Is AI vs humans the right question?
06 January 2026
Martin Walker, product manager at FIS, reflects on how the industry should look to use AI to supplement rather replace people, in improving efficiency, quality, and operational risk in securities finance
Image: stock.adobe.com/AntonKhrupinArt
One of the major claims made for artificial intelligence, in particular generative AI, is that it can replace humans in many routine tasks. Many commentators question whether the entry-level jobs in a lot of business areas will simply disappear. Looking across the business processes involved in stock borrow loan (SBL), a more fundamental question arises: do you want to improve the way problems in the trade lifecycle are resolved or avoid them happening in the first place?
Successful firms in sectors such as manufacturing, have shown continual improvements in productivity and quality by looking for the root cause of problems and fixing them. For decades banks and other financial sector intermediaries have invested billions in technology, process re-engineering, organisational change, and offshoring. Huge improvements have resulted in some areas but at an industry level the return on investment has been poor. In the United States, the Bureau of Labor Statistics data on banking productivity shows an improvement in recent years, but the overall level has yet to return to levels of 15 years ago. The data from the Office of National Statistics in the UK shows an even worse picture.
Figure 1: US banking productivity (2017 = 100)
Source: Bureau of Labor Statistics
Some of that reduction in productivity clearly results from the burden of increased regulation, but it can also be argued that the costs of complying with regulation were magnified by failing to deal with pre-existing problems in systems and business processes. In much of capital markets but particularly in SBL there has often been a focus on improving the process for identifying and resolving issues, rather than avoiding them. Some may point to issues in getting budget or creating business cases for more fundamental change but there is an underlying historic reason.
Made to fail
Modern infrastructure in capital markets is based on the concept of straight-through processing (STP). Trades (or other transactions) are captured or executed in one system and the relevant data is fed through to one or more other systems to generate confirmations, settlements, accounting entries, risk adjustments, and regulatory data. A completely straight-through process does not require human intervention. In practice allowance is made for exceptions i.e. deviations from automated processing where a human being can intervene to either resolve a problem or fill a functional gap.
Figure 2: The many places for breaks
This is a pragmatic way of designing systems and business processes, it recognises that the cost of designing systems to deal with every possible scenario can be more expensive than a system that relies on humans to resolve the most complex or difficult issues. Unfortunately, allowing exception-based processing opens the door to customised manual processes, hides flaws, and creates key person dependencies. It can also make it easy to lose sight of the real costs, especially when those exception management processes create the need for tactical and third-party tools.
Firms with better infrastructure and processes have higher STP rates. However, many processes are bilateral so for example, one party鈥檚 processing cash marks can ultimately be as error free as that of the counterparty on the trades. How though, can we address the specific root causes of some of the costs in securities finance?
Onboarding
The trading process really starts with client onboarding 鈥� a process that generally remains complex and error prone. Fortunately, there are multiple industry initiatives around onboarding but firms should reflect on the key drivers of failure:
鈥� Non-standard agreements and inconsistently structured collateral schedules.
鈥⒙燗greements that only exist as text and have to be converted into data for ingestion into system.
鈥⒙燙ommunication of standard settlement instructions in documents rather than as data.
In addition to addressing the above root causes there are more radical approaches to reducing the costs and risks of onboarding. Clearing (with the right economic model) can potentially reduce the need for a web of bilateral relationships. Also there are potential models for recording the agreement of preferences between counterparties for processing trades in a consistent data driven model.
Trade execution and trade capture
If each party manually enters trades into their own trading system it invites errors and mismatches. Some asset classes have almost entirely removed post-trades exceptions through electronic trading or affirmation. These ensure both parties start with a consistent view of the trade. The clearing of trades, with rejection by the CCP of any trades that do not match, is also a proven method to eliminate most mismatches in trades throughout the lifecycle.
The cost equation for using electronic platforms should include:
鈥� Savings through error reduction,
鈥⒙燭he minimisation of platform fees for the simple cases where data such as availability is directed to known counterparties,
鈥⒙燬avings in front office costs where sophisticated logic in matching/directing orders replaces some of the efforts of traders.
Trade lifecycle management
A unique feature of SBL is the volume of lifecycle events. Every time a lifecycle event is entered into a trading system there is risk of a break that could add costs, create operational risk, and drive wrong trading decisions. The contract compare process adds much value but there are obvious costs, both fees and the costs of analysing breaks. The lesson from other asset classes is that consistent processing of trades between parties reduces breaks.
There are three main models for doing this:
鈥⒙燙learing all trades via a CCP that manages the lifecycle and provides the golden source of trade data.
鈥⒙燭he Common Domain Model (CDM), an industry initiative that aims to bring in consistent modelling of trades and events. Though progress on adoption since it started in the derivatives world in 2017 has been slow .
