The AI realists strike back
A few years ago, Blockchain evangelists told us that the technology would dis-intermediate almost everything in the post-trade value chain, from custody right through to clearing and settlement.
While Blockchain has scored some successes, most notably in the repo market, its overall impact on post-trade has been pretty limited. Barring a few proofs of concepts, most of the Custodians and Financial Market Infrastructures (FMIs) I speak to have put their Blockchain plans on hold.
So, is the Securities Services industry at risk of falling into the same trap with Generative AI as it did with Blockchain?
I spoke on a panel during TNF, where experts extolled the virtues of Generative AI.
Within Securities Services, people have repeatedly told me that AI can do all sorts of things, ranging from reducing the number of settlement fails, supporting cash optimisation, enhancing client reporting and communications, right through to making it easier for Network Managers to autonomously compare the responses from different due diligence questionnaires (DDQs).
While Generative AI will certainly have productivity benefits, i.e. summarising documents succinctly or helping people draft RFPs or PowerPoint presentations, I believe our industry is looking at the technology through rose-tinted glasses.
Arguing the case against AI is not always easy, especially in a month that saw Open AI, the organisation behind ChatGPT, valued at $157 billion, making it one of the most valuable privately held companies in the world. [1]
However, I explained to the audience at TNF that the Securities Services industry needs to think very carefully before adopting AI en masse.
Be diligent about AI’s risks
Firstly, Custodians and FMIs need to be on top of the risks, which AI could potentially pose.
Post-trade providers are guardians to vast amounts of proprietary and highly sensitive client information, yet a number of firms are using external Generative AI or Large Language Model (LLM) solutions to comb through these very same data pools.
Should client information be leaked or compromised because of badly managed AI, then firms risk falling foul of incoming EU regulations, namely the Digital Operational Resilience Act (DORA).
At MYRIAD Group Technologies, we will not roll out any AI solutions – proprietary or otherwise – unless they have been tested internally to the nth degree, not least because we are an ISO 27001 certified company.
If we were to integrate AI into our business operations, then we would most likely need to be re-certified to ensure our AI policies and procedures meet ISO 27001’s strict criteria.
Any output from AI also needs to be thoroughly scrutinised before it is used by businesses or clients alike. If AI tools consume bad data, then the analytics produced will be equally flawed. Additionally, firms need to monitor that their AI systems are not hallucinating, namely presenting falsehoods as facts, a common problem with the technology.
Will the status quo suffice?
There are other reasons to be sceptical about AI.
One of the (many) reasons Blockchain struggled to scale was because existing software and technology could do many of the things Blockchain was promising, if not better, and cheaper.
The same rings true for AI. For instance, on benchmarking DDQs, Network Managers do not need to leverage AI, but can simply rely on standard delta management solutions instead.
AI enthusiasts routinely argue the technology will lead to efficiencies and significant cost savings, but I am not quite so sure.
Companies are expected to spend $1 trillion on capex on AI, yet an MIT Professor told a Goldman Sachs report that only ¼ of AI exposed tasks will be cost-effective to automate within the next 10 years, suggesting AI will impact less than 5% of all tasks. [2]
Compounding matters further is that the same MIT professor added that AI advances are not going to happen nearly as quickly as many people think. [3]
And we have not even mentioned the massive energy costs which AI adoption will inflict on businesses due to the technology’s enormous power consumption, a yardstick which needs to be factored into any Return on Investment calculations.
With Securities Services’ overheads rising, providers need to ask themselves whether investment into AI solutions should be a priority, or if they should focus more on improving their existing product suites.
After an excellent TNF Americas, MGTL look forward to seeing you at TNFs in Qatar and Singapore in November, together with SIBOS in Beijing.
[1] Reuters – October 3, 2024 – Open AI closes $6.6 billion funding haul with investment from Microsoft and Nvidia
[2] Goldman Sachs – June 25, 2024 – Gen AI – Too much Spend? Too Little Benefit
[3] Goldman Sachs – June 25, 2024 – Gen AI – Too much Spend? Too Little Benefit
