In an interview with Traders Magazine, Carissa Robb, Managing Partner of Banking & Financial Services at SolomonEdwards, discusses the risks of weakening the leadership pipeline and why AI cannot replace the apprenticeship model that has traditionally defined banking careers.

Can AI replace the apprenticeship model that has traditionally shaped banking careers?

Yes, the AI-enabled model will be a force multiplier for banks’ experienced practitioners, but not a complete substitute for growing new ones. The institutions getting this right are using AI as a teaching tool, not a shortcut. They’re asking junior employees to engage with AI outputs critically – to interrogate the recommendation, identify its assumptions, and argue with it – rather than simply accept and execute. That’s a deliberate design choice, and most banks aren’t making it intentionally.

The apprenticeship model in banking is about transmitting judgment. A junior analyst who sits next to a seasoned credit officer learns how to read a borrower, how to stress-test an assumption, and how to push back on a deal that feels wrong before the data confirms it. That judgement transfer happens through observation, feedback, and iteration over time, not through access to a knowledge base. It’s about learning how to be wrong in recoverable ways, under supervision and with feedback.

What AI can do is surface precedent faster, flag anomalies a junior analyst might miss, and give early-career professionals access to synthesized expertise they’d otherwise have to wait years to accumulate. Used well, AI compresses certain parts of the learning curve. But compression is not replacement. The judgment layer – the part that decides which anomaly matters, which precedent applies, and which risk to escalate – still requires human mentorship to develop.

As banks reduce entry-level hiring, what does that mean for the future leadership pipeline?

The pullback in entry-level hiring is one of the most consequential and underreported structural risks in financial services today. We are losing bankers. Optimizing short-term efficiency quietly hollows out the bench that produces tomorrow’s executives, risk officers, and compliance leaders.

As executive leaders, we develop judgment through years of pattern recognition – the loan that looked clean and defaulted 3 months in, the model that passed validation but failed in stress, the compliance gap that surfaced three years after the control was designed, the repeat finding caused after a merged portfolio was boarded and overlooked. You can’t compress that learning into a prompt. When you eliminate the entry points, you don’t just lose analysts, you lose the 10-year trajectory that produces your next Chief Risk Officer.

Banks that are serious about long-term resilience need to rethink the talent funnel entirely. That means preserving selective entry-level pipelines even in lean periods, creating structured rotational programs that expose early-career professionals to risk, operations, technology, and compliance simultaneously, and building internal academies that accelerate development in ways that scale. The institutions that cut entry-level pipelines now will face a leadership deficit in the next cycle that no amount of AI can solve. The remote work era hurt entry-level employees and career progression from lack of real-time solutioning and exposure. Now, AI is compounding that.

How can banks prevent institutional knowledge from walking out the door with departing employees?

This is one of the most urgent and consistently underinvested areas across the institutions we work with. The problem isn’t that banks don’t know knowledge walks out – it’s that they rarely have a systematic approach to capturing it before it does.

Most knowledge transfer efforts are reactive. For example, someone announces a retirement, and the institution scrambles to document processes over the final 90 days. By then, you’ve already lost the nuance. The tacit knowledge, such as why a control was designed a certain way, what regulatory feedback shaped a specific policy, or which counterparty relationships carry unwritten history, lives in people’s heads, not in procedure manuals.

A more durable approach has three layers. The first is structured documentation practices built into the workflow, not bolted on at departure. These might include decision logs, annotated runbooks, and post-mortem records that capture context alongside process. The second is AI-assisted knowledge management leveraging tools that can ingest unstructured institutional content, identify gaps, and surface relevant precedent to incoming staff. We’re seeing real traction here, particularly in compliance and financial crimes functions where regulatory history is deep and complex. The third layer is deliberate succession architecture and identifying critical knowledge holders well before transition risk materializes, creating overlap periods, and building redundancy into key roles. Banks tend to treat succession as an HR function, but the institutions that get this right treat it as an operational risk discipline.

What are the biggest mistakes banks make when implementing AI initiatives?

The most common mistake is deploying AI as a technology initiative rather than a business transformation initiative. When AI is implemented through an IT lens, institutions end up with well-built tools that don’t solve the right problems. The use case selection gets driven by what’s technically feasible rather than what’s operationally valuable, and adoption stalls because the people closest to the work weren’t part of the design.

The second mistake is bypassing the governance infrastructure. Banks move fast on AI pilots, get promising results, and then scale before they’ve built the oversight architecture the model requires. We’ve seen institutions deploy AI in AML alert triage or credit decisioning without adequate model validation frameworks, explainability documentation, or vendor accountability structures. When the exam comes, the tool that was supposed to reduce risk becomes a finding. Even in cases where AI outputs are reviewed by humans, institutions often discover that reviewers are repeating the same error patterns rather than providing effective oversight. AI success ultimately depends less on the technology itself and more on the governance, controls and operational rigor surrounding it.

How can financial institutions use AI to encourage more cross-functional collaboration rather than create new silos?

AI, deployed thoughtlessly, accelerates the siloing it’s supposed to break down. When each function – compliance, risk, operations, finance – builds or procures its own AI tooling, you end up with incompatible data architectures, duplicated capabilities, and no shared intelligence layer. The silo problem gets worse, not better.

The institutions that are using AI to improve cross-functional collaboration invest in a shared data infrastructure first. More practically, the use cases with the highest cross-functional value are often the ones that don’t belong cleanly to any single function. Financial crimes is a good example, as effective fraud and AML programs require integration across transaction monitoring, customer operations, technology, legal, and line-of-business teams. When AI is deployed at that intersection, it creates a shared intelligence layer that makes collaboration a requirement rather than an aspiration.

Leadership governance matters enormously here. If AI strategy is being set function-by-function, you’ll get function-by-function outcomes. Banks that establish enterprise-level AI governance with cross-functional representation make better investment decisions and see more durable adoption. It’s a governance question as much as a technology question.

What will separate the banks that succeed with AI from those that struggle over the next decade?

The winners will be the banks that understand – and refuse to compromise on – defensibility standards. Over the next decade, success will come down to three things: governance maturity, talent strategy, and the ability to learn at institutional scale.

On governance, the banks that succeed will be the ones that built accountability structures before they need them — not in response to regulatory pressure or a model failure, but because they understood that AI in financial services is inherently a risk management challenge. That means model risk frameworks that extend to third-party AI vendors, explainability standards embedded in procurement, and board-level visibility into AI exposure. The institutions that treat AI governance as a compliance checkbox will spend the next decade remediating.

On talent, the winners will recognize that AI doesn’t eliminate the need for deep human expertise. Instead, it changes the profile of that expertise. AI is an accelerator, but the competitive advantage won’t come from having the most sophisticated models alone. It will come from having people who understand both the domain deeply enough to design the right use cases and the technology well enough to govern it responsibly. That’s a rare combination, and the banks investing in developing it now are building a durable edge.

On institutional learning, AI generates continuous signals about what’s working, what’s breaking, and where models are drifting. The banks that win will be the ones that have built the feedback loops to act on those signals in real time. That requires a culture of iteration, a willingness to challenge assumptions, and an infrastructure for continuous model monitoring that most institutions are still building.

The banks that struggle will be the ones that treat AI as a one-time modernization project. The ones that succeed will treat it as a permanent operating capability – something that must be managed, governed, and improved continuously, not just implemented and left to run. We are already seeing disguised versions of entry level positions being “replaced by AI” as short-term P&L plays. But that is shortsighted. The institutions that build lasting advantage will be the ones that use AI to elevate human judgement, not simply remove human capacity.