© Starpoint LLP, 2025. No part of this site, blog.starpointllp.com, may be reproduced or retransmitted, in whole or in part, in any manner without the permission of the copyright owner. Also, see our Legal/Disclaimer (this is a highly opinionated and partially informed blog). Enterprise readers, please consider Enterprise Subscription (not required for Starpoint Clients).
I’ll be honest, I’ve been watching the “agentic commerce” hype train with a healthy dose of skepticism. The idea that AI agents will soon handle all our shopping feels like a solution in search of a problem. Yet, looking at the data, I have to admit something massive is happening under the surface. We are in the midst of a fundamental change in how the internet works, and while the “Agentic Era” is still 3+ years away, the tremors are already breaking the internet’s business model.
Agentic Hype? Or is the web really breaking? (see blog)
Cloudflare CEO Matthew Prince recently dropped some staggering data: where a Google search might result in a 2-to-1 ratio of pages crawled to user actions, OpenAI consumes 250 pages for a single user action. For Anthropic? A mind-boggling 6,000 to one.
The old “quid pro quo” of the web—publishers create content, Google sends traffic is dead. AI is extracting value and returning nothing but a summary.
This isn’t a future problem. Between January and May of this year, AI-referred sessions to e-commerce sites exploded by 527%. Consumers are offloading the mental work of shopping to AI, with 36% of GenAI users already replacing search with these assistants. While AI conversion rates initially lagged, that gap is closing fast—from 43% lower conversion to just 9% as of February 2025.
Tom’s Projection 5% Traction starts in 2027
Despite the traffic explosion, we are in a holding pattern on agentic transactions. Today’s agentic commerce is struggling to break free from the “Device ID” paradigm that underpins our entire risk ecosystem.
- Stripe’s ACP works today because it pragmatically leverages existing device graphs to patch the lost data signals.
- Google’s AP2 is technically brilliant—using cryptographic mandates—but it’s fundamentally misaligned with how card rails currently handle liability and disputes.
This tech flux will get resolved, but the entire business model of everyone is evolving. Merchants are not geared to accept agentic transactions and walled gardens are building their own; new rules must be written, there is no role for mobile (yet) and Google is currently overhauling its core business model from a search/ad engine to a platform for the agentic economy. But until alignment happens between tech and all stakeholders, we are in flux. I don’t see agentic commerce gaining real traction until mid-2027.
Card Network Revenue Growth
Here is what investors are missing: Agentic commerce will not substantially change GDV (consumers won’t buy more), but the role of the networks greatly increase as they are the only entities capable of governance (and VAS) on this 10x growth in interactions.
Using updated 2024 data, let’s assume 80% of the $1.2T eCommerce spend runs on cards ($960 Billion).
As most of you know, network fees are currently just 0.05% (5 basis points) of volume. With an average transaction size of $40, the networks make fractions of a penny ($0.002) on the transaction itself.
Agentic Commerce changes the equation via Unbundling.
One human intent ($40 basket) unbundles into a swarm of specialist agents. Taking a page from JPM’s open banking fees, let’s baseline on two simple fees for illustration (a starting point)
- Authentication (The Entry Fee): Happens once per agent/consumer binding. ($0.05 fee).
- Authorization (The Usage Fee): Happens for every interaction (negotiation, inventory check, micro-payment). ($0.01 fee).
The Projections: Two Futures for 2029
The model for how often these fees are incurred is uncertain. In a “future web” decentralized model where millions of retailers compete for every order, or where 6 big retailers dominate in their “Walled Garden” (Centralized). For a 2029 estimate, I would bet that 70% of volume will remain with Amazon, Walmart and the other big retailers. But, I’ll leave it to the readers to bake out that model.
Scenario A: The Bull Case (Decentralized Swarms)
- Assumption: 25% Agentic Penetration ($240B volume / 6B Human Intents).
- Behavior: Agents are highly specialized and chatty. One human intent triggers 20 authorizations as agents negotiate across the open web.
- Revenue Math:
- Authentication: 6B $\times$ $0.05 = $300 Million
- Authorization: 6B $\times$ 20 calls $\times$ $0.01 = $1.2 Billion
- Total Net New Revenue: $1.5 Billion
Scenario B: The Conservative Case (Walled Gardens)
- Assumption: 25% Agentic Penetration, but large retailers (Amazon, Walmart, Shopify) dominate.
- Behavior: The “Universal Cart” consolidates the complexity. The agent talks to the retailer once. The chatter happens internally on the retailer’s servers, not the payment rails. The number of calls drops by 1/10th (e.g., 20 calls become 2).
- Revenue Math:
- Authentication: 6B $\times$ $0.05 = $300 Million (Trust is still needed). Note that this is where large platforms will work directly with the large retailers to manage joint authentication. So these fees are at risk, the control point for card network is the rules around authorization necessary (and a bank liability shift). .
- Authorization: 6B $\times$ 2 calls $\times$ $0.01 = $120 Million.
- Total Net New Revenue: $420 Million
VAS US Market Share Split (Visa vs. Mastercard)
Assuming split of 70% Visa / 30% Mastercard against 2024 annual revenues ($36B and $28B, respectively). Note that net revenue would be more appropriate comparison.
|
Scenario |
Visa Upside (70%) |
Impact on Visa Rev |
Mastercard Upside (30%) |
Impact on MA Rev |
|
Conservative (Walled Garden) |
$294 Million |
+0.8% |
$126 Million |
+0.4% |
|
Bull Case (Open Web) |
$1.05 Billion |
+3.0% |
$450 Million |
+1.6% |
The Verdict:
These fees are contrived, as well as volume estimates. But I see agentic as a significant VAS tailwind that investors should consider (starting in 2027).
Even in the conservative “Walled Garden” scenario where retailers compress the noise, the networks role in authentication is central to unlocking Agentic and should result in incremental VAS. While the “Open Web” vision of autonomous agent swarms won’t fly in the US, it is likely to win in many other markets, while ROW number were not assessed, Visa and Mastercard are should incur a tailwind of 3% incremental revenue (not assuming 0% eCom/agentic GDV growth).
A much BIGGER tailwind is possible if Networks can help unlock the Gordian Know of Agentic Monetization. Google is on track to solve it themselves.. Everyone else needs help (see long blog).
Related Deep Dives: