerce OS — and that Stripe used to move from payments into Treasury, Radar, and now agentic commerce. For payments operators, the challenge is that usage-based billing is technically harder than subscription billing: it requires real-time metering infrastructure, accurate consumption tracking, and revenue recognition that can handle partial-period and overage scenarios. The AI/SaaS companies driving this demand aren't willing to bolt on three separate tools to solve it, which is why Airwallex bundling it into existing pricing is a genuine competitive move against Stripe's Billing and Maxio.
Clink launched public access to its payments and billing platform for AI builders, enabling AI products to accept payments from humans today and process authorised agent-to-merchant transactions in real fiat — within user-defined spending limits — using existing Stripe, Adyen, or Checkout.com infrastructure. The platform supports 135+ currencies, 100+ local payment methods, and native integrations with coding agents including Cursor and Claude Code. Early customers AutoCoder and PollyReach are live.
Why it matters
Clink's architectural decision — routing agent transactions through standard card network infrastructure rather than building parallel rails — is the pragmatic bet that agent payments will be governed, auditable, and MDR-bearing rather than crypto-native and fee-free. That's both a compliance strength and a cost implication merchants haven't fully priced yet: if AI agents generate meaningful transaction volume, the interchange and network fees on those flows will be material, and the question of who pays them (end consumer, agent operator, merchant) isn't settled. The coding agent integrations (Cursor, Claude Code) are the most concrete near-term use case: developer tools that autonomously purchase API credits, compute resources, or third-party services on behalf of a user within a session budget. That's a narrow but real and immediately monetisable workflow. For vertical SaaS operators building commerce-adjacent tools, Clink represents the infrastructure layer that removes payment complexity from agentic feature development.
Stitch's 2026 Consumer Payments Report (3,000+ respondents) maps a materially shifted South African online payment landscape: 76% of shoppers spend R2,000+ monthly online; BNPL trial rate reached 38.9% with 71% of credit-active users employing it across categories including groceries; Capitec Pay captured 24.6% online payment preference in just three years; 62.5% now shop online as much as or more than in-store; and 48.5% use low-cost international platforms like Temu and Shein. Peak transaction hour is 20:00 SAST; the 25th of the month (payday) drives 35% of monthly e-commerce volume.
Why it matters
Two data points from this report deserve immediate operational attention for SA payment and merchant operators. First, the 20:00 peak and 25th payday concentration aren't consumer curiosities — they're infrastructure design requirements. Checkout flows, fraud scoring models, and payment routing need to be tested under the load profiles these windows create, not average daily traffic. Second, Capitec Pay's 24.6% share is a structural disruption signal: a bank-native A2A payment method built on a customer base that skews younger and more price-sensitive than traditional card users has become the second-most-preferred online payment method in three years. That's MDR compression in slow motion. The BNPL omnichannel gap — 48.7% of consumers want BNPL in-store, not just online — remains largely unmet and represents a genuine product opportunity for operators who can bridge physical POS and instalment credit.
Amazon launched Amazon Prime in South Africa on Tuesday at R59/month or R399/year, bundling unlimited same-day and next-day delivery (no minimum order), Prime Video, Luna cloud gaming, and Twitch. South Africa's first Prime Day is scheduled for June 23–29, 2026. The R59 price point undercuts Prime Video's standalone R79/month subscription, and TechCentral analysis notes AWS has a 15-year presence and R46 billion committed SA investment by 2029 — this is not a soft entry.
Why it matters
The competitive read here isn't just 'Amazon vs Takealot.' The R59 subscription is priced to anchor habitual purchase behaviour — the same structural move that made Prime a trillion-dollar moat in the US. Takealot's TakealotMore (R39/month) and Shoprite's Sixty60 are already embedded in different consumer segments; what Amazon brings is the logistics density, Prime Video content lock-in, and brand trust that makes switching cost compound over time rather than simply offering cheaper delivery. For payments operators, Prime Day (June 23–29) will be the first stress test of Amazon's SA checkout and payment method adoption — the payment preferences data from that event will tell the market a lot about how Amazon's default checkout competes with local payment methods and whether they enable Capitec Pay or PayFast natively. Merchants outside Amazon's ecosystem need to watch whether Prime's delivery promise changes consumer expectations about fulfilment timelines across the entire market.
PicPay reported Q1 2026 results: R$3.5 billion total revenue (up 70% year-on-year), adjusted net income of R$169 million (up 92% YoY), credit portfolio growth of 116% YoY to R$28 billion at a 3.7% cost of risk, and ARPAC climbing 55% to R$80.7. The efficiency ratio improved from 61.5% to 46.9%; 69% of revenue now comes from no-risk or low-risk products; collateralised credit revenues grew 272% YoY.
