Payment Pulse – Ep. 47

Here’s how vertical SaaS platforms can use AI to take on valuable customer work, build trust through defined actions, and connect payments to broader workflows.

Also Listen On:

TL;DR

Episode Transcript

Shannon: Hey, Michelle. How are you today?

Michelle: Doing good. How are you?

Shannon: Good, good. Hey, so I actually saw something this week that I think is perfect for what we’re talking about today

Michelle. Ooh, tell me about it.

Shannon: So Visa released some research on agentic commerce. 72% of US consumers that they surveyed said that they’ve used an AI assistant, but only 23% said they trust generative AI to actually handle a payment transaction for them.

Michelle: Interesting. Yeah, I’m, I’m probably in the other 77%.

Shannon: Yeah, me too. Me too, for sure. So I’ll let AI, like, help me research a hotel, but I’m not ready to say, “Here’s my credit card. You decide. Go book it.”

Michelle: Yeah, and there’s a pretty big leap between those two things.

Shannon: There is. So recommending something is one thing, but actually doing something for me is different.

Michelle: Especially when doing it involves money.

Shannon: And that’s really what we’re getting into today. So if you’re new to Payment Pulse, I’m Shannon.

Michelle: And I’m Michelle.

Shannon: And we spend a lot of time working with vertical SaaS platforms and looking at how payments, technology, and changing customer expectations are reshaping their businesses. And lately, it’s almost impossible to have a conversation without talking about AI.

Michelle: But so much of the conversation still starts with, “Where can we add AI?”

Shannon: And I wonder if that’s actually the wrong question. So if I’m running a vertical SaaS platform and my leadership team is saying, “We need an AI strategy,” where should I actually start?

Michelle: Probably not by opening the product roadmap and asking, “Which features can we add AI to?”

Shannon: Why? Why do you say that?

Michelle: Because you’re starting with the technology instead of the customer problem. I would start outside of the software. So what is the customer actually trying to accomplish? What are they still doing manually? Where are people still making repetitive decisions? Where are the handoffs, and where does the process slow down?

Shannon: Yeah, so we’re not really looking for AI features.

Michelle: No, we’re looking for work.

Shannon: Yeah, and work, it can really mean a lot of things. So it could be like following up with a customer, scheduling, reconciling something, um, reviewing information and deciding what happens next. It could also mean collecting a payment, issuing a refund, or taking some other kind of financial action.

Michelle: And that’s why looking at the whole workflow matters. If you only look inside your existing product, you see what your software already does. If you look at the customer’s entire workflow, you start seeing everything that still happens around the software.

Shannon: Mm-hmm. And potentially, that’s where some of the bigger opportunities are. So is there a danger that companies are going to spend the next two years putting together copilots when the bigger opportunity is actually changing the workflow?

Michelle: I mean, I think so. It is possible. A copilot can absolutely be useful, but “We added AI” versus “We materially changed the customer’s outcome” are two different things.

Shannon: Mm-hmm. Yeah. Okay, so I’m sold on looking at the work instead of the feature list, but now I’ve mapped this workflow and there are 20 things that people are doing manually. How do I know which of those is worth going after?

Michelle: Yeah, that’s a good question. I would start with value. Not, “Can AI do this?”, but “If we take this work off the customer’s plate, does something meaningfully improve?” So, do they save time? Are they making more money? Does it reduce costs, get paid faster, serve more customers, or make fewer mistakes?

Shannon: Yeah, because technically possible isn’t the same thing as valuable.

Michelle: Right. And I look for work that happens frequently enough that solving it matters.

Shannon: Yeah. So let’s make that more concrete. So say on the field service platform, a job gets completed, somebody has to do the work, communicate with the customer, generate the invoice, collect the payment, reconcile it.

Michelle: And then maybe schedule whatever comes next.

Shannon: So if I only look at one screen in the software, I miss most of that.

Michelle: Yes. So the better question is, “What has to happen from the beginning of this workflow all the way to the outcome?” Then you can start asking where software could take on more responsibility.

