Author

Andreea Pleşea

Published

23 April 2026

Estimated Read time

5Mins

AI is no longer seen as an add-on. It is expected as a standard in enterprise IT infrastructure explains Andreea Pleşea PhD, Co-Founder & COO at Druid AI

Digital transformation is often hailed as the answer to improve productivity, and yet, despite significant investment, the UK continues to lag behind similar markets such as the US, France and Germany in productivity growth.

The UK is recognised internationally for its financial and banking sector, and it sits at the heart of the UK economy. When banks operate efficiently, businesses move faster, but when banks are slowed by operational friction, the ripple effects are felt far and wide.

UK financial institutions operate a technology stack across core banking platforms, CRM systems, contact centre infrastructure, mobile apps, fraud systems, onboarding tools, compliance platforms and knowledge bases. Each was designed to solve a specific problem, but together they have created a fragmented set of solutions that require employees to constantly switch between applications to find the information they need to answer customer queries or understand how to make improvements to the business.

This fragmentation has created an orchestration gap, and agentic AI is the technology that can bridge it – not by adding another tool, but by becoming an essential part of the IT infrastructure.

Scripted Bots are Out – Autonomous Execution is in

The first wave of banking automation focused on what is referred to as ‘deflection’. Essentially, chatbots and Interactive Voice Response (IVR) systems were rolled out with the goal to reduce call volumes and answer basic account questions. But 61% of customers still escalate to human agents because these systems fail to resolve issues. Regulated banks cannot allow public Large Language Models (LLMs) to access core systems without strict governance. They generate responses, not orchestrate workflows.

The introduction of Generative AI tools in recent years has allowed for more natural language capabilities, but improving language alone does not complete work. Many, if not all, financial institutions don’t want public LLMs accessing their core banking systems, enforcing business rules, or figuring out whether they can stand up to the test of being audited in a highly regulated industry. Quite simply, they generate responses, they do not orchestrate business processes.

The Fundamental Difference with Agentic AI

AI agents are built from the ground up to be decision-capable and goal-oriented. They are capable of executing multi-step workflows or processes across different core platforms while operating within the strict boundaries of financial governance.

If a customer asks the question “What is the balance of my current account?” an AI agent will authenticate the customer, retrieve the necessary account data from core banking systems and provide the answer. They can also help with queries such as a card replacement, updating contact details or guiding a customer through the process of a loan application, to completion. Irrespective of whether the customer chooses to engage across chat, SMS, voice or mobile banking, the AI agent won’t lose the context of the request even if they switch platforms.

Retail banking customers interact with their bank approximately 150 times per year, and when those touchpoints are fragmented across channels, cost-to-serve rises and trust declines. However, when they are resolved quickly and securely in digital channels, efficiency and retention improve.

Making the Productivity Case for UK Banking

The productivity opportunity for UK banking lies in automating the high-volume, repeatable journeys – not through rigid, scripted chatbots, but through intelligent, governed workflow execution.

High-volume journeys such as account servicing, loan applications and fraud inquiries require secure verification, system checks and downstream actions. Yet customers are often forced to escalate to human agents to complete them.

By applying unified business rules across digital channels and legacy IVR systems, AI agents standardise this fragmented logic. A single workflow can be built once and deployed consistently across web, mobile, contact centre and messaging channels. This reduces repeat contacts, eliminates ‘start over’ frustration and frees human advisors to focus on complex cases, cross-sell opportunities and relationship management.

In a market where 17- 22% of UK consumers are actively looking for a new bank or considering switching their main bank account, consistent, frictionless service is not a luxury – it’s a competitive defence.

Improve the Infrastructure Rather than Replace it

The productivity impact extends beyond front-line customer enquiries and extends to how employees can navigate the maze of business applications to onboard suppliers, generate compliance reports, update policies or process internal IT requests. Agentic AI sits across these internal systems as well, automating repetitive processes and orchestrating tasks without forcing employees to switch between interfaces.

One of the biggest barriers to adopting this transformation from CIOs and IT leaders is a fear of ‘rip-and-replace’ programmes. Core banking systems are deeply embedded with the organisation, CRM systems anchor case management and Contact Centre as a Service (CCaaS) platforms manage routing and workforce engagement.

Agentic AI does not require these embedded systems to be replaced, it securely integrates with them, creating an operational layer that improves productivity.

Conversational AI Platforms with autonomous agents act as an orchestration layer across existing stacks. They plug into core banking systems, CRM and CCaaS infrastructure, performing governed actions while maintaining audit trails and role-based access control. This highly customisable approach allows finance and banking institutions to modernise customer journeys without destabilising foundational systems.

The AI Opportunity is Clear

This is where the infrastructure argument becomes clear. UK finance and banking institutions don’t need more applications layered onto already complex, data-sensitive, highly secure enterprise IT environments – they need intelligent systems that unify what already exists.

The UK’s next productivity gains will not come from incremental feature upgrades. They will come from rethinking how repetitive tasks move across enterprise systems. Agentic AI represents a shift from tools that respond to requests to an infrastructure that completes complex tasks, at scale. For mid-to-large retail banks and credit unions, the opportunity is clear: resolve more interactions digitally, scale capacity without expanding headcount, protect margins and strengthen customer trust.