Why it matters
PicPay's results illustrate what the maturation curve looks like for a digital wallet that successfully completes the transition to diversified financial platform. The critical operational data point is the combination of 116% credit growth with stable 3.7% cost of risk — that's not luck, it's evidence of underwriting infrastructure that scaled with the portfolio. Compare this to Nubank's simultaneous 75.7% loan loss provision increase during aggressive growth and you see two diverging execution paths in the same regional market. For SA fintech operators, the ARPAC trajectory is the KPI worth watching: R$80.7 per active customer per quarter, up 55% YoY, means PicPay is successfully moving customers from single-product (payment) to multi-product (credit, insurance, investment) relationships — the unit economics thesis that every super-app in SA is trying to validate.
Meta made its AI customer support and commerce agent available globally in WhatsApp and Instagram DMs on Wednesday, enabling SMBs to automate customer service, product recommendations, appointment booking, and lead qualification without code. The agent connects natively to Shopify, Zendesk, and Shopee; Meta is testing daily briefing summaries and plans to monetise through WhatsApp Business Premium tiers and token-based enterprise pricing.
Why it matters
Perfectly validating Michael Jordaan's fintech survival framework we covered earlier this week — specifically the principle to 'build for existing devices' like WhatsApp — this isn't just a new feature, it's a structural shift in how small merchants can scale customer operations without hiring staff. A spaza shop owner or township trader who already handles orders via WhatsApp can now automate the intake, qualification, and response workflow without an engineering budget or a separate SaaS subscription. The Shopify native integration means merchants running e-commerce can surface catalogue, inventory, and checkout through the same chat thread their customers already use. The monetisation model (WhatsApp Business Premium tiers) is worth watching: if Meta prices this aggressively enough to remain accessible to micro-merchants, it effectively becomes the commerce operating system layer for the informal economy that every fintech in Africa is trying to reach — and Meta gets there via a channel merchants already trust and use daily.
Agentic commerce moves from protocol to production Three separate threads converged this week: Worldline/ING/Mastercard completed Europe's first live end-to-end agentic payment, Visa launched its Intelligent Commerce Connect platform, and Clink opened public infrastructure for AI agents to transact in real fiat. The question has shifted from 'can agents pay?' to 'who owns the liability, the trust layer, and the routing logic when they do?'
African informal economy becomes the primary fintech frontier MTN's repositioned MoMo, the spaza digitisation MOU with Wakanda/UNDP, and Stitch's consumer data showing 48.5% of SA shoppers using Temu/Shein all point at the same structural opportunity: the unserved informal merchant and price-sensitive consumer segment, not the urban middle class, is where the next acquisition volume lives in South Africa and across the continent.
Payment infrastructure incumbents bundle upward to defend margins Airwallex launching billing/subscription management, Adyen winning the UK gov contract, Checkout.com integrating stablecoin settlement via Coinbase, and Mastercard expanding weekend/stablecoin settlement options all reflect the same defensive logic: transaction processing alone is commoditising, so platforms are moving into working capital, billing operations, and treasury to retain enterprise relationships.
SA consumer data signals a bifurcated market Stitch's 2026 report (BNPL at 38.9%, Capitec Pay at 24.6% online preference, peak spend at 20:00 and the 25th of each month) and NielsenIQ's Q1 data (traditional trade outperforming modern, trade-down in durables) paint a consistent picture: SA consumers are digitally active and payment-savvy, but deeply price-pressured — which shapes what loyalty mechanics, checkout flows, and merchant incentives actually work.
The trust and governance gap is now the primary AI adoption bottleneck Across four separate datasets this week — Sinequa's enterprise survey (only 10% have true multi-agent systems despite 51% claiming adoption), Salesforce CFO data (29% capturing ROI vs 97% perceiving value), Forrester's Agentic AI report, and TELUS Digital's CX survey — the consistent finding is that technology willingness outpaces governance readiness. For merchants and operators, this means the competitive advantage lies not in having AI but in having the measurement, audit, and control infrastructure that makes AI deployable at scale.
What to Expect
2026-06-15—SARB comment period closes on the third-draft Payment Ecosystem Modernisation (PEM) activity-based licensing framework — final SA fintech regulatory shape expected Q3 2026.
2026-06-23—South Africa's first Amazon Prime Day runs June 23–29, 2026 — the first major test of Amazon's subscription loyalty engine in the SA e-commerce market against Takealot and Sixty60.
2026-07-01—FNB's repriced bundle structure takes effect: EFT fees eliminated, RTP included in bundles, PayShap at R3 — sets new competitive baseline for SA transaction banking.
2026-07-08—Hang CEO Matt Smolin hosts webinar on AI-driven 1:1 loyalty personalisation for restaurants — practical session on prompt engineering and unit-economics results from deployed deployments.
2026-07-28—ModRetro M64 launches at $229 — AMD FPGA-based N64 clone enters retail, competing directly with Analogue 3D in the premium retro hardware segment.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
950
📖
Read in full
Every article opened, read, and evaluated
226
⭐
Published today
Ranked by importance and verified across sources
12
— The Merchant Desk
🎙 Listen as a podcast
Subscribe in your favorite podcast app to get each new briefing delivered automatically as audio.
Apple Podcasts
Library tab → ••• menu → Follow a Show by URL → paste