Today on The Merchant Desk: the agentic commerce permission wars are heating up, South Africa's payment mix has quietly reset, and McDonald's drive-thru AI has crossed one million transactions — here's what it actually means for merchants.
Pine Labs launched the Pine Labs Payment Protocol (P3P) on Thursday, enabling AI agents to execute UPI payments autonomously without requiring human authentication at transaction time. Built on NPCI's Reserve Pay mandate framework, the protocol lets users pre-authorise rule-based instructions — buy gold at a target price, purchase electronics below a threshold — after which agents execute within defined limits using Grantex for identity and delegated authorization. Early deployments are live with digital gold platform Gullak and electronics retailer Vijay Sales. The protocol is designed to extend beyond UPI to card networks.
Why it matters
This solves the single biggest friction point blocking agentic commerce at scale: the requirement for a human to authenticate every transaction. The mandate-based architecture is elegant — it doesn't bypass security, it relocates it upstream into rule-setting, preserving control while removing per-transaction friction. What makes this significant beyond India is the pattern: UPI is already the world's highest-volume real-time payment network, and P3P demonstrates that emerging market rails can leapfrog to support machine-initiated transactions without retrofitting incumbent bank infrastructure. The Gullak deployment is a particularly clean use case — autonomous micro-accumulation of gold triggered by price conditions is the kind of behaviour that is impossible with card-authorisation flows but natural with mandate-based agents. For payments operators thinking about agentic commerce architecture in South Africa, the analogous opportunity sits on PayShap: the SARB's current focus on real-time retail infrastructure creates a window to embed mandate and delegation frameworks before agentic commerce demand materialises locally.
McDonald's is testing ArchIQ — a Google-powered AI system — at five US locations, where it has processed over one million drive-thru transactions with approximately 90% completed without human intervention. Beyond order-taking, ArchIQ monitors restaurant operations in real time, identifying equipment failures (freezer breakdowns), kitchen bottlenecks, and performance anomalies. The deployment follows McDonald's discontinued IBM voice-ordering partnership, which ran from 2022 to 2024.
Why it matters
The order-completion rate is the headline, but the operational intelligence layer is the more consequential development. ArchIQ isn't just replacing a cashier — it's building a continuous data layer over the entire restaurant operation, from order intake through kitchen throughput to equipment health. That creates a platform dependency that will reshape how third-party payment, loyalty, and analytics providers integrate with QSR systems: McDonald's will control the performance visibility layer, and integrations will need to work within or alongside that intelligence rather than owning it. For QSR operators in South Africa, the jump from IBM's failure to Google's 90% success rate in roughly two years illustrates how quickly the technology has matured. The practical question for local restaurant groups is no longer whether AI ordering works; it's which infrastructure owner they become dependent on when they deploy it. The compare-and-contrast with Toast IQ — reported last briefing at 125,000 restaurants in 90 days — is instructive: centralized proprietary AI (McDonald's/Google) versus distributed platform AI (Toast) are both scaling fast, serving different operator segments.
Adyen has agreed to acquire Orb, an enterprise billing platform specialising in real-time usage tracking and complex pricing contract translation, for $335 million in cash, with the deal expected to close July 1, 2026. Orb co-founders will reinvest a portion of proceeds into newly issued Adyen shares. The acquisition connects Orb's billing infrastructure — designed for high-frequency, variable-cost consumption models — directly into Adyen's payment processing platform for digital enterprises managing AI-driven pricing complexity.
