Ping Payments (payment orchestration platform) is growing 40 % year-on-year

Building fast under heavy regulation

Ping Payments is a Swedish PSD2-licensed payment orchestration platform that lets tech teams launch fast, compliant bank-to-bank payments with a single endpoint. Developers can collect, hold, split and pay out funds—complete with built-in Know Your Customer and Anti Money-Laundry (KYC/AML), escrow-style client accounts and ledger-grade reconciliation—so marketplaces and public-sector portals go live with payments quickly and securely while Ping quietly handles the registrations and bookkeeping.

Founder-CEO Petter Sehlin has a simple scoreboard: reach 300 000 transactions a month before the end of 2025. He also has grand plans for Ping as it grows through its current market, He believes Ping can claim the “office-of-the-CFO” leadership position. But every percentage point of volume brings more metadata, stricter AML checks, and thicker KYC files.

BEAM Elixir/Erlang, generative AI and developers

Ping’s payments processing runs on the BEAM (Elixir/Erlang) platform, which is highly suitable for streaming massive volumes of transactions. The developer team at Ping are clear that any AI layer must support their tech stack choices, not weigh it down with bulky frameworks.

The team has built OCR powered by Machine Learning, which extracts KYC data from scanned documents and feeds it into rules engines. Their main issue has been that natural language creates new form layouts, and regulatory tweaks risk breaking the flow.

Ping focus on two initial AI features, while one is already done, work now begins on the next:

• KYC automation. Done and in production environment
• Chatbot guiding customers with information completion. Starting to build now.

The ambition of the AI features, in the words of the builder:

Q: What was the goal of the KYC AI feature?
A: Three FTEs have been very involved in doing renewal paperwork for 6 000 customers; ML has alleviated this but we’re seeing that our AI powered KYC pipeline makes this much more efficient. The goal in terms of % has never been discussed but let’s just say; “improve it a lot is the goal”.

Q: That's interesting, so this is purely a backend task?
A: Well, yes it is but we are now also building a chatbot that will walk each customer through gaps in their Oneflow agreements. This should automate a lot of the KYC information gathering and we hope we can build something that creates an excellent customer experience.

Petter and the developers aim to empower developers to utilize AI within their existing workflows, leveraging their domain expertise. Petter's vision is to keep AI capabilities integrated with the development teams, rather than centralizing AI in a separate unit that dictates frameworks.

The KYC AI feature in more detail

Ping put together a Elixir microservice with Opper’s REST API to create a document data automation pipeline. A vision-optimised model classifies incoming PDFs—contract, bank statement, board resolution etc. The task then delegates the file to the best task-specific LLM. For example, agreements pass through a metadata extractor that pulls dates, signatories, authority checks and so forth. Each stage returns a typed JSON object into Ping’s KYC tech stack.

In this feature, Ping currently uses three EU-hosted models - specifically chosen for vision, low-latency and instruction-following. They all run through Opper’s calls, evals, and traces, so quality never drifts.

The first custom Evals metric, specifically, is “Success”. Ping’s engineers use this to evaluate that they have extracted all information as well as capturing any drift on facts. This is highly useful as a first checkpoint in the KYC automation. It enables them to capture failures that go to manual processing. Tobias, one of the developers, tells us that he is quite sure that they will have further eval metrics in the KYC pipeline going forward as they refine the flows further.

Proof that shows up on the ledger

Agreement analysis—once something that became an irregular batch job—now runs continuously; transactions have full audit through traces, so compliance officers can sign off without reopening the PDF. Finance sees real-time token spend next to interchange fees, turning AI from a black-box cost to a line they can forecast.

Ping treats language models like any other microservice in its Erlang supervision tree—restartable, measurable, and replaceable.

The payoff is impressive: 87 % of documents are now automatically processed without human touch, and Ping is perfectly happy leaving the last 13 % to manual review because the ROI is already stellar.

What’s next on the roadmap

Ping does not have small ambitions when it comes to generative AI. From our dialogue, it becomes clear that the entire company is well aware just how much they can automate and improve. And it makes sense, Ping is going after becoming a micro-optimization engine for massive volumes of payments which should allow the CFO to calculate and forecast cost, margin and more for every single transaction - and then aggregate this into a very powerful financial orchestration.

When we asked Petter what the team is considering focusing on for the next 6-12 months on the AI roadmap, we got an impressive answer. Petter’s sights are on transaction monitoring: an AI advisor that reviews every statically generated alert, downgrading false positives and escalating real fraud.

The long game is bigger than payments; it’s a full stack CFO console where budget, forecasts, cash-flow insights, and compliance checks emerge from the same conversational layer that already renews KYC. As he puts it, “Efficiency isn’t the moonshot—it’s the launch pad. Once the basic parts run themselves, we can optimize for 300 % efficiency and still have time to skate where the puck is going, not where it is.”