Agentic commerce: what does it mean for consumer subscriptions and SaaS?

Read this blog and learn

• Why discovery is the one part of agentic commerce you can act on today, and how most of the SEO work you have already done still counts toward it.
• The data behind the shift: 47% of US holiday shoppers used AI last season, 94% of procurement professionals now use GenAI, and one company is seeing AI-sourced web forms convert four times better than organic traffic.
• Why the current checkout protocols, OpenAI's instant checkout and Google's UCP, were not built for subscriptions, and which of the two is closest.
• How agentic selling gets an order of magnitude harder in B2B, told through a medtech buyer asking an agent for an oncology billing solution.
• Why the renewal, not the first purchase, is where the real battle for subscriptions will be fought.
• The honest case for buying rather than building when your billing and catalog are on the line.
• How three leaders are actually getting their teams up to speed on AI, and why being professorial about it does not work.

Agentic commerce is one of those phrases that can mean everything and nothing at once. So for Limio's first ever webinar we tried to make it concrete. What does it actually change for businesses that sell subscriptions and SaaS, what can you do about it this quarter, and what is still hype you can safely ignore for now?

The session, Agentic Commerce: What does it mean for Subscriptions and SaaS?, brought together three people who are living the question rather than theorising about it. Errol Denger runs Zuora Commerce, Zuora's new monetization catalog and experiences layer. Chris Brenneman oversees the digital experience for OpenText's cybersecurity consumer and SMB brands, which means he owns e-commerce and web operations across two B2C sites and a lead-generation SMB site. Amaury de Closset, Limio's co-founder, hosted and moderated.

Between them they cover consumer subscriptions, complex enterprise SaaS, and everything awkwardly in between. What united them was refreshing honesty about where this technology really helps today and where it is still a demo waiting to grow up.

You are either overwhelmed or behind, and that is normal

The most reassuring part of the whole session was the admission that nobody has this fully figured out. Amaury described going down a rabbit hole after Anthropic released Opus 4.5, to the point that, in his wife's words, he was talking to Claude morning and night and was fairly insufferable for a few months. Errol, a year and a half into the shift, said his fiancée now asks whether he is talking to his robots again. Chris put himself at the other end of the spectrum, calling himself late to the party and describing OpenText's use of agentic tooling as being at a crawl.

That gap is the point. As Amaury summed it up, you cannot quite get it right. You are either feeling behind or you are overwhelmed and using it too much. If you feel one of those things reading this, you are in good company.

A quick definition to anchor everything that follows. An AI agent is a software system that uses a large language model to pursue goals and complete tasks on behalf of a user. Commerce is the buying and selling of goods and services. Agentic commerce is what happens when you introduce an AI agent as a party in that exchange, whether it acts as the buyer, the seller, or an assistant to either side. The rest of the conversation split neatly along the customer journey: discovery, then transaction.

Start with discovery, because it is the part you can act on today

Errol was unequivocal about where to begin. Discovery is where every subscription business should start, and the good news is that it is relatively easy. Before touching your site, look at how buyers now behave. During the last US holiday season, 47% of shoppers used AI. In B2B the number is higher still: Forrester puts GenAI use among procurement professionals at 94%. And in a Zuora survey, every single company said optimising their site for generative engine optimisation and answer engine optimisation was critical to their future.

The comforting part is that most of the groundwork is work you already know how to do. Everything you built for SEO still applies: deep, authentic content, case studies, FAQs, linked content, a clean sitemap, a sensible robots.txt, and now an LLMs.txt. Errol's one addition worth acting on is scenarios. If your platform serves thirteen industries, spell out all thirteen use cases explicitly, because that is the context an agent needs to recommend you well.

Chris shared what this looks like in practice at OpenText, and the numbers made everyone sit up. His team has been injecting structured data, JSON-LD, into the SMB site, first by hand and now moving to auto-publishing so reviews and pricing update themselves. Google Analytics has started reporting AI as a source medium, and while it is still a small slice, it is growing steadily. The striking part is quality. Web forms from AI-sourced traffic are converting at roughly four times the rate of organic, and organic used to be his gold standard for lead quality. His read is that people currently trust being pointed somewhere by an AI, so they arrive more ready to engage. For context, when Limio optimised online checkout the conversion lift was around 13%, so a 4x difference in lead quality is a different order of magnitude entirely.