Shannon: And I think payments gets interesting here because sometimes we talk about payments like it’s a separate workflow.

Michelle: Yeah, and you know, a lot of the time it’s actually part of completing the workflow. It becomes increasingly important as software starts doing more of the work.

Shannon: So here’s where I think people can get carried away. We start talking about agentic software, and suddenly we’re imagining an AI agent running the customer’s entire business. Is that really where a vertical SaaS platform should start?

Michelle: Well, I wouldn’t. I’d pick one clearly defined problem and one outcome that matters.

Shannon: Yeah, and that actually sounds a lot like the traditional vertical SaaS playbook.

Michelle: Yeah, it does.

Shannon: Yeah, so find a wedge, solve something really well, and then earn the right to expand.

Michelle: Yeah, so the technology is changing, but that logic still holds. You don’t have to automate an entire workflow on day one. Maybe the platform starts by identifying what needs attention, then it recommends an action. Eventually, it could take that action with defined parameters.

Shannon: So in essence, are you saying “start small” doesn’t mean “think small”?

Michelle: Yeah, exactly. You’re just proving that the platform can create the outcome reliably.

Shannon: Yeah, and you’re building trust.

Michelle: Yeah, which matters a lot more once we start talking about autonomy.

Shannon: So the wedge itself may be narrow, but you’re choosing it partly based on where it could eventually lead.

Michelle: Yes. I’d want to understand the adjacent workflow. If we successfully take on this piece of work, what could the platform reasonably do next?

Shannon: Yeah, so I wanna go one level deeper because I think this is where the conversation becomes much bigger than the product strategy. If software starts doing more of the work, does that change what the software is worth?

Michelle: Potentially, because now you’re not only comparing the value of the platform to another piece of software, you’re potentially comparing it to the cost and value of the work itself.

Shannon: And this is where some of the vertical AI research gets really interesting. Bessemer Venture Partners has predicted that vertical AI’s total market capitalization could eventually be at least 10 times the size of legacy vertical SaaS.

Their reasoning isn’t simply that AI companies get higher valuations, it’s that vertical AI can potentially address portions of the much larger services economy and not just traditional software budgets.

Michelle: Yeah, and that goes back to what we were talking about earlier. If you’re only looking at your software feature list, you may miss the work customers are spending money on outside your software.

Shannon: There’s an interesting parallel with payments too. Vertical SaaS has already gone through one expansion of the business model. A platform could charge for access to the software. Then embedded payments created another opportunity because the platform could participate in the economics of transactions happening through that software.

Michelle: Yeah, and AI potentially introduces another question. If the platform starts performing more of the work, how do you monetize the value it’s creating?

Shannon: Maybe it’s still subscription-based, maybe it’s usage, maybe it’s tied to a workflow or an outcome, or maybe transactions are part of the economics.

Michelle: Yeah, or some combination.

Shannon: Which raises a question I think a lot of software platforms are going to have to wrestle with. If your platform starts doing the work, should you still price it like software?

Michelle: Yeah, and right now I don’t think there’s one true answer to that yet.

Shannon: Yeah, I don’t either, but I think it’s a question worth asking now. So let me push back on all of this for a minute because it’s easy to sit here and say, “Let the software do more.” Would you actually let it though?

Michelle: I think it depends on what it’s doing.

Shannon: Yeah, I’d have to say that’s my answer too. I’ll let AI summarize a document for me right now, but would I let it issue a $10,000 refund without me asking? No.

Michelle: Yeah, and you know, those actions have completely different consequences.

Shannon: Yeah. So maybe the question isn’t just, “Can the agent do this?” It’s also, “How much authority have we earned the right to give it?”

Michelle: Yeah, I really like that framing. You can think about it as a progression. You know, at first, the software may identify something, then recommend what should happen, then maybe take a defined action with approval, and then eventually for certain actions, it could operate more independently with established rules.