鈥⒙燯se of a centralised trade and lifecycle management system such as the Loanet Full Service. Where both parties use Loanet Full Service as for trade processing it can reduce the number of breaks by 60-70 per cent.
Regulatory reporting
The regulation that probably causes more pain than any other is the 网爆吃瓜 Financing Transactions Regulation (SFTR). Processes work, in as much as data is processed and sent to trade repositories, but the costs of data enrichment, matching, and break resolution can be very high. Even then data is often unusable by regulators. The complexity of regulation drives much of the cost but there is scope to deal with root cause issues such as agreeing a Unique Trade Identifier (UTI) for trades at execution (some platforms provide for free). Not to mention consistent lifecycle management as discussed above.
Settlement
Improving any of the areas above can reduce errors in settlement processes; as would making greater use of regulatory mandated information such as UTIs and Legal Entity Identifiers (LEI). Another catalyst for solving root cause problems is making the data about fails more granular. The common practice of combining settlement of financing and cash trades can obscure securities lending specific problems. This does not mean securities finance settlement should have segregated processes and systems but it should be possible to see the specific data for SBL or repo. For instance, in the repo world, there is very limited industry-level data about fails on repo start or end legs, because it is intermingled with normal bond settlements, making it hard to see the root cause of problems.
AI versus people
Many of the outputs of generative AI tools such as ChatGPT, Gemini, and Claude have been so impressive that the technology has rapidly captured the imagination of millions. Microsoft provides a very clear and hype free explanation of generative AI 鈥� it 鈥渁nalyses large amounts of data and generates new content, including text, images, and code, that mirrors human expression鈥�. The downside to Gen AI is the statistical models that can produce convincing text that is not real (鈥榟allucinations鈥�) or based on incorrect data. The concept of STP aims to remove sources of error, not add them. A Gen AI process fixing a trade problem by making up a dividend rate would cause catastrophic problems. Gen AI needs to be carefully applied in SBL, starting with an understanding, by a human, of what type of problem needs to be solved and whether or not Gen AI can help. On the other hand, other forms of AI, such as deep learning, could help provide powerful and objective insights into the root cause of problems. They can also help to train systems how to fix recurring problems.
Conclusion
The forms of AI currently gaining the most attention are not necessarily those most suitable for dealing with problems in front-to-back SBL processes. Other forms of AI, such as machine learning, when used by the right people with the right data still have the potential to transform the industry. The future for SBL is to be more automated and efficient, but it's still a future with people in it.
Note 鈥� no bots or agents were harmed, or even used, in writing this article.
Successful firms in sectors such as manufacturing, have shown continual improvements in productivity and quality by looking for the root cause of problems and fixing them. For decades banks and other financial sector intermediaries have invested billions in technology, process re-engineering, organisational change, and offshoring. Huge improvements have resulted in some areas but at an industry level the return on investment has been poor. In the United States, the Bureau of Labor Statistics data on banking productivity shows an improvement in recent years, but the overall level has yet to return to levels of 15 years ago. The data from the Office of National Statistics in the UK shows an even worse picture.
Figure 1: US banking productivity (2017 = 100)
Source: Bureau of Labor Statistics
Some of that reduction in productivity clearly results from the burden of increased regulation, but it can also be argued that the costs of complying with regulation were magnified by failing to deal with pre-existing problems in systems and business processes. In much of capital markets but particularly in SBL there has often been a focus on improving the process for identifying and resolving issues, rather than avoiding them. Some may point to issues in getting budget or creating business cases for more fundamental change but there is an underlying historic reason.
Made to fail
Modern infrastructure in capital markets is based on the concept of straight-through processing (STP). Trades (or other transactions) are captured or executed in one system and the relevant data is fed through to one or more other systems to generate confirmations, settlements, accounting entries, risk adjustments, and regulatory data. A completely straight-through process does not require human intervention. In practice allowance is made for exceptions i.e. deviations from automated processing where a human being can intervene to either resolve a problem or fill a functional gap.
Figure 2: The many places for breaks
This is a pragmatic way of designing systems and business processes, it recognises that the cost of designing systems to deal with every possible scenario can be more expensive than a system that relies on humans to resolve the most complex or difficult issues. Unfortunately, allowing exception-based processing opens the door to customised manual processes, hides flaws, and creates key person dependencies. It can also make it easy to lose sight of the real costs, especially when those exception management processes create the need for tactical and third-party tools.
Firms with better infrastructure and processes have higher STP rates. However, many processes are bilateral so for example, one party鈥檚 processing cash marks can ultimately be as error free as that of the counterparty on the trades. How though, can we address the specific root causes of some of the costs in securities finance?
