Once a blockchain cheerleader, Axoni changes its playbook

The year is 2017. The Nintendo Switch has hit shelves worldwide, Brexit negotiations have begun, and distributed ledger technology is at the center of consortiums, innovation labs, and startups across capital markets.

As DLT proponents, pioneers, and enthusiasts made noise across capital markets, New York-based startup Axoni was carving out a niche. As financial firms sought to uncover pain points that might be remedied by the nascent, trust-centric technology, Axoni was looking to solve what it calls the data coordination problem, a post-trade issue in which data shared between counterparties is out of sync.

Across asset classes, the complaint is similar: “I want to see the same data you are seeing.”  With the US’s move to T+1 this past May—precipitating a possible move to T+0—and European countries now racing to shorten their own settlement times, the back office and has once again found itself under the spotlight.

According to data from McKinsey & Company, operations account for 15% to 20% of bank budgets, but Greg Schvey, Axoni’s chief executive and co-founder, tells WatersTechnology that the data coordination problem in reconciliations continues to cost the industry “billions of dollars per year [and] leads to huge operational costs and unknown risks flying around.”

In trade processing, sequence is everything. “Certain things can only happen once other things happen,” Schvey says. After counterparties have agreed to a trade, several events could occur in the post-trade sequence. Amendments can be made to the state of the trade, and cash flows can be generated, which can be either agreed or disagreed upon. Each step works almost as a gate that needs unlocking to progress to the next level, and this is where disputes, disagreements, and confusion can appear.

When Axoni set out in 2017 to solve this problem, it chose distributed ledger technology as the foundation of its platform. Seven years later, DLT has not made the grade originally envisioned by Schvey and his colleagues. Instead, Axoni’s offering has combined the “valuable parts” of DLT—namely, its principles, rather than its technology— with more familiar tech components like software and database technologies.

A lesson in adaptation

“The data issue exists in every corner of capital markets,” Schvey says. Axoni operates in securities lending and various OTC derivatives, and it has found the most value in contract-based assets because they can be outstanding for long stretches. “You might have a swap or a securities financing contract that’s open for months or years at a time, under which there can be many different events, different cash flows, lots of room for disagreement over a long period of time where you have to stay coordinated with another party,” he says.

A popular component of DLT is smart contracts. Schvey says smart contracts can solve the data consistency problem, but the workings of regulated financial markets bring complications. A smart contract is a piece of code that exists within a blockchain message that is then distributed out to participants on a network, effectively creating an application on the blockchain. Once that application is up and running on nodes, participants can send updates and validate them.

“In centralized infrastructures, we obviously have to trust the central party, which we have in this regulated financial market,” he says. “If you have one party doing it, they can do it with maximum efficiency.” In the distributed model of a blockchain, everybody is doing the same thing, and suddenly multiple parties are validating each other, bringing complexity.

The Axoni Data Platform looks to synchronize the data across the parties, cutting out the distributed consensus process because the parties are already known to each other. Additionally, the idea of shipping application code to someone else’s environment over the blockchain is not amenable to the way firms run their tech deployments.

“If you have to ship an update to people, they have to consume that code, and they have to run it through their security scans. They have to go through a deployment process,” Schvey says. The ability to make rapid updates to an application gets “hamstrung very quickly,” he says, because different parties move at different paces to make the same update.

In a simplistic scenario, where one might be doing a currency movement, Schvey says there’s a possibility it could work. With something more sophisticated, where consensus is required on a more complex application, you’re only going to move as fast as the slowest party on the network. The more parties that are added, the less scalable everything becomes.

Schvey acknowledges that the company should have foreseen the issue. “It’s one of those things that as we got it out there, we didn’t really realize until our fourth or fifth party on the network [that] we’re like, ‘Okay, this is getting harder and harder as we go.’”

That was in 2020. The team took a step back and decided they needed to figure out which pieces of the DLT-based platform were slowing them down and which were providing value. By the following year, the new data platform was being developed. It was deployed in 2023.

“Our core platform now still leverages a lot of the same technical concepts, just adapted more toward the things that make sense for large scale enterprises, particularly financial institutions,” he says.

Schvey sees one of the major limitations of DLT technology in its use of what is called a key-value store. Key-value stores are databases that use simple keys for data objects. These stand in contrast to relational databases, which allow for more complexity in the data they store and are more commonly used in traditional finance. “If you have something stored in a key value store, you’re going to have to do some sort of translation into something that’s easier to query, easier to build indexes around, and easier to generally operationalize,” he says.

Axoni also found that when issues surfaced in one user’s infrastructure, they impacted the user’s counterparties, but the “who” and the “how” weren’t easy to diagnose. Schvey says it wasn’t until multiple clients complained, that the affected counterparty could be identified. In one instance, a bank shut down its servers on a weekend, but didn’t spin them up fast enough on Monday morning. “We’d be getting calls from their counterparties like, ‘What’s going on? Why are my trades not going through?’” he says.

In another scenario, a client was running a custom version of Red Hat, an open-source technology provider, and had made some configuration changes. “It was basically not allocating enough memory to our software,” Schvey says. “We were only able to process a quarter of a transaction a second or some ridiculously low number.”