Why it matters
Adyen's rationale makes the strategic logic explicit: AI is shifting software economics from flat subscriptions to usage-based, outcome-based, and consumption pricing, and the billing system is now the bottleneck. When a SaaS platform charges per API call, per model inference, or per transaction outcome, billing must track millions of micro-events in real time and feed that directly into payment authorisation and fraud decisioning. Orb does that. The acquisition positions Adyen to offer genuinely end-to-end monetisation infrastructure — from usage event to invoice to settlement — which is a capability no other payment processor currently bundles. The competitive implication for Stripe is direct: Stripe has long owned the developer-centric billing stack, but Orb's enterprise-grade pricing flexibility was a recognised gap. Adyen is acquiring into that gap. For enterprise merchants and SaaS platforms building on AI-era pricing models, this signals that the billing-payments integration problem will increasingly be solved at the infrastructure layer rather than requiring bespoke engineering. The incubator operating model (Orb as an indirect subsidiary) suggests Adyen intends to preserve Orb's product velocity rather than absorb it into core platform bureaucracy.
Shopify is rolling out native Shopify Payments infrastructure across 17 new markets in three waves between July and September 2026, including Vietnam, Colombia, Romania, the Philippines, and Kenya. Each market receives local acquiring infrastructure and reduced interchange costs. Merchants in pilots report 8–14% checkout conversion lifts versus legacy gateway setups. Kenya's inclusion with M-Pesa integration marks Shopify's most significant push into sub-Saharan Africa and signals the platform's commitment to mobile money-first commerce architecture.
Why it matters
The Kenya M-Pesa integration is the headline for African commerce operators: one of the world's largest e-commerce platforms is now treating mobile money as a primary payment method, not an add-on. That's a normalisation signal that will eventually create pressure on SA merchants — if Shopify Payments extends south, the multi-vendor stack (gateway + APM layer + FX + fraud) that currently serves SA merchants faces competitive compression. More immediately, the 8–14% conversion lift data from existing markets quantifies the cost of fragmented checkout infrastructure: merchants currently managing separate gateway, APM, and FX relationships are giving up real revenue. For third-party payment app partners in emerging markets, the geographic arbitrage window — where Shopify's absence justified standalone payment integrations — is narrowing wave by wave. The question for SA-focused payment operators is not whether Shopify Payments arrives locally, but what differentiated value they offer once it does.
Adding hard conversion data to the 155% digital wallet surge we recently tracked from PayFast, Stitch's 2026 consumer payments research — based on ~3,000 SA consumers and live transaction data — reveals Apple Pay has officially overtaken card as the highest-volume e-commerce payment method on its platform. Digital wallets convert at 92–95% versus ~80% for card. Nearly one in four card transactions fail before reaching the bank, mostly from invalid details and 3DS abandonment, while BNPL adoption sits at 71% among credit-active consumers. Separately, Variable Recurring Payments are now live with major retailers, enabling bank-to-bank subscriptions without card rails.
Why it matters
This is actionable intelligence with a direct revenue implication: card-only checkout strategies in South Africa are now actively destroying conversion. The failure rate data is particularly striking — most of those card failures are recoverable before they ever reach the issuer, meaning intelligent retry logic and method fallback (routing to wallet or EFT on a 3DS abandonment, for instance) is pure margin recovery, not a product roadmap aspiration. The 92–95% wallet conversion rate versus ~80% for card quantifies the cost of not supporting Apple Pay and digital wallets at checkout. For payments operators and merchants, the priority is clear: multi-rail routing with failure-reason awareness, not just adding wallet logos to the checkout page. The VRP signal is worth watching closely — bank-to-bank variable recurring payments bypassing card rails for subscriptions is the structural shift that will pressure card network volumes over the next 18–24 months in SA, mirroring what the UK's UKPI is formalising at the scheme level.
BioCatch's Future of Digital Trust survey of 80 South African fraud, AML, and compliance leaders finds that 96% report AI has already increased fraud sophistication, while 86% identify AI agents as the most exploitable financial system vulnerability over the next year. 80% believe distinguishing legitimate AI-assisted actions from malicious ones will be very challenging, 98% are very concerned about accelerating fraud speed, and 44% report annual fraud losses exceeding $10 million. 91% support inter-bank intelligence sharing as a mitigation mechanism — well above global averages across all metrics.