Learn more at druidai.com

  • Artificial Intelligence in FinTech
  • Cybersecurity in FinTech
  • InsurTech
  • Neobanking

Author

Richard Doherty

Published

8 April 2026

Estimated Read time

6Mins

Richard Doherty, Head of Wealth & Asset Management, Publicis Sapient, on how asset managers must redesign their enterprise for AI-driven decision intelligence

The asset management industry is entering a structural inflexion point. The first wave of AI focused on improving productivity through copilots and automation. The next wave will fundamentally reshape how decisions are made, executed, and governed across the enterprise. This is not a technology upgrade. It is an operating model shift.

Despite significant investment, many firms remain trapped in fragmented AI experimentation. A majority are yet to realise meaningful economic returns from AI, not due to lack of capability, but due to a failure to redesign how intelligence is applied across the organisation. The gap between ambition and outcome is not a technology problem. It is a structural one.

From Automation to Decision Intelligence

The industry conversation has evolved. The question is no longer whether to adopt AI, but how to scale it across the enterprise. However, most firms are still approaching this challenge through the lens of automation, identifying tasks that can be executed faster or at lower cost. This delivers incremental value, but does not address the underlying constraint: the structure of decision-making within the organisation.

Traditional operating models are built around sequential workflows. Work moves from function to function: research, compliance, operations, and distribution, each dependent on the previous stage. This creates latency, duplication, and fragmentation. Agentic operating models shift the focus from tasks to decisions.

Instead of asking “Which processes can we automate?”, leading firms are asking: “Which decisions can be augmented or owned by intelligent systems?”

This shift enables organisations to move from sequential workflows to parallel decision systems; from human-led analysis to AI-assisted reasoning; from periodic insight to continuous intelligence. The result is not a marginal improvement. It is a step-change in how the enterprise operates.

The Pressures Driving Change

This transformation is not happening in a vacuum. Asset managers face mounting structural pressures: margin compression driven by fee pressure and passive competition; rising operational complexity from regulation and product proliferation; and advisor capacity constraints that limit scalable growth. Agentic operating models directly address all three.

By automating complex workflows, rather than individual tasks, firms can significantly increase advisor and analyst capacity without proportional cost increases. Parallel decision systems reduce the time required to launch products, respond to market events, and deliver client insights. This compresses cycles from months to days. Continuous monitoring of guidelines, portfolios, and operational processes reduces exposure to regulatory breaches and operational failures.

These are not theoretical benefits. They represent measurable improvements in cost-to-serve, time-to-market, and operational resilience.

Not all Intelligence is the Same

To scale AI effectively, organisations must recognise that not all problems require the same type of intelligence. Enterprise AI operates across three distinct layers, and conflating them is one of the primary reasons AI initiatives fail to scale.

Deterministic systems execute predefined rules with complete consistency. They are essential for functions where there is zero tolerance for error, trade validation, settlement processing, and regulatory reporting. If a business outcome must be identical every time, deterministic logic remains the correct approach.

Predictive systems use historical data to forecast outcomes. Applied in areas such as portfolio risk modelling, fraud detection, and client churn prediction, they generate probabilities and insights, but they do not interpret context or make decisions independently.

Agentic systems operate where problems require interpretation, judgment, and contextual understanding, investment guideline interpretation, regulatory document analysis, portfolio insights, and client communication. These systems can reason across complex information, generate insights, and take action within defined boundaries.

The ‘Different but Valid’ Dilemma

A critical challenge in adopting agentic systems is understanding how they behave. Traditional software produces identical outputs. Agentic systems produce reasoned outputs.

This introduces what I call the ‘different but valid’ dilemma. An agent may take a different reasoning path from a human and arrive at a different, but still correct, conclusion. This variability is not an error. It is inherent to reasoning systems.

The real risk lies in hallucination, outputs that are not grounded in data or evidence. Managing this requires organisations to clearly define where variability is acceptable. All AI-driven processes sit on a spectrum: deterministic actions with no variability (trade execution), predictive actions with controlled variability (risk scoring), and agentic actions with higher variability (investment insights).

Leading firms design systems where agents perform reasoning, deterministic systems enforce execution, and humans retain oversight on high-consequence decisions. This balance enables both flexibility and control.

The Operating Model Shift

The most significant change is not technological; it is organisational. Traditional models are built on functional workflows. Agentic models are built on coordinated decision systems.

Consider what launching a new investment product looks like under each model. In a traditional model, it involves sequential handoffs between teams, compliance reviews the guidelines, operations configures the systems, and distribution drafts the client narrative. Each stage waits for the last.

In an agentic model, intelligent systems operate in parallel: compliance agents interpret guidelines, operations agents configure constraints, distribution agents generate client narratives, and governance agents validate outputs. This orchestration compresses timelines, reduces friction, and enables continuous decision-making. It represents a fundamental redesign of how work is performed.

Governance: the Foundation for Trust

Trust is the prerequisite for scaling AI. Without it, adoption stalls, not because the technology fails, but because the organisation cannot adequately explain or defend the decisions it makes.

Leading firms implement governance models built on three principles. First, explainability: every decision must be traceable and auditable. Second, authority boundaries: agents operate within clearly defined limits. Third, human oversight: high-consequence decisions remain under human control.

Regulatory expectations will continue to evolve, but one principle remains constant: organisations must be able to explain how decisions are made.

