by Joe Logan, CIO, iManage
Over the past two years, a number of large enterprises made sweeping workforce cuts on the premise that AI could absorb the workload once handled by people, without any real cost to business performance.
That premise did not hold. Within a short span of time, several of the same organisations found themselves rehiring people – often at higher salaries – to regain the human skills they needed to get the most out of their AI investments.
This pendulum-like swing from firing to hiring provides a useful lesson for any financial services companies with an eye on trimming their human capital: resourcing decisions built around AI cannot be reactive. They require discernment.
Zoom out and take a wider view
The instinct to treat AI as a substitute for headcount is understandable but misguided. A better approach is to step back and ask what skills a particular function needs as AI becomes embedded in everyday work.
Whether we’re looking at the coding teams developing the backend that supports quicker financial transactions or the client-facing professionals preparing performance analyses or quarterly reports, the basic requirement is the same: financial enterprises need to understand what AI can do, what it should be trusted to do, and where a human-defined checkpoint still needs to sit in the process.
In part, this means identifying where over-trust in AI outputs is a risk, where validation needs to be built into a process, and where agentic tools might operate beyond their intended scope if left unchecked.
You need humans to find the blind spots
A concrete example is helpful here in illuminating what we’re talking about. Many automated processes today still run on broad-access service accounts built for a less autonomous era of automation. Extending that same access model to an AI agent is a governance risk – which means agents need their scope deliberately narrowed to the function they are meant to perform, rather than inheriting the wide permissions of legacy automation.
Here’s the key part, though: Determining where those boundaries sit, and splitting access accordingly, is itself a human skill that only comes from someone with years of experience.
Just as important as addressing any access issues is questioning whether the process being automated is still the right one in the first place, rather than layering AI onto a workflow that was designed for an earlier, manual-first era. Recognising where a process needs to be redesigned – not simply automated – is a further judgment call, and one that depends on people who understand both the old way of working and the tools now available to replace it.
In other words, in-depth knowledge of a process, and of what genuinely needs to be done, continues to sit with people. This is exactly why treating the introduction of AI as an opportunity to fire people is short-sighted. Valuable institutional knowledge walks out the door – knowledge that is costly to replace and, in a market where every organisation is competing for the same AI-literate talent, costly to re-hire.
Rather than throw the baby out with the bath water by getting rid of employees, a better approach is to look at the skills required for a particular function from a holistic point of view, then consider where agents could be introduced and where the guardrails might be needed.
Ultimately, it takes judgment and discernment to decide where AI adds value and where it introduces risk, and to build in checks that catch the difference. This requires thinking about where the human delivers value and where the AI agent might deliver the value, to combine human and automation in a meaningful way.
Avoid the churn, reap better business outcomes
AI offers many capabilities, but resourcing cannot be a knee-jerk reaction. A discerning approach is necessary to determine the right balance between human and AI capabilities for optimal business performance – never more so than when enterprises are increasingly evolving towards an ‘AI-ubiquitous’ business environment.
Taking this thoughtful approach enables organisations to bend the hiring curve and get more out of the people they already have rather than triggering a wholesale swap of people for technology. The alternative is churning talent in and out of the organisation – which will only hinder the ability to reap the better business outcomes that led financial services enterprises to AI in the first place.