Why it matters
The 96% figure is notable not as a scare statistic but as a procurement signal: SA financial institutions are nearly unanimous that AI-driven fraud is already operational, not theoretical. The 44% losing over $10M annually quantifies the addressable market for behavioural intelligence and real-time fraud scoring embedded in payment flows — which maps directly to what platforms like Feedzai (whose $9T network rollout we covered yesterday) offer. The 91% support for inter-bank intelligence sharing is the most structurally interesting data point: it suggests the SA market may be receptive to a collaborative fraud network model. For payment operators, the implication is that fraud prevention is moving from a cost-centre to a competitive differentiator — and the urgency around agentic AI (validating the 1,210% global surge in AI scams we noted from Visa) is now matched by practitioner-level action at the operating level.
Two fintech valuation stories this week illustrate divergent routes out of the interchange-compression trap. Ramp raised $750M at a $44B valuation — roughly 30x revenue — by explicitly repositioning itself as a 'Financial AI lab' rather than a corporate card company, inverting traditional card-issuer incentives by aligning economics with customer savings while capturing volume. Separately, Slash scaled from $10M to $300M annualised revenue in 24 months (now valued at $1.4B) by pivoting to vertical banking for e-commerce and Amazon FBA operators after losing 60% of revenue when its primary customer Yeezy collapsed — and by investing to process directly with Visa rather than using BaaS intermediary layers.
Why it matters
Read together, these cases define the two viable escape routes from horizontal fintech commoditisation. Ramp's path: own the AI narrative and data moat so convincingly that capital markets price you as infrastructure rather than fintech — 15–30x revenue multiples versus Bill.com's 2x. Slash's path: go deep into a specific vertical with workflow ownership and direct rail integration, eliminating BaaS margin layers and building switching costs through industry-specific features that horizontal competitors can't replicate. Both paths share a common thread: the pure transaction-fee business without either AI repositioning or vertical workflow ownership is being priced by the market at distressed multiples. The Lightspeed/Marqeta/Payoneer consolidation thesis covered in c_55 is the third outcome — for operators who pursue neither escape route. For SA-focused fintech operators, the Slash playbook is the more immediately applicable model: vertical specificity, direct rail investment, and ruthless opex discipline through AI-driven back-office automation.
DoorDash launched Ask DoorDash on Thursday — an in-app AI chatbot that converts recipe photos, handwritten lists, and plain-language prompts into grocery carts and restaurant recommendations. Early data shows chatbot-assembled carts average 35% higher value than standard app-interface carts and complete five times faster. DoorDash plans to license the technology to external partners including grocers and retailers, signalling that conversational commerce infrastructure may become a B2B software category.
Why it matters
The 35% AOV lift is the most compelling unit economics data point in merchant commerce AI this week — and it comes not from a checkout optimisation tool but from an intent-capture interface. DoorDash is demonstrating that the AI value in food commerce isn't at the payment moment; it's in the discovery and basket-building step upstream. The licensing strategy is the more strategically interesting signal: DoorDash is building what could become a B2B conversational commerce layer for grocers and retailers, effectively positioning itself as infrastructure rather than just a delivery app. For merchant tech operators, this creates both a competitive benchmark and a distribution question. If conversational interfaces systematically lift AOV by 30%+ and a major platform is willing to white-label the capability, the pressure on traditional search-and-browse menu interfaces in QSR, grocery, and hospitality becomes structural rather than aspirational. The comparison to Zalando's disclosure (at Shoptalk Europe this week) that 90% of online content is now AI-generated completes the picture: AI is restructuring the entire discovery-to-checkout stack, not just individual features within it.
Following our coverage of MTN's pivot to an Ant International-powered super app and its pursuit of Nigerian PSSP/PTSP licences, multiple reports confirm the group is finalising the structural separation of its Nigerian and Ugandan fintech operations into a standalone entity. MTN is seeking up to a 30% minority investment from Mastercard and strategic partners to back this transition, moving beyond payment facilitation into balance-sheet lending to target Nigeria's $236B MSME financing gap.
