In the May/June issue of Monitor, The Alta Group examined one of the most immediate risks created by generative AI inside equipment finance organizations: the temptation to rely on AI-generated language in credit narratives, board materials, customer communications and policy drafts without a disciplined process for validation. The article’s core message was not that lessors should avoid generative AI, but that they should govern it, matching the depth of human review to the risk and consequence of the output.
That principle does not stop at the edge of
an organization’s own workflow. It extends to every AI capability a lessor is asked to buy into, often for the next decade. If “verify before you trust” is the right posture when using AI to draft or analyze, it is equally important when evaluating the AI claims embedded in a software vendor’s RFP response, product demo or implementation roadmap.
Nearly every platform provider in equipment finance can describe some form of AI-enabled functionality. But the label can mean very different things: true machine learning, rules-based automation, predictive analytics, workflow orchestration, document intelligence or a chatbot interface layered onto a process that has not fundamentally changed. For technology committees evaluating a platform, the question is no longer whether a vendor mentions AI. It is whether the claimed capability is real, relevant, explainable, already in production and appropriate for the lessor’s own portfolio and operating model.
A Market Moving Faster Than Buyers Can Verify
The pace of AI-related investment in leasing technology is not simply a marketing trend. It reflects real capital allocation. When private equity firm Diversis Capital acquired LTi Technology Solutions in March, the firm’s leadership described the opportunity in terms of product innovation and bringing AI-powered capability to LTi’s client base. That kind of statement is not unique to one platform or one transaction. Across the vendor landscape, AI investment theses and product roadmaps are becoming inseparable from the pitch a lessor hears in an RFP process.
This creates a genuine dilemma for buyers. Platform decisions in equipment finance carry long tails, often five to 10 years from selection to replacement. A lessor that gets the AI assessment wrong is not making a minor mistake correctable in the next product cycle. It is living with that decision, and its consequences, for a very long time. Yet most technology committees do not have the internal expertise to independently evaluate a vendor’s AI architecture the way they would evaluate core lending logic or integration capability. The result is that AI claims are too often taken on faith, evaluated by demo polish rather than by the same diligence rigor applied to everything else in a platform decision.
Five Questions Worth Asking Before You Believe the Claim
Governance, not avoidance, is the right response. The following questions give a technology committee a practical way to separate substance from positioning, without requiring a data science background.
Is this actually machine learning, or is it automation with a new label?
Ask the vendor how the system’s outputs change or improve as more data flows through it, versus following a fixed set of configured rules. A capability that behaves the same way today as it did a year ago, regardless of volume or pattern, is likely automation wearing an AI label. A vendor confident in its own architecture should be able to explain this distinction clearly and specifically, not in marketing language.
What data was the model trained on, and is it relevant to your portfolio?
A model built on generic commercial lending data will not necessarily understand equipment specific dynamics such as residual value behavior, asset-class default patterns or secondary market liquidity. Ask for specifics: which asset classes are represented, how many transactions are included, how recent the data is and whether it reflects your market segment or a broader one.
Where exactly is the human in the loop, and where is there none at all?
This is the same principle the earlier article applied to internal generative AI use: match the depth of human review to the consequence of the output. Ask the vendor to walk through, step by step, where the AI’s recommendation ends and human judgment begins for credit decisions, valuations and servicing actions. Ask whether that boundary is configurable by your organization or fixed by the vendor’s design.
“There’s a tendency to treat AI as a straight replacement for automated business processes, when in practice the strongest outcomes come from combining the two,” Andrew Flegg, chief technology officer at Alfa says. “Regulated operations still require deterministic controls, clear audit trails and consistent policy enforcement. The most effective architectures use AI to assist people and accelerate decision-making, while relying on business systems and workflow to enforce policy, manage risk and deliver consistent outcomes. This layered approach allows organizations to realize the benefits of AI without sacrificing control.”
What is the actual track record, versus the roadmap?
A vendor recently backed by new capital with a stated AI thesis, as in the Diversis and LTi example, may be describing where the platform is headed as if it already describes where the platform is today. Ask specifically which AI capabilities are live with paying clients right now, request reference clients using that specific feature, not the platform in general, and separate what you are buying today from what you are being asked to believe about tomorrow.
“Production versus roadmap is a good question to ask, and any buyer should know what’s actually live with competitors who look like them,” Jennifer Martin, chief strategy officer at LTi Technology Solutions, says. “But in a decision you live with for the better part of a decade, the feature list you can verify today is the least durable thing you’re buying. What really determines whether you capture AI value over that horizon is how quickly a vendor can keep turning your data into new capability, and how ready your own organization is to change the way it works to use it. The best AI diligence I’ve seen is as much a mirror as a microscope.”
How does the system handle its own uncertainty?
A well-designed AI capability should be able to flag when its confidence is low, rather than presenting every output with the same authority. This is precisely the failure mode the earlier article warned about: AI-generated content that sounds definitive but has no grounding in verified fact. A system that can say, in effect, “I am not confident here; a human should look at this,” is architecturally more trustworthy than one that cannot.
What Disciplined Diligence Looks Like in Practice
Lessors handling this well share a few common habits. They require written, specific answers to questions like these as part of formal RFP scoring, not just verbal reassurance during a product demo. They ask for reference calls focused specifically on the AI feature in question, rather than the platform relationship generally. And they build a pilot or proof-of-concept period into the selection process, to validate AI claims under real conditions before signing a long-term contract.
Cory Damm, chief marketing officer and chief technology officer of Elevex Capital, a lender actively building and buying platforms while weaving AI into its own operations, puts it directly: “AI claims are cheap to make and expensive to get wrong on a 10-year contract, so we ask vendors to show us the feature working today, not on a roadmap. But the real question underneath that is build versus buy — it has never been easier to build on the same models a vendor is using, and their implementation timeline often rivals doing it ourselves. If a platform’s AI is going to be sold to every competitor we have, we need to know it is earning its place over building it in-house, not just checking a box.”
That last point is worth sitting with. A capability every vendor sells to every lessor in the market is not, by definition, a competitive advantage for any one of them. The diligence question is not only whether the AI works but whether buying it beats building it or simply doing without it.
None of this requires a lessor to become skeptical of AI itself. The capability taking shape across the industry, from document intelligence to workflow-embedded recommendations to more natural interfaces for customers and internal users, is real and worth pursuing. But real capability and real diligence are not in tension. If anything, the more central AI becomes to a platform decision, the more that decision deserves
the same rigor a lessor would apply to core lending logic, integration architecture or long-term total cost of ownership.
Governance Does Not Stop at Your Own Workflow
The earlier article’s central message was governance over avoidance: generative AI can help an organization think faster, but it cannot determine what is true, and the responsibility for accuracy stays with the people using it. The same is true on the buy side. A vendor’s AI claim, however polished the demo or however credible the funding behind it, does not carry inherent authority. It carries whatever authority a lessor’s own diligence process gives it.
The vendors making the boldest AI claims deserve the most scrutiny, not the least. Equipment finance companies that build that discipline into their platform selection process now will not be the ones surprised, three years into a 10-year contract, by the gap between what was promised and what was actually delivered.