Scaling AI is a Leadership Challenge

Executives must take a deliberate approach across four areas:

  • Define the intelligence model: map business problems to deterministic, predictive, or agentic systems.
  • Build the foundation: invest in data, infrastructure, and orchestration capabilities.
  • Redesign the operating model: shift from workflows to decision systems.
  • Implement governance to ensure transparency, control, and compliance.

Start with high-value use cases and expand rapidly across the enterprise. The firms that act now will establish a structural advantage in cost, speed, and decision quality. Those that do not risk being constrained by legacy operating models that cannot scale with the demands of modern markets.

The Question is not if, it is Who

The industry is not simply adopting new technology. It is redefining how decisions are made. The firms that succeed will not be those that deploy AI tools in isolation. They will be those who design the right form of intelligence for each problem, redesign their operating models around intelligent systems, and scale agentic capabilities across the enterprise.

This shift is already underway. The question is no longer whether it will happen. The question is which firms will lead, and which will be forced to follow.

Learn more at publicissapient.com

  • Artificial Intelligence in FinTech
  • Blockchain & Crypto
  • Data & AI
  • Digital Strategy
  • Fintech & Insurtech

Author

Thomas Benjaminsen Normann

Published

27 March 2026

Estimated Read time

6Mins

Thomas Benjaminsen Normann, Product Director at Paymentology on the future for agentic payments and the progress still to be made

Santander and Mastercard’s live AI-agent payment pushed the industry past the stage of talking about agentic commerce as a future use case and into the reality of a transaction moving through live banking infrastructure. In doing so, it placed an AI agent at the point of spend within a system that still assumes the person initiating the payment is also the one making the decision and carrying the liability.

That assumption is far easier to sustain when a payment draws on existing funds than when it creates a debt that someone must later repay. And may dispute. As soon as an agent moves from guiding a choice to completing the transaction, the usual alignment between instruction, authorisation and liability becomes harder to see.

Card authorisation has long rested on a simple premise: the person using the card is the one deciding to spend. Even when the transaction runs through a wallet, an app or a stored credential, the model still relies on a cardholder who is directly involved in the act of payment.

Agentic payments

Agentic payments stretch that arrangement. The customer may have set the rules, the budget or the merchant preference in advance, but the point of execution can now sit with software acting later and at speed. The question then extends beyond whether the transaction was authenticated to whether the debt it created was taken on with the kind of consent and clarity card systems have traditionally relied on.

Mastercard has responded by building a stronger trust layer around delegated intent. Once software acts on a customer’s behalf, the usual signs of presence and intent at the moment of payment carry less weight than they do in an ordinary card transaction. Santander’s pilot showed that this can be handled inside a tightly controlled framework with predefined permissions.

The challenge becomes very different once the same model moves into ordinary credit flows, where issuers are dealing with borrowing, repayment and dispute risk rather than a bounded test case.

Risk models built on human behaviour

Fraud systems and credit models have been trained to read people. How they spend, how quickly they move, where they buy, and what tends to happen before repayment trouble begins to show. An AI agent, even when acting entirely within a customer’s instructions, is unlikely to look much like that. It may search more widely, compare more aggressively, transact at unusual times and behave with a consistency that looks odd against a human baseline. Some legitimate payments will appear suspicious. Some suspect ones may look routine. Signals that once separated ordinary behaviour from risky behaviour will arrive in forms the system is not used to reading.

Research from Capgemini indicates that 71% of consumers want generative AI integrated into shopping interactions. Meanwhile, 58% say they already use generative AI instead of traditional search for recommendations. That does not mean autonomous purchasing becomes mainstream overnight, but it does suggest the move from AI-assisted discovery to AI-executed transactions will not stay theoretical for long. For issuers, that means transaction systems are about to encounter a new behavioural signature without much history behind them.

The pressure does not sit only with fraud screening. Credit decisioning is built on assumptions about how people build balances, revolve debt, repay over time and run into repayment trouble. An AI agent may be acting entirely within a customer’s instructions while still producing patterns those models were never trained to read cleanly. A sudden increase in spend, an unusual merchant mix or a burst of late-night activity may deserve scrutiny when a person generates it.

The same signals may be perfectly consistent with a software agent searching widely, responding instantly to price changes or executing against preset rules with much greater speed and regularity than a person would. Once that behaviour starts landing in the credit book, signals that once carried meaning around affordability, intent or emerging repayment risk become less reliable as indicators.

Signals the authorisation layer does not carry

The transaction also arrives with gaps that matter more once software is involved. Existing payment messages can identify the merchant, the amount, the credential used and the authentication path. What they do not natively describe is whether the action came from a customer or an agent, what spending authority had been delegated, whether that authority was limited to a category, merchant or price threshold, and whether the funding source was intended to be debit, charge or revolving credit. A payment can be technically valid while still leaving the issuer with too little context about how the decision was made.

A controlled pilot can solve some of that by imposing rules around the transaction from outside the standard message, which is effectively what bounded testing is for. Everyday credit use is less forgiving. If the issuer is expected to approve the payment, apply the right controls, score the exposure and later defend the outcome in a dispute, those signals have to be legible inside the flow rather than reconstructed around it after the event.

At that point, the question is less about whether the payment experience works and more about whether the issuer-side controls underneath it can carry the weight. That includes the ability to apply rules in real time, restrict how a credential can be used, and keep a clear record of how the transaction was authorised and what kind of exposure it created.