Shannon: Which is basically “crawl walk run”.

Michelle: Yeah, exactly. And you don’t get to run because the technology suddenly becomes capable of it. You get there because the system has demonstrated enough reliability, and you put the right permissions and guardrails around it.

Shannon: Mm-hmm. Yeah, that’s also why I thought that the Visa research was so interesting because people are already using AI, but the trust drops when you ask whether they’re comfortable letting it actually handle the transaction.

Michelle: Yeah, because now there are consequences.

Shannon: Right. The more consequential the action, the higher the trust threshold. So let’s cross that line, okay? So what happens when completing the work requires money to move?

Michelle: That’s where the connection between agentic AI and payments really gets interesting. So we’ve spent years talking about embedded payments, bringing the transaction inside the software experience. But if software can increasingly take actions on someone’s behalf, a transaction can potentially become one of those actions.

Shannon: Mm-hmm. Yeah, and I think that’s an important evolution. Embedded payments put the transaction inside the software, but agentic commerce could make the transaction part of that action.

Michelle: With an important qualifier: authorized action.

Shannon: Yeah, that’s critical for sure. So we’re not talking about giving an AI agent unlimited access to a bank account and hoping for the best.

Michelle: Yeah. So we’re talking about permissions, spending limits, approval requirements, authentication, defined conditions, and knowing who’s authorizing what.

Shannon: Mm-hmm. And the payments industry is already working on those problems. So this isn’t just a theoretical discussion anymore.

Michelle: Yeah, and, you know, think about what that could mean inside a vertical workflow. Maybe an agent identifies inventory that needs to be reordered. Today it alerts someone. Tomorrow it might recommend the order. So eventually, within the right permissions, it could potentially place the order and complete the transaction.

Shannon: Yeah. Or think about our field service example. So the platform knows the job is complete, it knows what’s owed, it communicates with the customer, the customer authorizes the payment, the transaction happens, then reconciliation happens. So we’re getting closer to the platform orchestrating the outcome rather than just providing the tools somebody uses to get there.

Michelle: Yeah, and I think that’s the bigger point. You know, agentic commerce isn’t interesting just because AI can buy something. It’s interesting because commerce can become part of an agent completing a larger workflow.

Shannon: So that’s the piece that I would be watching if I were running a vertical SaaS platform. Not just can the agent transact, but where does a transaction sit inside the work my customer is trying to accomplish? So let’s make this practical now. If somebody is listening and has a product meeting next week, what do we want them to do differently?

Michelle: Sure. So I would pick one important customer workflow and map it from beginning to end. So don’t just map what happens inside your product, map the actual work.

Shannon: Yeah, and then start asking questions. Where are people still doing things manually? Where are they making repetitive decisions? Where does the work stall? Where does money move? And where are the handoffs?

Michelle: Which of those steps, if the platform took it on, would actually create a better outcome?

Shannon: Then pick one, not 10.

Michelle: Yeah, just one valuable outcome.

Shannon: Yeah. So prove it.

Michelle: You go ahead and build trust.

Shannon: Yeah. And then earn the right to do more.

Michelle: Yeah, that’s probably the part I would emphasize the most. Becoming more agentic doesn’t mean automating everything. It just means being deliberate about where the platform can take on more responsibility and create more value.

Shannon: And if you want to go deeper, we put our full executive discussion with Luke Sophinos, Mark Passifione, and Daniel Burton together with our practical from vertical SaaS to agentic roadmap. You can get both at go.xplorpay.com/agentic. So Michelle, final question. What’s the one question you want people thinking about after this episode?

Michelle: Sure. I would stop asking where can we add AI, and instead ask, what work are our customers paying people to do today that our platform could start doing for them?

Shannon: Yeah, I think that’s a pretty good place to end. Well, thank you, Michelle.

Michelle: And thank you, Shannon.

Shannon: Yeah, and thanks everybody for listening. We’ll see you next time on Payment Pulse.

FAQs