Becoming AI-first: What it actually means in securities finance

Ben Challice, CEO of Pirum, looks at how firms can move beyond superficial adoption to embed artificial intelligence across workflows, data infrastructure, and culture to deliver measurable outcomes in securities finance

The noise problem

Every major conference in our industry now features at least one session on artificial intelligence. Every vendor deck has been updated to include it. Boards are asking for AI strategies. Meanwhile, the evidence on actual production deployment tells a rather different story.

Deloitte's 2026 survey of more than 3,000 senior leaders found that only 34 per cent of organisations are genuinely reimagining their business with AI — the remaining two-thirds are either redesigning isolated processes or using it at surface level with little change to how work gets done. Gartner found that despite 60 per cent reporting some forms of adoption among finance functions, 91 per cent of respondents reported only low or moderate impact.

This is not a story about technology failing. It is a story about the gap between positioning and execution — and most importantly for today's securities finance practitioners, what is required to close it.

At Pirum, we are not approaching AI as an experiment. It is a fundamental and complementary shift that raises the bar for how we build, support, and operate. AI materially changes what clients can achieve, how quickly firms can roll out new capabilities, it reimagines workflows and means every interaction across the customer lifecycle starts delivering more value than it did before.

In what follows, I want to move past the headlines and into the operational reality. What does the evidence actually show about where AI is delivering returns in securities finance? What does it require of the firms — and the technology partners — that want to benefit from it? And what does it mean to build an organisation that is genuinely AI-first, rather than one that has simply added AI to the agenda?

This piece is written from our perspective, but the stakes are the same for every participant in securities finance.

Rethinking value for clients

The most important conversation we can have with a client about AI does not start with models or platforms. It starts with a question: is your securities finance data standardised, complete, and flowing in real time across your operations?

The use cases generating real, measurable returns in trading, post-trade, and collateral management today share a consistent pattern. They are either high-frequency, high-volume tasks that have disproportionately consumed human time — exception triage, break identification, collateral substitution matching — or they are prediction problems where historical patterns in structured data allow a system to anticipate a costly outcome before it happens or to identify and seize an opportunity with greater speed and conviction.

Settlement fail prediction in a T+1 environment is the clearest example: models trained on clean, real-time trade data can flag likely fails hours before the settlement window closes, enabling resolution that would otherwise require emergency manual intervention at best, or settlement fails, cash penalties, and overdraft charges at worst.

The evidence that forward-thinking firms have already grasped the problem and are busy building the solution is accumulating rapidly. Citi's most recent securities services industry survey, drawing on 537 respondents, found 86 per cent of firms are already piloting generative AI in post-trade operations. BNY has 125 AI solutions in production through its Eliza platform, including 'digital employees' (i.e. AI agents with defined roles and governance controls) handling payment instruction validation autonomously across its global operations. State Street's Alpha platform delivered a 25-times productivity improvement in post-trade data exception processing — not by deploying a sophisticated model, but by building a system that could distinguish meaningful exceptions from noise in a dataset that previously flagged more than 31,000 items, of which only 250 were genuine problems.

None of these results came from AI deployed on top of fragmented or batch-based infrastructure. They came from organisations which had already invested in the underlying plumbing. A Riverbed survey from February 2026, covering 1,200 financial services decision-makers, found 92 per cent agree that improving data quality is critical to AI success — and yet only 43 per cent are fully confident in the accuracy and completeness of their organisation's data. That gap is where most AI projects stall — and why only 12 per cent of AI initiatives in these firms have achieved full enterprise-wide deployment.

"We are already working on evaluating how our unique data, combined with insightful intelligence layers, will help clients uncover completely new insights about their business"

For firms with complete lifecycle connectivity already in place — real-time, standardised, across pre-trade and post-trade, enterprise-wide — the question shifts from 'can we do this?' to 'what do we build first?'. That is a genuinely different position, and one that every firm must reach if they want to keep up with their competitors in an increasingly AI-native marketplace.

And that is just the beginning: we are already working on evaluating how our unique data, from processing globally US$6.5 trillion trades daily, combined with insightful intelligence layers, will help clients uncover completely new insights about their business.

Time to value: The gap between pilot and production

At Pirum, we have embedded this across our own organisation — accelerating product development cycles through agentic AI engineering, from capability to impact, and redesigning core workflows in how we onboard, support and service clients. Our clients have shared their AI ambitions with us openly. We are on this journey alongside them and hold ourselves to the same ambitious and exacting standards.

Speed of deployment matters more in AI than it has in almost any previous technology cycle. Historically, the gap between a pilot that impresses a technology team and a tool that changes how a trader or operations analyst does their job tends to be wide. Usually, the gap between deployment and demonstrated commercial impact is wider still.