The missing context does not stop at authorisation. It follows the transaction further down the line, when an issuer has to explain why a payment was approved, whether the agent acted within its delegated scope. And how that scope should be evidenced if the customer challenges the transaction. Card systems are used to relying on the credential, the authentication path and the transaction record.

Digital versus Traditional Wallets

Agentic payments demand something more granular: a clearer account of who or what acted, under what limits, and with what right to create a liability on the customer’s behalf. The control layer around that decision, including how credentials are restricted and how delegated authority is defined, starts to matter much more than it did in a conventional wallet or stored-card journey.

Infrastructure many issuers built out for tokenised wallets now looks more like part of the control architecture for agent-led spend. Because the payment credential itself may need tighter restrictions than the market has been used to applying.

Santander and Mastercard have shown that an AI agent can now make it all the way through a live payment flow. What follows from that is less about whether software can reach the point of spend and more about what the rest of the stack needs to know once it gets there. If agentic payments are to move beyond controlled deployments and into ordinary credit use, issuers will need clearer ways to tell who acted, under what authority, against which funding source, and with what liability attached. Until those signals travel cleanly through the flow rather than being inferred around it, agentic payments on credit will remain easier to demonstrate than to absorb into everyday card operations.

Learn more at paymentology.com

  • Artificial Intelligence in FinTech
  • Digital Payments
  • Embedded Finance
  • Neobanking

Author

Martijn Gribnauis

Published

23 March 2026

Estimated Read time

5Mins

Martijn Gribnauis, Chief Customer Success Officer at Quant, on why Agentic AI will redefine financial services

A recent Google Cloud survey showed that only 13% of finance organisations are currently using agentic artificial intelligence. This number needs to, and will rise when you consider that 88% of financial leaders are seeing ROI from generative AI already. Agentic is the next and most advanced evolution of artificial intelligence the world has ever seen. 

Agentic AI is not on the way. It is here and already reshaping how forward-leaning financial institutions operate. In 2026, for IT and finance leaders to build an insurmountable competitive lead they must deploy agentic AI in every area where it can safely and effectively create value. The institutions that hesitate will find their business models under threat from familiar competitors and newcomers alike.

Reinvention of Core Processes

Agentic AI is poised to reinvent core financial processes. Bookkeeping, record maintenance, and period-end close are nearing complete automation. Month-end processes that once required late-night, stress-filled marathons will evolve into continuous, largely automated cycles. IT teams will no longer spend evenings on high alert waiting for failures. 

This shift also frees IT leaders, finance teams, and operations functions from monotonous repetitive tasks. Instead of focusing on system uptime and manual reconciliation, they will collaborate with the C-suite on strategic initiatives that drive growth and revenue. 

Understanding Why Adoption Is So Low

Despite the promise of Agentic AI, there is understandable caution. Some 80% of organisations have reported ‘risky behaviour’ from AI agents, and in the world of finance that is an alarming number. Finance is one of the most regulated, risk-averse sectors in the world. The fear of losing control remains the primary reason so few in the industry have embraced Agentic AI.

Loss of control and fear of catastrophic error

Financial leaders fear that an autonomous system could go ‘off script’, mis-route payments, misinterpret rules, or inadvertently cause compliance breaches. In finance, even small errors can trigger major financial or regulatory consequences.

Security and data privacy concerns

Large AI models require huge quantities of sensitive data. Organisations worry about breaches, cyber-attacks, or manipulation. An AI agent with improperly configured permissions could, in theory, execute fraudulent transactions or expose confidential customer information.

Bias and fairness risks

If AI agents make decisions using incomplete or fragmented data, they risk perpetuating or amplifying bias. At scale, biased decision-making can undermine customer trust and expose firms to legal and regulatory challenges.

Regulatory ambiguity and audit difficulty

Regulators are still determining how to govern agentic AI. Some organisations fear that early adoption could unintentionally violate rules or create future audit vulnerabilities.

These fears are legitimate, but not insurmountable.

Tackling the Adoption Barriers: A Practical Blueprint for Finance Leaders

To capitalise on Agentic AI’s immense potential, leaders must take a structured approach grounded in business value, security, and trust.

1. Start With Clear, Measurable ROI and Efficiency Gains

In finance, adoption accelerates when decision-makers see proof of value.

Start by automating repetitive processes. Agentic AI can handle tasks like data entry, reconciliation, invoice matching, and initial fraud checks faster and more accurately than humans. This leads to reduced operational overhead as automation lowers labour costs, shortens processing times, and reduces error rates. Demonstrating these savings through case studies or internal pilots is critical to changing minds. 

AI agents can enable revenue growth by analysing huge data sets to identify new investment opportunities, optimise trading strategies, and generate personalised product recommendations. Each of these capabilities directly impacts top-line growth.

2. Strengthen Risk Management and Compliance Through AI

Agentic AI will improve risk management when deployed responsibly. This starts with real-time fraud detection. AI agents can monitor transactions continuously, identifying patterns that suggest fraud long before traditional systems would detect an anomaly.

Continuous monitoring is also incredibly helpful when it comes to compliance. AI agents excel at ensuring adherence to KYC and AML regulations. They can automatically maintain audit trails, identify missing documentation, flag anomalies, and escalate issues instantly.