One practical note from Chris on the frameworks themselves: schema.org will tell you whether your structured data is valid, but not whether it is the right structure for the page. Homepages, product pages, and content pages each want a different shape, and the rules are still being worked out by the various AI engines. Expect to iterate.

The checkout layer is not ready, especially for subscriptions

If discovery is the encouraging half of the story, the transaction layer is the wait-and-see half. Errol was blunt about the two big contenders.

OpenAI's instant checkout, in his assessment, was poorly executed. It could not handle complex carts, discounting, or tax calculation, it carried a 4% fee, and OpenAI wanted to be the merchant of record. He described it as effectively dead before it launched.

Google's UCP is closer. It offers two models worth understanding. In the first, Gemini acts as a kind of headless cart: you discover products inside Gemini, but the transaction routes back to your commerce platform, your cart does the heavy lifting on tax and calculation, and you remain the merchant. The second is closer to a modern version of punch-out, where a buyer starts in Gemini, identifies what they want, and then completes the purchase back on the merchant's own site. It is still early, with Shopify, Target, and a handful of large vendors involved.

Here is the catch that matters most for anyone reading this. None of these standards were built for subscriptions. The temporal element, advanced subscription types, ramps, usage, and contract terms simply are not represented yet. Limio and Zuora have both mapped and normalised the existing standards and are working with Google and the standards bodies to get subscription concepts included. As Errol put it, the more subscription businesses unite behind that effort, the better the chance of shaping it. For now, though, the transact layer is not ready for us.

Consumer subscriptions: the renewal is the battleground

Chris, who straddles B2C and B2B, made the point that most agentic commerce research today is retail-first, and that adoption will track your ideal customer profile. A base skewing 18 to 24 will move faster than one skewing 50 and over.

The more interesting shift he sees is in how demand gets created. Seasonal campaigns take a back seat when the trigger is no longer the time of year but the individual's intent, spotted by an agent. If an agent has visibility into your purchases and sees you bought a laptop, it might surface antivirus or security software without any human ever seeing a Facebook ad. The job stops being about capturing a human's attention and starts being about being the answer an agent reaches for.

Then there is the renewal, which is where subscriptions differ sharply from retail. The familiar playbook of a deep first-year discount followed by an auto-renewal at full price relies on human inertia. Chris expects that to come under pressure once an agent acting in the customer's interest knows the renewal date, quietly compares alternatives, and even negotiates with other companies' agents. That raises a new question for marketing and product teams: what do you offer to keep a customer when the thing deciding is not sentimental about your brand? Amaury's summary was that the first purchase will look a lot like retail, but the renewal is where the real battle for subscriptions will be fought, and nobody is quite sure yet how it plays out.

Chris threw in one more prediction worth noting: pay-to-play review lists may lose their power. If a human is no longer reading the top-ten roundup, an agent probably is not weighting it either, especially the ones that read as advertising rather than genuine user reviews.

B2B is an order of magnitude harder

The models are impressive. Errol pointed out that today's frontier models understand context well enough to tell the difference between a summer dress for a reggae show and one for the Belmont Stakes. But B2B, he argued, is orders of magnitude more complex.

His example landed the point. One Zuora customer is a medtech provider preparing for agents to arrive with requests like, "I need a multi-office billing and patient management solution for an oncology practice." Serving that means identifying the number of offices, selecting the right EMR module, and recognising that an oncology practice needs a level of collaboration, imaging, and file management that a general practitioner's office simply does not. To handle it, Zuora is building an agent orchestrator that expects agents to arrive with dozens of terms and conditions at once and to need the best product at the right price back in milliseconds.

There is healthy scepticism too. Sales-led growth carries an enormous amount of human and institutional knowledge in the solutioning process, and many of Zuora's customers doubt that agent-to-agent interfaces will handle truly complex deals any time soon. Chris agreed, listing the real-world friction: multiple decision makers who may each end up with their own agent, high switching costs, existing contracts with their own SLAs and payment terms, and legacy billing systems where an agent has to work out which platform to even transact on.