Why it matters
With the corporate architecture now clearer, the spinoff, the Mastercard minority stake, and the lending licence applications are revealed as one integrated move to transform MoMo from a payments utility into a full financial infrastructure layer with its own balance sheet. The Mastercard minority stake is particularly significant: it gives MTN a credentialing anchor for the agentic commerce frameworks Mastercard is rolling out globally, and gives Mastercard African mobile money distribution at scale. For OPay, PalmPay, and Moniepoint, this signals the return of a better-capitalised competitor with telco distribution, a super app interface, and Alipay's product depth. The structural question for the broader SA market is whether MTN applies the same separated architecture south.
Paymentology CTO Tim Joslyn argues in a Thursday analysis that the strategic value in agentic commerce has already migrated away from payment rails — which are commoditised — toward orchestration and programmable trust controls: who sets the spending rules, who verifies agent identity, and who governs autonomous authorization. Joslyn notes Alipay has already deployed delegated AI purchasing capabilities to over 100 million users, and frames the transition from human-speed to machine-speed commerce as requiring a complete redesign of fraud, identity, and authorization frameworks. J.P. Morgan Payments Global Head of Merchant Services Michael Lozanoff made the same pivot explicit separately: the capability questions are largely solved; governance infrastructure is now the competitive battleground.
Why it matters
This framing cuts through the noise of the week's agentic commerce announcements. Visa's Agentic Directory, Mastercard's AP4M credentialing layer, Pine Labs' P3P mandate framework, and the FCA's 'Know Your Agent' signals are all expressions of the same structural shift: whoever controls the permission layer controls the commerce. For merchant tech platforms, this has a concrete implication — order fulfillment, reconciliation, loyalty, and chargeback models that were designed for discrete human transactions are inadequate for continuous, machine-speed spending authority. The fraud and chargeback redesign problem is the most urgent operational gap: current card network dispute frameworks assume a human who authorized a transaction, not an agent operating within pre-set rules across hundreds of micro-transactions. The operators who build governance infrastructure now — not after agentic commerce scales — will hold the leverage.
Spar announced plans this week to deploy a new SAP Finance platform during the 26 weeks ending September 2026, aimed at eliminating manual reconciliation in claims and settlement. This comes as the retailer continues to absorb fallout from a failed SAP rollout at its KwaZulu-Natal distribution centre — deployed February 2023, costing R1.6B in operational disruption and triggering a R168.7M lawsuit — with KZN contributing R123M to Spar's latest operating profit decline. Separately, CEO Reeza Isaacs announced a reboot of the 2U quick commerce app, explicitly stating Spar will not compete with Sixty60 on its own terms.
Why it matters
These two stories are actually one story: Spar's structural competitive disadvantage against Shoprite stems from the same root cause — operational fragmentation across a decentralised franchise model that cannot support the data infrastructure modern commerce requires. The KZN SAP failure illustrates that backend execution failure cascades directly into consumer-facing revenue loss (stock unavailability, poor service levels, lost loyalty). The 2U concession is honest management communication, but it also signals that the quick-commerce battle in SA is effectively over as a competitive contest — Sixty60's centralised logistics and data infrastructure is too far ahead for a franchise operator to replicate on the same model. For payment and merchant tech operators, the Spar situation is a useful case study in sequencing: Spar invested in commerce-layer products (quick commerce app, promotions) before getting its supply chain and systems foundations right — exactly the inverted stack problem identified in this week's Coresight retail data showing retailers spending on analytics before deploying foundational shelf-level visibility. The lesson is not unique to Spar.