Enhanced stress testing and scenario modelling can both be completed via Agentic AI. It can simulate complex market environments more dynamically than legacy tools, providing deeper insights into vulnerabilities and improving resilience. When showcased and presented in this context, agentic AI becomes a risk-reduction tool in the eyes of decision makers. 

3. Directly Address Security and Trust Concerns

Trust is the cornerstone of adoption. Implement enterprise-grade security architecture that includes encryption, secure APIs, strict access controls, and continuous monitoring of agent behaviour. And, use explainable and transparent AI systems (XAI) so your finance teams understand the reasoning behind decisions. XAI helps provide interpretable outputs that support auditability and regulatory compliance.

Start small with a controlled, low-risk pilot. A proof-of-concept in a non-critical workflow helps teams understand the technology, gather evidence, and build internal support before scaling. Produce numbers based reporting that speaks the language of the people who make the decisions. Show, don’t just tell them how agentic will move the business forward.

4. Highlight the Competitive Advantage

Agentic AI adoption is not just an efficiency upgrade. It is a competitive imperative. AI agents create faster innovation cycles by accelerating product development, service delivery, and operational improvements.

They also provide superior customer experience. From instant account servicing to personalised financial recommendations, Agentic AI delivers the speed, personalisation, and convenience customers expect. Plus, it scales exponentially. No matter how many people call in at the same time, an agentic agent will answer immediately. Agentic AI reduces up to 86% of time spent in complex workflows that were traditionally handled only by people. This will be huge in getting ahead of your competition. 

5. Build Momentum Through Internal Champions

Adoption increases when respected leaders advocate from within. Mid-level managers, AI-literate staff, or members of the C-suite who understand the technology can serve as champions. Use them and their beliefs to drive alignment, communicate benefits, and counter misconceptions. The more people from different departments and levels of the organisation that talk up the technology, the more likely you are to get buy-in. 

Your Time is Now

Agentic AI will redefine financial services. The organisations that act today will build capabilities, insights, and competitive advantages that late adopters will not be able to replicate. Finance leaders must begin asking where agentic AI can support their business, where it can remove friction, where it can unlock growth, and where it can transform operations. The firms that act now will lead the industry. Those that hesitate will not get the chance to catch up.

The only remaining question for finance organisations is not whether agentic AI will change the industry, but how quickly they choose to deploy it.

Learn more at quant.ai

  • Artificial Intelligence in FinTech
  • Data & AI
  • Digital Payments
  • Digital Strategy

Author

Dr Megha Kumar

Published

17 March 2026

Estimated Read time

6Mins

Dr Megha Kumar, Chief Product Officer and Head of Geopolitical Risk at CyXcel, on whether our risk and regulatory frameworks and institutional cultures can keep pace with Agentic AI

Within the next couple of years, Agentic AI is likely to progress from early stages of operation to be fully embedded within systems. Its expansion will be subtle rather than spectacular. It will integrate steadily into enterprise platforms, logistics networks, compliance workflows, cybersecurity operations centres and executive decision-support tools. Processes will move faster, operating expenses will decline and performance indicators will trend upward.

Yet these visible improvements mask a deeper challenge. The regulatory exposure, data governance pressures and erosion-of-trust risks associated with Agentic AI are being misjudged.

Unlike earlier AI applications designed primarily to generate outputs – whether text, imagery, or predictive insights – agentic systems are built to act. They sequence decisions, draw from multiple data environments, initiate consequential processes and function at scale with differing levels of human supervision. In sandbox environments this can seem contained and controllable. Over extended periods in live environments, however, sustained oversight, traceability and effective governance become significantly more complex.

Evolving Operational Complexity

There are two key challenges that businesses must address.

First, how do organisations monitor what agentic systems are doing once deployed? These systems evolve through updates, integrations and retraining and they interact with new data environments.

Second, how do you ensure responsible behaviour throughout the lifecycle? Regulators, policymakers and customers will likely expect firms to shift from compliance assurance to risk assurance and demonstrable evidence of trust and transparency.

The prevailing assumption is that human oversight will mitigate these risks. Human in the loop or human over the loop has become the default reassurance. In practice, however, that assumption breaks down far faster than many anticipate.

When a system works 95 per cent of the time, human reviewers limit their scrutiny. Behavioural science tells us that automation bias and complacency occur when automated systems are high-performing. Employees often become validators of AI outputs rather than critical examiners. The diligence gap widens gradually and then suddenly.

Facing Up to Difficult Questions

How do you incentivise employees to remain diligent checkers when the system mostly ‘works’?  And how much time does effective oversight actually require? True review is not a cursory glance at a dashboard. It involves interrogating assumptions, validating inputs, checking context and assessing downstream consequences. In many cases, meaningful oversight may take nearly as long as performing the original task manually. When checking becomes more costly than doing the job yourself, pressure to ‘trust the system’ intensifies.

And what happens to accountability when oversight exists on paper but not in practice? Governance documentation may show layered review structures, escalation pathways and audit processes. Yet if humans are functionally disengaged, responsibility becomes dispersed. When errors surface, organisations may struggle to attribute fault – was it the model design, the data, the integrator, the operator or the reviewer who signed off without fully scrutinising?