The shift to agentic AI is already reaching post-trade workflows. Forward-looking firms are deploying 24/7, always-on, real-time assistants that can reason through a multi-step problem and take action without human initiation. Wells Fargo, for example, has AI agents within its corporate and investment bank to answer, triage, and resolve complex foreign exchange post-trade inquiries. These are tasks that previously required manual navigation across multiple internal systems and data sources. IDC research published in March 2026 found that 80 per cent of capital markets firms now cite building AI agents as their top IT spending priority, with those that have adopted them reporting 2.3-times return on investment (ROI) within 13 months.

The implication for our product development is direct: capabilities need to reach users inside the workflows they already operate, not behind a separate interface that demands a behavioural change. A trader managing a live book does not need a new dashboard, and neither does an operations team managing a recall. They need the insight — whether that is a pricing anomaly against the market, a predicted deadline conflict, a counterparty re-rate signal, an auto-generated recall instruction, or a corporate action election recommendation — surfaced within the environment where they are already making decisions.

AI is central to how we are shortening every stage of the interval that we measure as time to value: from a faster, more intuitive onboarding experience for clients going live, to shorter product development cycles that get new capabilities into clients' hands sooner, to more responsive and better-informed client support. That metric is more useful than feature counts or go-live dates, and it focuses on the right engineering questions.

Client interactions in 2026: Deeply insightful and impactful

The second dimension of an AI-first posture concerns client engagement: how technology changes the quality of the conversations and touchpoints that define a relationship. HSBC AI Markets offers a useful reference point. The platform uses natural language processing (NLP) to give institutional clients real-time access to HSBC's research, pricing, and analytics across asset classes at the point of decision. The standard it sets — turning data into something immediately actionable for the person receiving it — applies equally to how a post-trade platform should engage with its clients.

In practical terms, this means moving from reporting to anticipation. Receiving a reconciliation exception report after the fact informs a client. Receiving a flag 12 hours before settlement, with a probable cause identified from the pattern of the mismatch and a resolution pathway already pre-populated, serves one. The technology required to deliver the latter is available today. The constraint is usually not the model; it is the timeliness and cleanliness of the underlying data, and how thoughtfully the user experience is built around it.

For clients, this should also universally improve their interactions with vendors. Support teams can now arrive at every interaction better informed — aware of a broader aperture of a client's operational patterns, recent activity, and potential pressure points before the conversation begins. Cases can be resolved faster and with less effort on the client's side, but with the same high bar for service and quality of interaction. And increasingly, the ecosystem can be more proactive in outreach than ever before — because data now generates intelligent insight on demand. We are focused on these ambitious standards across every touchpoint.

Culture: The dimension that is most often underestimated

The technology, however, is rarely the hardest part of this transition.

Citi has deployed AI tools to 140,000 employees and made AI fluency training mandatory across a workforce of 175,000. Goldman Sachs is deploying thousands of autonomous AI coding agents across its 12,000-strong developer workforce, with its chief technology officer projecting productivity gains of three to four times compared to previous AI tools. BNY's AI agents have defined identities, access controls, and dedicated workflows — with its chief data and AI officer describing the shift as moving humans from executing tasks to training and nurturing those agents.

These are cultural changes as much as technical ones, and the cultural shift is harder to replicate than the technology. To quote BCG: "70 per cent of AI's impact comes from changing how work gets done — not the technology itself."

McKinsey's analysis of the banks generating the most value from AI identified a consistent organisational differentiator, and their report merits quoting at length: "Capturing value from digital and AI transformations requires a fundamental rewiring of how a company operates."

Cross-functional teams with clear ownership, sustained by central governance rather than isolated by functional boundaries, outperform organisations where AI sits in a specialist unit and requests arrive from the business. The difference is not the sophistication of the model. It is whether the people closest to the problem — the operations analyst, the client services lead, the product manager — are empowered to identify where AI can improve their work and are supported to act on it.

At Pirum, we are orchestrating a shift from reactive to proactive, from surfacing issues to delivering solutions, and from automation to intelligence. We are investing in this deliberately: in the tools our own teams use daily, in the processes we have redesigned around AI-augmented workflows, and in the expectation that every part of the organisation — product, commercial, operations, client services — operates with AI embedded in how work gets done. We are doing this because we believe our clients deserve a partner whose own capabilities reflect the standards we are helping them reach.

Becoming AI-first in securities finance is not a project with a go-live date. It is a continuous operating model evolution, sustained by a culture that treats AI capability as a shared responsibility rather than a specialist function. The firms that embrace that reality now — in their products, their client relationships, and their internal operations — will find themselves operating at a level that would simply not have been possible before. Not because of the models they have chosen, but because of the organisation they have made AI-native.

If you'd like to learn about Pirum's Complete, Connected Lifecycle platform, and how it enables clients to unlock an AI Virtuous Cycle, get in touch with Ben and our AI team.