A developer has successfully ported Celeste — the critically acclaimed 2018 indie platformer — to the TI-99/4A home computer from 1981. The port, shared to AtariAge forums on Friday, preserves the core climbing-and-dashing mechanics across 11+ levels, with optimisations including removal of particle effects and secondary parallax layers to fit within the TI-99/4A's constraints. The original Celeste was itself built in the fantasy console Pico-8 — making this a port-of-a-port across four decades of hardware.
Why it matters
Anyone who spent time with a TI-99/4A will appreciate the absurdity and craftsmanship of this in equal measure. The TI-99/4A was already considered underpowered relative to its 1981 contemporaries (fighting against the Atari 400/800 and Commodore VIC-20 on a 16-bit CPU that couldn't address its own memory efficiently), and yet 45 years later it's running a platformer that won game-of-the-year awards. The port-of-a-port lineage — Celeste Classic on Pico-8, then Pico-8 logic onto TI-99/4A — is a neat demonstration of how constraint-driven design compounds across generations of hardware. Worth five minutes of your weekend.
The permission layer is becoming payments' most valuable real estate From Pine Labs' P3P on UPI to Paymentology's CTO framing to J.P. Morgan Payments' governance pivot, the consensus is hardening: rails are commoditised, and the competitive moat is now in credentialing, delegated authorization, and spending controls. Every major network announcement this week — Visa's Agentic Directory, Mastercard's AP4M — is really a land-grab for permission infrastructure, not transaction processing.
Africa's fintech maturity phase rewards compliance architecture over growth velocity Three separate storylines this week — enza's B2B-only Ghana entry, Nigerian banking profit normalization, and the BCG telco AI report — converge on the same thesis: the next competitive advantage in African fintech is regulatory depth and institutional trust, not market expansion speed. Operators who treated compliance as a back-office detail are being outflanked by those who built it as a product feature.
South Africa's payment mix has irreversibly fragmented Stitch's 2026 consumer data showing Apple Pay overtaking card by volume, 92-95% digital wallet conversion rates, and a 25% card transaction failure rate before bank contact is the clearest signal yet that SA merchants running card-first checkout strategies are leaving recoverable revenue on the table. This isn't a trend to watch — it's an operational gap with a measurable cost today.
Billing and pricing architecture is the new payments M&A battleground Adyen's $335M acquisition of Orb for usage-based billing infrastructure, combined with the Ramp $44B AI-era valuation narrative and the Stripe Billing margin teardown, signals that the monetisation layer — not just settlement — is where enterprise payment value is concentrating. As AI shifts software economics from seats to consumption, whoever owns real-time billing owns the revenue recognition layer.
AI in commerce is bifurcating: orchestration winners vs. pilot-mode losers Gartner's 80% enterprise agent adoption figure sits alongside the finding that 19% of deployments never reach ROI and 40% of pilots will be scrapped by 2027. McDonald's ArchIQ crossing one million drive-thru transactions at 90% automation is on the winning side; Australian retailers where only 12% have AI-ready data foundations are on the losing side. The differentiator is data infrastructure and workflow redesign, not model quality.
What to Expect
2026-07-01—Adyen's acquisition of Orb (enterprise billing platform, $335M) expected to close — first major test of Adyen's usage-based billing strategy for AI-era enterprise merchants.
2026-Q3—MTN MoMo super app with Ant International launches in Nigeria — first live test of the payments-led super app model with mini-app capabilities for merchants and consumers.
2026-Q3—SARB expected to publish final directives on non-bank payment institution access and Banks Act exemptions — will determine whether independent SA wallet providers can operate without mandatory bank partnerships.
2026-07—Pleo AI agents for spend management enter beta (policy enforcement, invoice processing, cash flow monitoring) — early signal of vertical SaaS agentic finance product uptake in Europe.
2026-07-09—Shopify Payments first-wave expansion into Vietnam, Colombia, Romania, Philippines and others goes live — first real test of whether native Shopify acquiring in emerging markets compresses third-party gateway volumes.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
914
📖
Read in full
Every article opened, read, and evaluated
225
⭐
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