Regulators are only beginning to grapple with these realities. In jurisdictions such as the European Union, the EU AI Act introduces risk-based obligations, documentation requirements and human oversight provisions. These are important steps, however, the operationalisation of those requirements in dynamic, agentic environments remain untested at scale. Compliance on paper will not automatically translate into resilient governance in practice.

Addressing the Trust Challenge

Beyond regulatory exposure, there is a broader trust challenge emerging.

As Agentic AI systems scale across industries, they will generate vast volumes of automated outputs – reports, communications, risk assessments, content, decisions and transactions. If errors or manipulations spread through interconnected systems, confidence in digital outputs may erode.

In geopolitically sensitive contexts, this has profound implications. Agentic systems interacting with external data sources could amplify disinformation, introduce biased datasets or make decisions based on manipulated inputs. The speed of automation may outpace the speed of verification. Trust, once diluted, is difficult to restore.

Data protection risks will also intensify. Agentic systems frequently require broad access privileges to perform tasks effectively. They may access internal databases and personal data and interact with third-party platforms. Each interaction creates potential exposure points. A single misconfiguration or prompt injection attack could trigger cascading consequences across systems.

The next phase of AI adoption will not simply amplify productivity: it will amplify regulatory, legal and reputational risk. This moment therefore demands serious scrutiny before agentic AI becomes deeply embedded in business infrastructure.

The Moment for Action has Arrived

So, what should organisations be doing now?

To begin with, organisations need to look past superficial, tick-box compliance. Effective governance cannot live solely in policy documents – it must function in day-to-day operations. This means investing in continuous monitoring capabilities, robust audit trails and real-time anomaly detection tailored specifically to Agentic AI behaviours.

In parallel, incentive structures should be redesigned. Meaningful human oversight will not happen if it is treated as secondary to speed or output. If employees are expected to provide meaningful review, organisations must allocate time, training and authority accordingly. Performance metrics should reflect risk management responsibilities, not just output rate.

Clear lines of accountability are equally important. Senior leadership and boards should determine who carries ultimate responsibility for outcomes produced by agents. Where third-party vendors are involved, responsibilities must be contractually and operationally defined. Incident response mechanisms should be rehearsed in advance, rather than presumed to work when pressure is high.

Expertise must also be integrated across functions. Legal, risk, compliance, cybersecurity, data protection and operational teams should be engaged from the outset. Deploying Agentic AI is not simply a technical upgrade – it reshapes the organisation’s risk profile.

Finally, resilience demands deliberate stress-testing. Leaders should examine not only pathways to success but how models fail at scale. How would the organisation respond if a system update embedded systemic bias, if an integration vulnerability enabled unauthorised activity or if automated actions eroded customer confidence? Rigorous scenario exercises, however uncomfortable, are essential to building genuine preparedness.

As Agentic AI advances, Risk Management Should Match its Pace

None of this is an argument against adoption. Agentic AI presents meaningful productivity improvements and the potential for sustained competitive differentiation. Organisations that deploy it with discipline and foresight may secure a measurable advantage. The danger lies not in adoption itself, but in pursuing acceleration without knowing the risks and putting the right guardrails in place.

The coming two years are critical for businesses. Before these systems become deeply embedded in core processes, organisations have an opportunity to shape the control environment around them.  However, once agentic systems are fully embedded, retrofitting controls will be far more difficult and costly. Leaders must therefore treat this period as a design phase for oversight, not merely a race for competitive advantage.

Agentic AI is advancing rapidly. The defining question is whether our risk and regulatory frameworks and institutional cultures can evolve just as quickly.

Learn more at cyxcel.com

  • Artificial Intelligence in FinTech
  • Data & AI
  • Digital Strategy

Author

Alessandro De Leonardis

Published

5 December 2025

Estimated Read time

6Mins

From banking to alternative funds, modular architecture is the missing link for effective adoption of artificial intelligence, writes Alessandro De Leonardis, CIO of Armundia Group

The global banking industry is approaching a strategic crossroads – one that will prove expensive for those who choose the wrong direction. Financial institutions stand to lose USD 170 billion in profits over the next decade if they do not adapt rapidly to the evolution of artificial intelligence, according to the McKinsey Global Banking Annual Review 2025. Yet the report’s most provocative insight isn’t about AI itself, but the infrastructure required to leverage it effectively.

Agentic AI has the potential to reshape banking at its foundations. Early adopters will strengthen long-term advantages, potentially boosting returns on tangible equity by up to four percentage points. On the other hand, laggards face structural declines in profitability. The difference between these outcomes won’t be determined by who adopts AI first, but who has the architectural foundations to implement it effectively. Increasingly, those foundations are modular.

From Generative to Agentic AI: Revolution not Evolution

To understand why architecture matters so deeply, we must distinguish between the two paradigms reshaping financial services.

Generative AI, the star of 2023-24, excels at creating content: automated reports, document summaries, customer-service response, and so on. It is powerful, but fundamentally reactive. GenAI requires human prompts and produces outputs that must still be reviewed and acted upon by humans.

Agentic AI represents a step-change. These systems combine autonomous reasoning, planning, and execution. They don’t only generate recommendations, they act on them. An Agentic AI system can autonomously manage an entire loan-approval workflow: collecting documents, verifying information, assessing creditworthiness, checking regulatory compliance, and making approval decisions, all without human involvement at each step.