Underneath all of it sits data hygiene, which Amaury called the real first step. Before you touch a framework or publish a catalog, an agent needs clean data, and that is hard at scale when you are carrying multiple product lines, legacy tech debt, grandfathered subscriptions from a decade ago, and reseller permissions. Chris's conclusion was that organisations need to look at their own house first and ask how they are set up internally to transact with agents, because for many that will not be trivial.

Will agents replace salespeople? Not while the golf course exists

On whether B2B buying goes the way of consumer e-commerce, Chris was measured. As long as there are senior decision makers who value relationships, and who want the lunch and the round of golf, there will be salespeople. This is not a light switch that flips overnight.

Where agents earn their place first is the long tail. Discoverability gets a prospect to your site. Most buyers do not want to talk to sales on day one, so the site handles awareness and education, and a well-trained agent can nudge someone to raise their hand by answering a niche question they cannot resolve alone. Chris was candid that most AI chat today is not very helpful yet, and the interesting work at OpenText is training an agent by pulling in the domain expertise sales reps carry in their heads, the practical answers that never make it into a white paper.

Amaury framed Limio's own view here. An agent can hold a conversation about a product in a way a pricing page never can, because there are only so many scenarios you can fit on a page. It will not replace the golf course, but it can take on more of the complex long-tail sale, which is exactly why Limio is launching its selling agent.

Buy versus build: why you probably should not vibe code your billing

The buy-versus-build question drew the firmest answers of the session. Chris described OpenText's position: at a large company there is real rigour around AI controls and policy compliance, so just because you can vibe code something does not mean you should, particularly where data is involved. Leadership's view is to spend valuable engineering time on core competencies, the SaaS products themselves, and to lean on companies that focus squarely on one thing. He was open that OpenText is a Limio customer for exactly this reason: it let them hand off checkout efficiency and free their own engineers for the work only they can do.

Errol was blunter. People will try, and if you look at some of the early throwaway sites they are getting hacked. Zuora augments its own development with AI and recommends everyone do the same, but the pattern-matching that makes these models fast also makes them miss the critical, deterministic details, the ASC 606 revenue rules and the dependencies that billing lives or dies on. That still needs extensive human oversight. Maybe in five years the picture changes, but for now the precision is not there.

Amaury tied it back to the core issue. These are probabilistic models, and billing is one of the areas where you really do not want the system guessing. Vibe coding gives you an easy and flattering day one, because the model indulges you, and then reality arrives in the form of sales tax, card handling, and merchant relationships. The unglamorous jobs turn out to be real jobs.

Bringing your team along

Amaury closed on leadership, because agentic commerce is as much a change-management problem as a technical one. His founder's view was simple: you have to use the technology yourself and go deep, because you cannot credibly tell your team what to do with a tool you have never touched.

Errol offered the most practical playbook of the day. Start with a twenty-dollar Claude account, which will effectively train you and tell you what to learn next. Lean on the material Anthropic publishes, the upper-level computer science courses Stanford puts online, and Andrew Ng's work. Then build a rhythm inside your organisation. At Zuora that takes the form of a recurring call they call the AI-Empowered Product Manager, where everyone gets hands on, rotates who leads, and pools what they are learning. His warning was pointed: if you try to be professorial and simply instruct your teams, it will not work, because the field is moving too fast for anyone to be the sole expert.

Chris's version was grounded in web operations. Focus on discoverability as the easy win that gets your foot in the door, then widen the conversation across marketing, product, and engineering, because this is no longer only about the website. His closing image was the one that stuck: you can be part of the wave, or you can be underneath it, so start the planning conversations as soon as you can.

Which is really the whole message of the session. You do not need to have it figured out. You do need to start experimenting, because if you wait for certainty, you will be behind. The goal, as Amaury put it, is to help people surf the wave rather than drown in it.

Watch the full webinar

Watch the full discussion with Errol Denger, Chris Brenneman, and Amaury de Closset for the complete conversation, including the frameworks each of them recommends, the live audience questions, and more detail on what to build now and what to wait on.

Watch the Agentic Commerce webinar recording →