Why AI Consulting Projects Frequently Underperform in Financial Services — And How to Fix It

Financial institutions have poured billions into AI consulting engagements since 2023. Despite widespread adoption and numerous successful pilots, a significant number of these projects fail to deliver meaningful, sustained business value at scale.

This is not primarily a failure of technology. It is a failure of approach. The traditional consulting model — built around large transformation programs, extended timelines, and technology-first thinking — is poorly suited to the realities of regulated, legacy-heavy financial services organizations.

Why Most AI Consulting Projects Underperform

Here are the primary reasons AI initiatives in banking and insurance frequently fall short:

1. Overly Broad Scopes and Misaligned Incentives

Many engagements start with vague objectives such as “develop an AI strategy” or “drive AI transformation.” These are attractive to sell but difficult to execute. Consulting firms are often incentivized to propose large, multi-year programs rather than focused interventions with clear ROI.

2. Underestimating Data and Legacy System Complexity

Data quality, accessibility, and integration remain the most cited barriers across industry reports. Yet many proposals treat data readiness as a secondary issue. In financial services, where core systems can be decades old, this leads to projects that stall or deliver limited value once they move beyond the pilot stage.

3. Insufficient Early Governance Involvement

Risk, compliance, legal, and model risk teams are frequently brought in too late. This either slows projects dramatically or forces them to be significantly scaled back, eroding the original business case.

4. The Persistent Pilot Trap

A large proportion of AI consulting work ends at the proof-of-concept stage. While this reduces short-term risk for the client, it creates an illusion of progress. Many banks now have portfolios of successful pilots but very few production systems generating measurable returns.

5. Weak Organizational Change and Capability Building

Successful AI deployment requires new ways of working, new skills, and new decision-making processes. Most engagements significantly under-invest in the change management and internal capability components, leaving organizations unable to sustain or expand solutions after the consultants depart.

6. Lack of Focus on Measurable Value

Too many projects are measured by deliverables (reports, models, workshops) rather than business outcomes (reduced fraud losses, faster origination times, lower operational costs). Without clear outcome accountability, value realization often falls short.

What Actually Works

Projects that deliver strong results tend to share these characteristics:

• They begin with specific, high-value business problems rather than broad AI transformation mandates.
• They treat data readiness and legacy integration as first-class workstreams.
• They involve governance and compliance functions from the very beginning.
• They are designed with clear paths to production deployment, not just pilot success.
• They combine external expertise with deliberate internal capability building.
• They prioritize quick wins with measurable ROI while building toward larger transformation.

A Better Approach: How Specialized Finance AI Consultancies Like Skirr AI Are Addressing These Issues

Not all AI consulting is created equal. Firms that have adapted their model to the specific constraints of financial services are achieving better outcomes. One example is Skirr AI, a UK-based AI consultancy specializing in regulated sectors including banking and insurance.

Skirr AI’s approach directly targets several of the common failure points:

• Fixed-price, short-cycle AI Audits (typically one week) that deliver a prioritized, ROI-ranked roadmap instead of open-ended transformation programs. This reduces scope creep and aligns expectations early.
• Workflow-first discovery that maps real front-, middle-, and back-office processes to identify high-impact automation opportunities before any technology is recommended.
• Compliance and regulation built in from day one, with explicit consideration of requirements such as FCA, PRA, AML, and Consumer Duty. This helps avoid the common problem of governance bottlenecks later in the project.
• Pragmatic use of existing technology stacks — we review what the client already pays for (e.g., Microsoft 365 Copilot or core system capabilities) before recommending new tools. This addresses legacy constraints more realistically.
• Strong focus on measurable value by ranking opportunities according to impact, effort, and time-to-value, with examples such as fraud detection, regulatory reporting automation, document processing, and KYC/onboarding acceleration.
• 12 months of included post-engagement support, helping clients move from roadmap to governed production rather than leaving them after the initial phase.

By leading with discovery and governance rather than technology sales, and by focusing on deployable, regulated use cases instead of broad transformation, this type of specialized approach helps financial institutions avoid many of the classic pitfalls that cause AI projects to underdeliver.

Bottom line

The issue is not that AI doesn’t work in financial services. The issue is that many consulting engagements are still structured around outdated assumptions about how transformation happens in complex, regulated environments.

The organizations getting the best results are moving away from large, open-ended AI transformation programs and toward focused, outcome-driven initiatives that properly account for data realities, regulatory requirements, and the hard work of organizational change.

As the industry matures from the experimentation phase into the value realization phase, the consulting models that succeed will be those built around speed to insight, governance by design, and accountability for real business outcomes — not just deliverables.

Want to discuss how this affects your workflows? Book a call →