For banks moving artificial intelligence from pilots into day-to-day operations, the key question is no longer whether to use AI. It is how to judge whether an investment is worthwhile. Measuring value solely through job reductions or hiring freezes can produce an incomplete—and potentially misleading—business case.
In a recent two-part analysis on the true cost of AI, Chris Skinner argues for a fuller view of AI economics. His central point is practical: an AI programme has costs beyond software licences, but it may also create value beyond labour savings. For banking leaders, that means assessing both sides of the equation with the same rigour.
Why headcount is not enough for AI ROI in banking
Headcount reduction is easy to place in a spreadsheet. If an automated process reduces manual work, finance teams can estimate a cost saving. Yet that calculation does not necessarily show whether the organisation has made better use of people, improved a customer journey, reduced operational risk or created a capability it did not previously have.
Skinner’s argument challenges leaders to distinguish between removing work and creating value. A bank may use AI to help colleagues retrieve information, draft routine communications, detect exceptions or support decisions more quickly. The financial benefit may be realised through faster turnaround, fewer avoidable errors, greater capacity or more consistent service—not simply fewer employees.
This matters because a narrow cost-cutting target can shape implementation choices. If teams are rewarded only for reducing staff numbers, they may overlook applications that strengthen controls, improve responsiveness or free specialists to focus on higher-value judgement and customer interactions.
Chris Skinner’s AI speaker lens: count the full cost base
The other half of the business case is cost. AI may appear inexpensive at the point of use, particularly when a tool is purchased as a cloud service. But Skinner’s analysis directs attention to the wider enterprise requirements behind a dependable deployment: preparing and cleaning data, rebuilding or adapting data architecture, cloud infrastructure, GPUs and ongoing operational support.
For a regulated financial institution, leaders should also consider the work required to integrate AI into existing processes and controls. A model is not a finished business service simply because it can generate an answer. It needs appropriate data access, governance, monitoring, security, user training and clear accountability for how outputs are used.
The implication is not that banks should avoid AI. It is that they should avoid treating an initial pilot cost as the total cost of ownership. An AI business case should separate one-off implementation spending from recurring costs, then test how both may change as usage scales.
A broader AI business case for financial services
A more useful AI ROI banking framework asks whether a use case improves measurable business outcomes across several dimensions. The precise measures will vary by function, but leaders can assess:
- Productivity and capacity: Are teams completing defined tasks faster, with less rework or with more time for complex cases?
- Speed: Has the time to respond to a customer, resolve an exception or complete an internal process changed?
- Service quality: Are answers more consistent, accessible or useful to customers and colleagues?
- Risk and control: Does the application help identify anomalies, document decisions, escalate uncertainty or reduce process errors?
- New capabilities: Does AI make possible a service, insight or workflow that the bank could not reasonably deliver before?
- Total cost of ownership: What are the data, infrastructure, compute, integration, governance and maintenance costs over time?
These categories do not guarantee a return. They create a disciplined way to test it. They also make trade-offs visible: a use case with modest labour savings may still justify investment if it materially improves speed, control or customer service at an acceptable ongoing cost.
Questions executives should ask before scaling
Before approving a scaled AI deployment, a CFO, CIO or programme owner can ask a small set of sharper questions:
- What specific business decision, workflow or customer problem is this use case improving?
- What baseline will we use to measure time, quality, error rates, risk events or service performance?
- Which benefits are cashable savings, which are capacity gains, and which are strategic capabilities?
- What data preparation, integration and governance work is required before the tool can operate reliably?
- How will compute, cloud and support costs behave if usage grows?
- Who owns monitoring, human oversight and remediation when outputs are incomplete or wrong?
- At what point will we review the case and decide whether to expand, redesign or stop?
These questions shift the conversation from AI as a technology purchase to AI as an operating-model decision. They encourage leaders to make assumptions explicit before a pilot becomes a costly programme.
A useful agenda for banking and fintech events
For banking conferences, executive offsites and transformation forums, Skinner’s perspective offers a focused discussion beyond AI hype. A session can help audiences examine the economics behind their own portfolios: where value is genuinely being created, what costs are being overlooked and how governance and infrastructure affect the final result. It is particularly relevant to mixed audiences of finance, technology, operations, risk and people leaders, who must evaluate the same investment through different lenses.
To enquire about Chris Skinner for a conference, leadership meeting or corporate event, contact Speakers Connect at info@speakersconnect.com.