The impact is already measurable. MIT Technology Review Insights found that 70% of banking leaders are implementing agentic AI through production deployments (16%) or pilot projects (52%). Deloitte reports early adopters achieving 30–50% cost reductions in specific workflows. McKinsey anticipates the emergence of a “disruptive agentic business model” within three to five years, with potential cost reductions of up to 70% in some categories. But the benefits are far from evenly accessible.

Why Monolithic Architecture are Incompatible with AI

The uncomfortable truth is that most banks are attempting to deploy twenty-first-century AI on twentieth-century infrastructure. And it doesn’t work.

Legacy systems still absorb around 60% of banks’ technology budgets, according to a 2024 Bloomberg Intelligence survey. These monolithic architectures were never designed for the rapid iteration, continuous integration, and granular governance demanded by AI deployment.

Monolithic systems require release cycles lasting months; AI models require continuous retraining and fine-tuning based on real-world performance. The mismatch is structural. Modern Agentic AI relies on orchestrating multiple specialised agents… One for data collection, another for risk evaluation, a third for decision execution. Monolithic architectures struggle to support this level of inter-system communication.

Governance is another barrier. AI systems require differentiated risk controls depending on the level of autonomy. A fully autonomous fraud-detection agent needs different guardrails than a customer-service chatbot. Monolithic systems offer all-or-nothing governance, not graduated controls.

Financial institutions cannot transform everything at once; they need incremental adoption. Starting with high-impact use cases, learning, then expanding. Monolithic architectures force “big-bang” transformations that almost never succeed.

This architectural misalignment explains why so many AI initiatives stall in pilot purgatory, never reaching production scale.

Modular Architecture as an Enabler of AI

Modular, service-based FinTech architecture solves these problems by design. Instead of monolithic platforms, modular systems are composed of independent, interoperable functional blocks connected via APIs. Each module can be developed, updated, or replaced without affecting the whole.

The key is the concept of the service: a module that does not expose standardised technical interfaces simply does not function. Services are the technical objects enabling interoperability:

  • A compliance module exposes services for regulatory checks,
  • A data-ingestion module exposes services for data collection and structuring,
  • An Agentic AI module exposes services for executing autonomous workflows.

This architecture creates an ecosystem where each component has clear responsibilities and well-defined interfaces.

For AI deployment, this translates into concrete advantages. Banks are implementing Agentic AI systems into specific processes – KYC/AML screening, credit-memo generation, collections monitoring, intelligent communication routing – without rebuilding their entire stack. Service-based modularity allows AI agents to be activated on circumscribed workflows, with impact measured before expansion.

Because agents operate within discrete modules, failures remain contained. A malfunctioning fraud-detection agent does not propagate into customer-facing systems. This isolation allows institutions to experiment more boldly.

Service-based architectures also enable integration of best-of-breed AI solutions. One module may use Anthropic’s Claude for document analysis, another Google’s Gemini for customer interaction, a third proprietary models for highly specialised credit scoring. Monolithic systems lock institutions into single-vendor dependencies.

Different modules can carry different levels of AI autonomy, aligned with risk profiles and regulatory requirements: high autonomy for customer-service bots, human-in-the-loop supervision for lending decisions.

As McKinsey notes, the winners of this transformation will practise “precision over heft”- implementing AI surgically where it generates measurable bottom-line impact. Service-based modular architecture is the technical manifestation of such precision.

Techfin vs FinTech: When Architecture Comes First

There is a fundamental difference between starting from finance and adding technology, and starting from technology and specialising in finance.

In the first case, solutions are built top-down – gather functional requirements, then find the technology to satisfy them.

In the second, solutions are built bottom-up – design the architecture before the functional requirements, optimising for flexibility rather than feature completeness.

When designing wealth- and asset-management platforms – such as FundWatch or 360 FUNDS – this distinction becomes tangible. Being AI-ready does not mean adding an ‘AI layer’ on top of an existing platform. It means the modular architecture allows AI capabilities to be integrated precisely where needed.

Modularity operates along two dimensions:

  • Process modules (compliance, analytics, reporting, client engagement) that can be activated independently;
  • Target modules tailored for different market participants: custodians, asset servicers, alternative-fund managers, wealth advisers—each activating different module combinations.

AI governance is embedded in the architecture, not layered on top. A fully autonomous reconciliation agent operates under different guardrails than a semi-autonomous investment-recommendation agent—different approval workflows, audit trails, and supervision requirements.

This approach does not remove the need for transformation, but it changes its rhythm. Instead of three-year platform-replacement projects, institutions can transform progressively: start with a high-impact module, prove value, learn from deployment, scale outward.

The key managerial shift is conceptual: the question is no longer “When will our digital transformation be finished?” but “Which module do we activate this quarter, and what do we learn?”

The $170bn Question

McKinsey’s warning – USD 170 billion of potential profit erosion – is not inevitable. Avoiding it requires strategic decisions today about the technology architecture of tomorrow.

The institutions that will thrive are not necessarily the largest or the earliest adopters of AI. They will be those building modular infrastructures engineered for precision, capable of integrating AI surgically, experimenting rapidly, scaling intelligently, and governing rigorously.

They will recognise that AI is not merely a technological deployment, it is an architectural imperative. And they will understand the deeper truth: in the Agentic AI era, precision beats scale.

The question faced by every financial institution is not whether to adopt AI, but whether its architecture can support it. For most legacy systems built on monolithic foundations, the honest answer is no.