The problem wasn’t identifiable from the outside, and it took getting into the client’s environment to figure out what had gone wrong. While those changes may have been necessary for the client, they impacted everyone with whom they interacted on the network.

The team also wanted to use technology that was familiar to and trusted by large enterprises. With Ethereum blockchains, as an example, a programming language called Solidity is required for the development of smart contracts. “It requires a new set of understanding around what it can and can’t do,” Schvey says. So Axoni leveraged existing databases and event feeds that could better interact with well-known programming languages like Python and C. It made information security easier and removed the burden on clients to learn a new piece of technology.

As an example, Schvey says most of Axoni’s clients use PostgreSQL, a common open-source relational database. The new platform can replicate permitted data from one user’s Postgres to other parties on the network, even if they aren’t using Postgres. MongoDB, another common database, can also receive the data.

The new platform, as Schvey describes it, is a piece of software that sits next to a relational database. It looks at data contained in that database, and each piece of the data is tagged with the identity or identities permissioned to receive it. The software notes the permissions and distributes the data to the appropriate end users. The end user ends up with a replicated database of the data to which they’re entitled.

“That’s a really powerful concept because you can then start to build your infrastructure off of that internally,” he says. “If you’re a bank, a hedge fund, an asset manager, it’s like you are directly accessing the database at the clearinghouse, and you’re able to build all your stuff on that with the same level of accuracy and the same exact datacenter.”

There’s also a real-time element to the technology Axoni is bringing to its clients. “Most of what our clients are trying to replace are things like end-of-day batch reports,” Schvey says. In the conversations around moving to T+1 and eventually same-day settlement, vendors and market participants have expressed a need to shift away from end-of-day batch to intra-day or real time. Schvey says Axoni’s platform can keep clients up to speed on events throughout the day, reducing the need to calculate risk and net position at the end of the day.

“Those are the types of guarantees that you can get on a blockchain network. We just dropped a lot of stuff, the consensus pieces and the complicated programming language,” he says. “We kept the really valuable parts and just kind of got rid of the complications.”

Tried, true, and fried

Axoni is not the only vendor to make changes to its tech after losing bets on blockchain. UTrade Solutions, a trading technology provider based in India, sought to capitalize on blockchain’s promises of immutable recordkeeping and enhanced security by launching UClear, a real-time clearing and settlement solution, and KYChain, a know-your-customer platform, both based on distributed-ledger technology, in 2016. The solutions were on the market for less than a year, said Kunal Nandwani, CEO and co-founder of UTrade, who told WatersTechnology in 2022 that he was somewhat misled by early blockchain zeal.

“When I start building KYChain, I totally believed in the world narrative and the people saying it would work, but as I built it, I realized it doesn’t work,” he said. Nandwani dug his heels in further. “You can’t have scale, speed and efficiency—you can have one of the three but not all three. With blockchain, by definition, decentralization brings slowness. Scale is not easy, and because of the speed and scale problems, blockchain will actually not solve any problem anywhere.”

Others have had more public and more dramatic pivots. Perhaps the most famous project using DLT was the upgrade of the Australian Securities Exchange’s cash equities clearing and settlement system, Chess. After years of development work costing hundreds of millions of dollars, ASX abandoned the project in 2022—or “paused” it, in corporate speak. ASX later commissioned Tata Consulting Services to run the replacement, with DLT not a part of the new core system.

The project’s scrapping shook the world of DLT. One blockchain startup founder told WatersTechnology that it “cast a very skeptical light on blockchain projects as being a waste of money and time. … I think it brought a lot of institutional skepticism on DLT as being the next shiny thing that never really worked out.”

“Every technology goes through a similar arc in that sense. I mean, pick a technology, I can take you through almost the exact same story,” Schvey says today. “As people understand the potential of these types of things, it tends to create a lot of interest, a lot of people trying to experiment, getting past that first experimental hump into the big things.”

Virginie O’Shea, founder and CEO of Firebrand Research, argues that all new technologies must contend with limited attention spans. “If they don’t work immediately, or they don’t work within a few years, people move on,” she says. Dwindling mentions of the technology among some of its original champions demonstrates this. “They do pilots and proofs-of-concept just to see if something works, so that’s why there is a high level of experimentation.” If the project doesn’t move beyond that stage, momentum can be hard to maintain.

For Axoni, it took until parties were already live on the network for operational and scalability issues to rear their heads. In Schvey’s eyes, those events—missteps or milestones—were essential to create the company’s offering today. “The same thing happened to the internet… There’s a reason we all dropped Flash and older versions of HTML. And we are just in a different world now, [but] that doesn’t mean that the things that happened in the 90s weren’t relevant to help everybody understand what could work—what will work.”

The only constant is change, and by that measure, Axoni is still re-shaping itself. The number of pure DLT startups has shrunk over the last decade, but many of them have become something else. Schvey, for his part, now puts his company’s purpose in the much broader category of data integration.

“There’s huge, huge amounts of very well-established competitors, and we’re actually taking a newer technical approach to solving some of those problems that they haven’t yet solved,” he says.

He laughs, maybe because he’s been here before.

“Out of the frying pan, into the fryer.”

Additional reporting by Bernard Goyder