Straight Through Processing, AI and the Process of Innovation

A discussion with James Karat about his invention of Straight Through Processing, the process of innovation, and the uses and abuses of AI to improve capital markets

Sometimes in the world of Fintech, it seems like there is an inverse relationship between noise generated about a particular innovation and its real-world impact. For over a decade we have been hearing about the transformational impact of blockchain aka DLT aka “Tokenisation”. The world has also been through many periods of excitement about artificial intelligence (AI) followed by prolonged AI winters as reality fails to meet expectations. None of this means those technologies are without merit but the truly transformational ideas can take hold amazingly quickly and become so ubiquitous, people take them for granted and forget there was a time they did not exist.

One of those ubiquitous ideas in capital markets technology and business processes is “Straight Through Processing” (STP). An idea that shapes the technology of virtually every firm in capital markets and many beyond. Though it might be harder for the younger generations to believe, computers have been around for a long time. Even in the 1980s they were in use in virtually every broker and bank. The major difference was they generally did not talk to each other. Also many of the core business processes, including that around executing, confirming and settling equity trades relied on people and paper to fill the gaps.

When James Karat, started in the City of London as a 16-year-old “runner” at a brokerage house the trading process was as follows,

· Salesperson took an order from a customer via telephone and wrote a paper deal ticket

· The deal ticket was delivered to a trader who would execute the trade

· Once the transaction was executed the trader would write the trade economics on the deal ticket and hand it to the “runner”.

· The runner input the ticket into a middle office system, and a contract note was printed out

· The contract note was put into an envelope and posted to the customer

· At some point during the ten business days (T+10) before final settlement of the trade the customer would check the contract note and contact the brokers if there was an error

It was a process open to all manner of human errors, not to mention fraud and the human (and postal) bottlenecks made speeding up the process close to impossible. Modern settlement cycles of T+1 or T+2 were then inconceivable.

Karat’s idea was to simplify and speed up the process of confirming the details of the trade through improved connectivity between systems and the parties involved. After the middle office entered the trade details, the confirmation would be electronically sent the client, who could confirm the details (or reject them) almost immediately. Once details were confirmed the exchange of funds and securities could safely take place. In large part because of this idea it was possible to shrink the settlement cycle for UK equities from T+10 to T+5. It also reduced risk of settlement failure and fraud.

Pitching the Idea

Karat’s opportunity to pitch his idea to the chairman of the London Stock Exchange, came about because he had been asked to talk to him about reducing the settlement cycle from T+10 to T+5. Today getting in front of a senior executive to talk about an innovative idea would likely depend on seniority, credentials in a management consulting or ability to talk about innovation. Karat’s manager sent him to talk to the chairman because he thought he was knowledgeable about settlements and would provide useful information. Karat himself came up with the idea of STP, not because he was trying to be “innovative” but because he had a problem to solve. The people that typically got blamed for the losses and general confusion that frequently arose in trading then were runners like Karat. Even though errors in booking trades were usually the result of miscommunication between sales and traders.

As I was the junior every time there was an incorrect booking I got blamed. I always had to buy drinks in the pub to apologise and I was spending more money on drinks than I was earning. I knew I had to come up with an idea which would stop me being the scapegoat.

The push to reduce the settlement cycle made the potential benefits of Karat’s idea even more attractive and he was given a team to implement his idea.

Unbelievably after a few more meetings the chairman told me that he wanted me to build the system that would allow Straight Through Processing and offered me an IT team who would build my design. The concept was quite simple. I needed to be able to send a message to someone asking them to confirm the details of a transaction were correct and then once I had that confirmation, funds needed to be moved from one account to another.

Today the concept of linking together parties to a trade using interconnected systems has spread across most parts of the financial sector. In most cash equities, foreign exchange and futures trading, the process from trade execution through to settlement has no human involvement at all apart from the trader inputting the trades. For a large proportion of trades, where a computer programme makes the trading decisions and books the trades (Algo Trading), there is no human involvement at all.

Achieving 100% STP (i.e. no trades require manual intervention) is easier said than done and so I asked Karat about the persistence of errors in many areas of capital markets. He explained STP is only as good as the information provided but the need for ever higher STP rates and quicker processing of trade data is now being driven by regulatory pressures. Under the terms of the Markets in Financial Instruments Directive II (MiFID II), equity trades need to be reported withing 1 minute of execution and those reports contain up to 65 fields per transaction. It was to improve the STP rate for trade reporting that Karat has been using AI to do root cause analysis of errors. The use of AI, more specifically machine learning, has been used to successfully identify the root causes of breaks. Finding and resolving root causes of issues is a far more efficient way to improve STP rates that simply trying to speed up the rate at which breaks are fixed. Trying to apply Generative AI (gen AI) to fix problems in operational processes is considerably more problematic. Since any process based on Gen AI is based on statistical models, it can always introduce errors into the process. A single piece of plausible but essentially “made up” data from gen AI can potentially cause huge losses if money or securities are sent to the wrong places.

Innovation in the modern age

I asked Karat what he thought about innovation today in Fintech. He said it was simply done differently today rather than better. “Who would listen to a 17-year-old in our industry or even a 20-year-old?” He also thought the cost of making technological changes in the modern industry was holding back innovation.

More generally, the idea that people in any role can come up with solutions to real problem in areas they understand, are listened to by the right person, and are given the resources to build their solution almost seems quaint. Yet historically it is the way so much invention has happened. For over a decade innovation has been progressively more focused in innovation labs and start-ups that often do not have sufficient domain expertise and spend their time looking for use cases for miracle technologies rather than trying to come up with solutions to real problems.