Why the most useful payments AI is built from the inside out

Most of the AI tools being built for payments today fall into one of two camps. The first puts AI on top of the experience: chatbots for customer service, dashboards you can query in plain language, reports generated on demand. These are useful, but on their own they aren’t much of an advantage. Anyone with access to a frontier model and a good engineer can build an interface like this.

The second camp builds AI that changes financial outcomes. For example, think of routing that improves authorisation rates through payment orchestration or fraud detection that learns from real transaction behaviour instead of synthetic data, and tools that spot a drop in conversion or a spike in chargebacks days before it shows up in the accounting software. This is where data becomes the deciding factor, and where payment infrastructure providers hold ground that’s harder to replicate.

The interface is easy, but the data underneath isn’t

It’s getting easier and easier to enter the first camp, but entering the second gets harder over time because it depends on years of payments data and on knowing what that data really shows: why a card issuer declines on a Saturday afternoon, what a duplicate charge looks like, which chargebacks point to fraud and which point to a checkout bug.

A natural language question is only as good as the answer behind it. If an AI tool can’t see your transactions, it can only guess. A tool that sits inside your payment flows can tell you exactly what happened, why it happened and how much it cost you in rands.

Where Stitch Pulse fits

That idea is behind Stitch Pulse, our new suite of AI-driven software for South African enterprises. Pulse doesn’t sit on top of payments as an extra layer. It’s built on the transaction data Stitch already processes for our clients.

The first product, Pulse Analytics, gives every team member their own AI agent that works across the business’s payments data around the clock. How it helps depends on who’s using it. Take, for example, a national retailer partway through a big spring sale:

The payments lead: finding a weekend problem on Monday morning

On Saturday afternoon, card conversion on one bank’s cards drops by two points. Nobody notices over the weekend, but on Monday, the payments lead’s morning brief puts it at the top: how many customers couldn’t pay, and how much in sales didn’t come back. She asks Pulse, “Which issuing bank had the biggest drop in card approvals, and why?” Pulse narrows it down to three card ranges. Now she can go to the bank with evidence, and not just a hunch.

The CFO: explaining the numbers before the board asks

Revenue is up 12% month to date, and the CFO needs to know why before the next exco meeting. She asks, “What’s driving the bump in revenue?” Pulse splits the increase three ways: the sale, existing customers spending more per visit, and first-time buyers. She turns that answer into a live dashboard with one prompt and has a summary sent to her inbox every Monday.

The risk team: knowing which chargebacks to look at first

Chargebacks have nearly doubled on last month, so the risk manager sets up an alert in one sentence: “Alert me if our chargeback rate goes above 0.75%, and flag any customer with more than two chargebacks in a month.” Pulse shows that a small group of customers is behind most of the increase. That means the team can investigate a handful of accounts instead of going through the whole book, saving time.

Customer support: sorting out a double charge in one conversation

A customer says they were charged twice, so the support agent asks Pulse to refund the transaction. Pulse confirms it’s a real duplicate with the same card and the same basket, charged again seconds later after a checkout retry. It then prepares the refund within the agent’s refund limit and nothing moves until the agent confirms. Once the refund is paid, Pulse generates a proof of refund with the network reference, ready to send to the customer. Every step is recorded in the audit trail.

The marketing lead: checking whether the sale actually worked

The marketing team wants to know if the 20% off promotion was worth it. Pulse compares category sales against the same week last month and splits out new and returning customers. It also shows which channel the new buyers came from, so the team can plan the next campaign based on concrete results instead of guessing.

Each person starts their day with insights that are important to their role, ranked by Rand impact, before they’ve asked a single question.

Pulse Analytics does have a plain-language interface, just like tools in the first camp, but the difference is what sits underneath it. Because Pulse runs on data Stitch already processes, there’s no new integration or data project. Teams can see their full payments history from the day they switch it on.

What this means for enterprise and e-commerce businesses

The AI tools that create the most value in payments are unlikely to come from general-purpose AI vendors who have bolted on a payments module. They’ll come from companies that are part of the payment flow and have spent years learning how it works.

That changes what “good” looks like when you evaluate AI vendors. A polished demo is less important than the answers to a few harder questions:

• What data does the tool actually see: your real transactions, or an export someone has to keep up to date?
• Does the team behind it understand payments, or are they applying general machine learning to an industry they don’t know well?
• Can it explain why something happened, or does it only describe what happened?
• Where does your data live, and is it used to train third-party models?

For Pulse, the answer to that last question is no. Pulse runs inside the same ISO 27001 and PCI DSS Level 1 certified environment Stitch uses to process payments, and client data is never used to train third-party AI models.

The underlying logic

The AI advantage in payments comes from how the industry is structured. The companies that process payments are the ones closest to the data that makes payments AI useful.

Payment orchestration that routes intelligently, online payments infrastructure that works reliably across multiple payment methods, and fraud detection that learns from real signals all improve over time when they’re built by teams that understand both the technology and the domain.

The next wave of useful financial AI will be built from the inside out, by companies that were already deep in the data before the AI opportunity became obvious. Pulse is our first step in that direction.

FAQs

What is Stitch Pulse?

Stitch Pulse is a suite of AI-driven software products for South African enterprises. It includes Pulse Analytics, Pulse Fraud and Pulse Onboarding (KYC and KYB), with billing and communications products in development.

What does Pulse Analytics do?

Pulse Analytics comes with an AI agent that briefs each team every morning on what changed in the business and why. Teams can ask questions in plain language, set alerts in a sentence, build custom dashboards, export reports and start permitted actions like refunds, all based on their payments data.

Why does payments data matter so much for AI in payments?

An AI tool is only as useful as the data it can see. Tools built into payment flows can explain what happened, why it happened and how much it cost. General-purpose tools without that access can only work with whatever data they’re given.

Do I need a new integration or data project to use Pulse?

No. Pulse works on the data Stitch already processes, so your full payments history is available from the day you switch it on.

Is my data used to train AI models?

No. Pulse runs inside the Stitch ISO 27001 and PCI DSS Level 1 certified environment, and client data is never used to train third-party AI models.

What should businesses ask when evaluating AI tools for payments?

Ask what data the tool actually sees, whether the team behind it has real payments expertise, whether it can explain why something changed (not just what changed), and how your data is stored and protected.