Onboarding
The trading process really starts with client onboarding 鈥� a process that generally remains complex and error prone. Fortunately, there are multiple industry initiatives around onboarding but firms should reflect on the key drivers of failure:
鈥� Non-standard agreements and inconsistently structured collateral schedules.
鈥⒙燗greements that only exist as text and have to be converted into data for ingestion into system.
鈥⒙燙ommunication of standard settlement instructions in documents rather than as data.
In addition to addressing the above root causes there are more radical approaches to reducing the costs and risks of onboarding. Clearing (with the right economic model) can potentially reduce the need for a web of bilateral relationships. Also there are potential models for recording the agreement of preferences between counterparties for processing trades in a consistent data driven model.
Trade execution and trade capture
If each party manually enters trades into their own trading system it invites errors and mismatches. Some asset classes have almost entirely removed post-trades exceptions through electronic trading or affirmation. These ensure both parties start with a consistent view of the trade. The clearing of trades, with rejection by the CCP of any trades that do not match, is also a proven method to eliminate most mismatches in trades throughout the lifecycle.
The cost equation for using electronic platforms should include:
鈥� Savings through error reduction,
鈥⒙燭he minimisation of platform fees for the simple cases where data such as availability is directed to known counterparties,
鈥⒙燬avings in front office costs where sophisticated logic in matching/directing orders replaces some of the efforts of traders.
Trade lifecycle management
A unique feature of SBL is the volume of lifecycle events. Every time a lifecycle event is entered into a trading system there is risk of a break that could add costs, create operational risk, and drive wrong trading decisions. The contract compare process adds much value but there are obvious costs, both fees and the costs of analysing breaks. The lesson from other asset classes is that consistent processing of trades between parties reduces breaks.
There are three main models for doing this:
鈥⒙燙learing all trades via a CCP that manages the lifecycle and provides the golden source of trade data.
鈥⒙燭he Common Domain Model (CDM), an industry initiative that aims to bring in consistent modelling of trades and events. Though progress on adoption since it started in the derivatives world in 2017 has been slow .
鈥⒙燯se of a centralised trade and lifecycle management system such as the Loanet Full Service. Where both parties use Loanet Full Service as for trade processing it can reduce the number of breaks by 60-70 per cent.
Regulatory reporting
The regulation that probably causes more pain than any other is the 网爆吃瓜 Financing Transactions Regulation (SFTR). Processes work, in as much as data is processed and sent to trade repositories, but the costs of data enrichment, matching, and break resolution can be very high. Even then data is often unusable by regulators. The complexity of regulation drives much of the cost but there is scope to deal with root cause issues such as agreeing a Unique Trade Identifier (UTI) for trades at execution (some platforms provide for free). Not to mention consistent lifecycle management as discussed above.
Settlement
Improving any of the areas above can reduce errors in settlement processes; as would making greater use of regulatory mandated information such as UTIs and Legal Entity Identifiers (LEI). Another catalyst for solving root cause problems is making the data about fails more granular. The common practice of combining settlement of financing and cash trades can obscure securities lending specific problems. This does not mean securities finance settlement should have segregated processes and systems but it should be possible to see the specific data for SBL or repo. For instance, in the repo world, there is very limited industry-level data about fails on repo start or end legs, because it is intermingled with normal bond settlements, making it hard to see the root cause of problems.
AI versus people
Many of the outputs of generative AI tools such as ChatGPT, Gemini, and Claude have been so impressive that the technology has rapidly captured the imagination of millions. Microsoft provides a very clear and hype free explanation of generative AI 鈥� it 鈥渁nalyses large amounts of data and generates new content, including text, images, and code, that mirrors human expression鈥�. The downside to Gen AI is the statistical models that can produce convincing text that is not real (鈥榟allucinations鈥�) or based on incorrect data. The concept of STP aims to remove sources of error, not add them. A Gen AI process fixing a trade problem by making up a dividend rate would cause catastrophic problems. Gen AI needs to be carefully applied in SBL, starting with an understanding, by a human, of what type of problem needs to be solved and whether or not Gen AI can help. On the other hand, other forms of AI, such as deep learning, could help provide powerful and objective insights into the root cause of problems. They can also help to train systems how to fix recurring problems.
Conclusion
The forms of AI currently gaining the most attention are not necessarily those most suitable for dealing with problems in front-to-back SBL processes. Other forms of AI, such as machine learning, when used by the right people with the right data still have the potential to transform the industry. The future for SBL is to be more automated and efficient, but it's still a future with people in it.
Note 鈥� no bots or agents were harmed, or even used, in writing this article.
NO FEE, NO RISK
100% ON RETURNS If you invest in only one securities finance news source this year, make sure it is your free subscription to 网爆吃瓜 Finance Times
100% ON RETURNS If you invest in only one securities finance news source this year, make sure it is your free subscription to 网爆吃瓜 Finance Times
