Why AI projects stall between pilots and proof

89% of organisations across asset, auto and equipment finance are already piloting, scaling or embedding AI. Only 14.5% report significant, proven ROI.

That is the central tension in the AI Adoption, Workforce Redesign and Operational Readiness report from Resilient Management Solutions and Finance Connect.

The sector has crossed the experimentation threshold. It has not yet crossed the evidence threshold.

Understanding why that gap exists, and what it takes to close it, is the most practical question facing leaders right now.

The three blockers

The report asks respondents directly about the biggest barriers to ROI. Three answers dominate.

The first is the absence of a clear business case or metrics, cited by 41.8% of respondents, indicating that this is a design problem, not a technology problem. Many AI programmes begin with a proof of concept that demonstrates technical feasibility, then stall because nobody has defined in advance what success looks like, how it will be measured, or who is accountable for the outcome.

The second blocker is limited internal expertise or resources, cited by 36.4%. Knowing that AI can help and knowing how to turn that into an operational reality are different capabilities. The people who can bridge domain knowledge, platform understanding, change management and governance are scarce. Without them, pilots often stay as pilots.

The third is data quality and system integration, at 34.5%. In asset finance, data lives across originations platforms, legacy servicing systems, third-party providers and manual processes. Getting clean, consistent, connected data into an AI workflow is rarely straightforward, and underestimating that work is one of the most common reasons value gets delayed.

The attribution problem

There is a fourth dynamic the report identifies that sits beneath all three blockers. Even where AI is technically working, isolating its contribution to business outcomes is hard.

Workflows in asset, auto and equipment finance cut across departments, systems and counterparties. When time-to-fund drops, or NIGO rates fall, or collections cure rates improve, attributing that improvement cleanly to an AI intervention is difficult. And if you cannot attribute it, you cannot build the board-level business case to scale.

This is where many programmes get stuck. The operational pain is understood. The technology is deployed. But benefit tracking is immature, and that makes the next investment decision harder than it needs to be.

Compliance friction compounds the gap

Regulatory requirements add another layer. In lending environments, Consumer Duty, fair lending expectations and model explainability requirements do not disappear because AI is involved. They intensify.

Technical feasibility does not equal operational value when mandatory human oversight, audit requirements and governance obligations slow the path to scale. That is not a reason to avoid AI in credit, decisioning or risk processes. It is a reason to design for those constraints from the outset rather than treating them as late-stage considerations.

Aysha Ellis-Aziz, Managing Director UK at Teamwill, captures what this means in practice:

“The ambition is there, the pilots are running, but the gap between proof of concept and operational reality is where projects stall. Closing that gap requires more than technology. It requires specialists who combine genuine asset finance domain expertise with hands-on platform knowledge.”

What crossing the threshold actually requires

The firms that move from pilot to proven ROI tend to share a few characteristics.

They define success metrics before deployment, not after. They start in areas where value attribution is cleaner, typically document-heavy workflows where time, error rates and rework hours are all measurable. They treat integration and data readiness as part of the programme scope, not a dependency to resolve later. And they build internal capability alongside the technology, rather than expecting the tool to carry the programme on its own.

Philip Benke, Growth and Innovation Director at CGI, frames the underlying challenge clearly:

“AI won’t transform outcomes unless businesses redesign how work actually gets done. Those who do will move faster from pilots to real ROI, while others risk adding more technology without changing their work.”

The report makes clear that the question of whether to adopt AI has already been answered across asset, auto and equipment finance. The question now is whether organisations can build the measurement discipline, integration depth and internal capability to turn activity into evidence.

That is the work that separates the 14.5% from the 89%.

Download the AI Adoption, Workforce Redesign and Operational Readiness report here.