The modular imperative is clear. The question remains: are you building for yesterday’s challenges or tomorrow’s opportunities?

Find out more at armundia.com

  • Artificial Intelligence in FinTech

Author

Dan Brightmore

Published

13 March 2025

Estimated Read time

5Mins

Fouzi Husaini, Chief Technology & AI Officer at Marqeta, answers our questions about Agentic AI and its applications for businesses

Agentic AI is emerging as the leading AI trend of 2025. Industry figures are hailing Agentic AI as the broadly transformative next step in GenAI development. The year so far has seen multiple businesses release new tools for a wide array of applications. 

The technology combines the next generation of AI tech like large language models (LLMs) with more traditional capabilities like machine learning, automation, and enterprise orchestration. The end result could lead to a more autonomous version of AI: Agents. These agents can set their own goals, analyse data sets, and act with less human oversight than previous tools. 

We spoke to Fouzi Husaini, Chief Technology & AI Officer at Marqeta about what sets Agentic AI apart whether the technology really is a leap forward in terms of solving AI’s shortcomings, and how Agentic AI could solve business problems.

1. What makes AI “agentic”? How is the technology different from something like Chat-GPT? 

“Agentic refers to the type of Artificial Intelligence that can act as agents and on its own. Agentic AI leverages enhanced reasoning capabilities to solve problems without prompts or constant human supervision. It can carry out complex, multi-step tasks autonomously.

“GenAI and by extension Large Language Models, the most famous example being ChatGPT, require human input to solve tasks. For instance, ChatGPT needs user prompts before it can generate content. Then, sers need to input subsequent commands to edit and refine this. Agentic AI has the capability to react and learn without human intervention as it processes data and solves problems. This enables it to adapt and learn much faster than GenAI.”

2. Chat-GPT and other LLMs frequently produce results filled with factual errors, misrepresentations, and “hallucinations”, making them pretty unsuited to working without human supervision – let alone orchestrating important financial deals. What makes Agentic AI any better or more trustworthy? 

“All types of AI have the possibility to ‘hallucinate’ and produce factually incorrect information. That being said, Agentic AI is usually less likely to suffer from significant hallucinations in comparison to GenAI. 

“Agentic AI’s focus is specifically engineered to operate within clearly defined parameters and follow explicit workflows, making it particularly well-suited for having guardrails in place to keep it on task and from making errors. Its learning capabilities also allow it to recognise and adapt to its mistakes, ensuring it is unlikely to hallucinate multiple times.”

“On the other hand, GenAI occasionally generates factually incorrect content due to the quality of data provided, and sometimes because of mistakes in pattern recognition.”

“In fintech, Agentic AI technology can make it possible to analyse consumer spending data and learn from it, allowing for highly tailored financial offers and services that are more accurate and help to create a personalised finance experience for consumers.” 

3. How could agentic AI deployments affect the relationship between financial services companies and their customers? What about their employees? 

“The integration of Agentic AI into financial services benefits multiple parties. First, 

integrating Agentic AI into their offerings allows financial service companies to provide their customers with bespoke tools and features. For instance, AI can be used to develop ‘predictive cards’. These cards can anticipate a consumer’s spending requirements based on their past behaviour. This means AI can adjust credit limits and offer tailored rewards automatically, creating a personalised experience for each individual.

“The status quo’s days are numbered as consumers crave tailor-made financial experiences. Agentic AI can allow fintechs to provide personalised financial services that help consumers and businesses make their money work better for them. With Agentic AI technology, fintechs can analyse consumer spending data and learn from it. This allows for more tailored financial offers and services.   

“As for employees, Agentic AI gives them the ability to focus on more creative and interesting tasks. Agentic AI can handle more routine roles such as data entry and monitoring for fraud, automating repetitive tasks and autonomous decision making based on data. This helps to reduce human error and enables employees to focus more time and energy on the creative and strategic aspects of their roles while allowing AI to focus on more administrative tasks.”

4. How would agentic AI make financial services safer? 

“Agentic AI has the capability to make financial services more secure for financial institutions and consumers alike, by bringing consistency and tireless vigilance to critical financial processes. With its ability to analyse vast strings of information, it can rapidly identify anomalies in spending data that indicate potential instances of fraud and can use its enhanced reasoning and ability to act without human prompts to quickly react to suspicious activity. 

“While a human operator will be susceptible to decision fatigue, an AI agent could always be vigilant and maintain the same high level of precision and alertness 24/7. This is vital for fields like fraud detection, where a single missed signal could lead to significant consequences.

“Furthermore, its capability to learn without human interaction means that it can improve its ability to detect fraud over time. This gives it the ability to learn how to identify new types of fraud, helping it to adapt as schemes become more sophisticated over time.” 

5. What kind of trajectory do you see the technology having over the next year to eighteen months?

“In fintech, Agentic AI integration will likely begin in the operations space. These areas manage complex, but well-defined, processes and are perfect for intelligent automation. For instance, customer call centres where human agents usually follow set standard operating procedures (SOPs) that can be fed into an AI system, which makes automation easier and faster than before.

“In the more distant future, I believe we will see Agentic AI integrated into automated workflows that span entire value chains, including tasks such as risk assessment, customer onboarding and account management.” 

  • Artificial Intelligence in FinTech