Tom Purves – Product Management, Fintech & AI

Some of my favorite take-aways from our Payment’s Operator Series April 2026 event week on “Rethinking the Merchant Stack for Agentic Commerce”.

The merchant perspective:
– AI referrals disproportionately drive net new customers
– Conversion overall much higher compared to search but payment abandonment is higher
– Payment drop could be due to mix of new customers, lack of pricing clarity in initial AI search (not sure why)
– Accurately tracking AI referal is hard, you are probably significantly under-counting, a good tip is to ask customer’s post check-out how they heard of you
– AI queries are much longer and open ended. At Seat Geek it’s ‘What shows are playing in my town this weekend that could be good for a date night?’ vs google ‘Tickets for Tame Impala Seattle’
– For recommendations AI isn’t just sourcing from a few top sites like reddit, it’s pretty fragmented and segment specific
– AI recommendations are more deterministic than you think. There is variance if you ask only a few times but recommendations cluster around a stable distribution of responses if you ask 20 times.

Insights from the network:
– AI is the fastest growing digital channel (by user adoption) we’ve ever seen
– The biggest power usecases (by token consumption) are Software development (50%), Back-office-Automation (9%), Marketing+Sales/CRM (8%), Finance/accounting (4%), Data Analytics (4%), long tail of everything else 22%. So hint: if you wanted to guess where true angent to agent commerce is going to take off first, maybe think about these segments and usecases first?
– ‘Agentic Commerce’ depends on what surface you are talking about. Alex Chiang models at least 5 distinct flavors: Consumer Platforms (GPTs), First Party Agents (Like Amazon Rufus), Vertical Specific Agents, Enterprise AI Platforms, and “Builder Agents” (e.g Claws)

AI Transformation Perspective:
– Don’t confuse Agentic AI projects with magically solving all your business process automation gaps, if anything fix the latter first as a pre-requisite
– Most people aren’t ready to fully let AI take the wheel on decision making, and particularly on pressing the buy button (yet)
– But it’s highly person and context specific on when you trust the AI more than yourself not to make mistakes, and inevitably we’re all going to start leaning on AI more and more. Esp as the world itself gets overrun with more complexity, AI fakes and new threat environments, it may be crazy not to rely on your own set of agents.
– AI is exploding our ability to generate content, products and new software, which requires new roles and organizational thinking. Editorial and curation skills become more important than ever, it’s not getting cheaper to support in production, to secure them and to go to market. But prototypes are 1000x easier to build. [my view: not launching 90% of your AI prototypes is expected behavior!]

Thanks for Visa for graciously hosting, thanks to Grace Wu for pulling another great event together as well as inviting me to MC

It’s April 2026 and these are the trends to be watching in Agentic Commerce. And it may not be in the places you’re expecting. Is your company’s strategy ready?

1. Websites are still alive and well? AI-originated traffic converts better than search, and ChatGPT pivots from ‘Instant Checkout’ to merchant referral model
2. Merchant’s own AI agents are starting to click with customers
3. You are probably sleeping on the full breadth of B2B agentic opportunity
4. Skills and CLIs are the new MCP. Where’s yours?
5. The wild ride of OpenClaw gives us a glimpse of what truly native agent-to-agent marketplaces can look like

Looking for deeper dives on any of these? follow for coming updates, or just reach out and set up a call!

A joke around SF fintech circles recently “Which are more numerous Agentic Commerce protocols or actual transactions?”. My own agents are tracking about 42 different protocols as of this week. But then again, it’s only Monday.

This much is clear. Consumers and buyers are adopting AI in general for commerce. For inspiration, for the broad knowledge power or AIs and as more efficient interface. AI-led journeys are already eating the traditional commerce funnel from the top down. Merchants are seeing organic and traditional channel click traffic disappearing and are crying out for answers on how to not to just maintain traffic and visibility in a new world of AI intermediation, but also strategically on implications to brand, loyalty and margins.

But those are not, exactly, the questions that the protocol makers, namely the tech companies, AI Vendors and payment processing giants seem to be intent on answering. Rather, many of the specifications launched so far, suspiciously seem bent on solving for a slightly different question. A question that sounds more like, how best could we lock in our own position as a dominant central point of control in this coming new agentic economy. So this is not exactly answering what the actual end users (buyers and sellers) were asking for.

And so we are seeing some slow adoption, and a number of false starts. OpenAI backing away from owning checkout as Walmart reporting that ChatGPT checkout converting 3x worse than it’s own website.

Of course, and in defense of my fellow technologist, we have to start somewhere. And many of these protocols ARE stating to define some useful and necessary primitives, and starting to build up from there. But necessary is not the same as sufficient.

Payment primitives are needed. However, stateless, one-time, one-item purchases is not what makes online business work. LTV is as important as CAC. Recognizing and authenticating returning customers is as important as acquiring new. Just as it’s hard to build a business of off one-time driveby anonymous customers. Sellers are trying to figure out how to build or maintain differentiation, at the same time protocols are trying standardize. There’s a disconnect.

The other thing I see a lot of is, over-thinking about how to port existing commerce flows over to AI. Agentic commerce commerce transition isn’t going to be smoothly spread like peanut butter. It’s going to explode in certain categories and new use cases, while taking much longer than others.

Realize that some things are going to be a lot easier to sell through agents than other things. e.g. Remember how online shopping rolled out the first time. Selling books, travel, auction listing were immediate ecommece success stories, took another decade before Spotify/netflix/Steam business models got sorted out for media, somehow it took 20+ years and a pandemic before digitally ordering fresh food/grocery etc. really got dialed in.

The real solution to Agentic commerce isn’t more solutions. It’s listening to users.

For Sellers, the AI strategy that I advise clients on right now is:

1. Yes invest in GEO/Discoverability. Your AI content strategy matters, and there is a lot how you optimize you catalog for AI readability.
2. But also consider, are you selling the right catalog? The skus, the bundles, the payment terms that work best for agentic flows will not be the same
3. Do invest in differentiating your customers end-to-end experience with your own agents. How to lift and shift your existing business into a new channel, is less likely to succeed that asking properly, how could we treat our customers much better or deliver a version of our products/services that never would have been possible without the help of AI?

For companies focused on buyer enablement:

1. Focus on what is your buyer’s actual goal? For buyers, simply transacting is never the goal itself, it just a step in the way of the goal. Buyers asking for breadth and immediacy of AI models to help them make smarter more informed choices. Or buyers are increasingly turning to agents, or swarms of agents to plan and even run whole projects which might likely entail baskets of purchases, repeat purchases or managing to a budget.
2. Build yourself an end-to-end view of an entire commerce journey. Solving any one step in a buyers’ journey won’t grow you any volume if adjacent steps are still blocked. Completely unblocking journeys
3. Niche-down on where agentic commerce is actually happening, then scale upwards and out. These will categories, usecases and markets where the barriers are lowest, the gains from agentic intermediation are highest, and where the buyers and sellers are most willing. I’m watching amazing AI-native commerce patterns are emerging first, kindof where you’d expect: agents buying api access, access to data, or hiring other ai agents to delegate tasks. Think developer tool, digital marketing and sales stacks and services. There are patterns and playbooks rapidly evolving in these early-adopter niches that absolutely will be extensible to broader and larger segments in short order.

Doing it right: hat tip Vicky Fu’s ClawRouter demo at Lobster.cash demo night 03/04/26. Differentiating by rethinking buying experience for agents.