Most enterprise sourcing processes are built to compare. They produce a structured list of requirements, issue an RFP to a broad field of providers, score responses against a compliance matrix, and select the vendor that best matches the specification.
The problem isn’t the process. The problem is that in an AI-shaped market, the things that matter most don’t show up in a compliance matrix.
The Gap Between What Providers Say and What They Deliver
Across the IT services market today, virtually every major provider claims AI leadership. They reference automation rates, publish thought leadership, and highlight AI-enabled service desks in their proposals. From the outside, the claims are nearly indistinguishable.
In practice, the gap between what providers say about AI and what they actually deliver has never been wider. Some providers have genuinely restructured their delivery models around automation, built governance frameworks for AI-driven operations, and developed architectures that create real productivity outcomes. Others have applied AI capabilities to portions of delivery while leaving core operations largely unchanged.
For enterprises entering long-term managed services agreements, the difference between these two profiles is not marginal. It shapes cost trajectories, automation value capture, and operational flexibility for the entire contract term.
Why Static Evaluation Methods Can’t Surface the Difference
Traditional IT vendor evaluation is designed for a market where providers compete on similar dimensions, including scale, geographic coverage, labor models, and pricing structures. When the key differentiators are well-defined and relatively stable, a structured RFP with predefined criteria works reasonably well. AI delivery capability doesn’t behave that way.
It isn’t a feature that can be checked off a list. It reflects the depth of a provider’s automation architecture, the maturity of their AI governance, how they approach Human-in-the-Loop oversight, and how they plan to evolve capability over the life of a multi-year engagement. None of that is visible through proposal language alone.
Organizations that evaluate AI capability through self-reported claims are essentially asking providers to grade their own work and select based on the result.
Architecture-Led Evaluation as a Better Standard
Windsor Group’s AI Engagement Model takes a different approach. During the Architect phase, pre-qualified providers receive the client’s actual operational baselines — AI readiness, service performance, and cost position — and are asked to design a real solution against that environment.
This means providers must produce their proposed operating model, AI and automation strategy, transition path, and commercial structure. They aren’t responding to a generic specification. They’re designing against a real client’s real situation.
The design sessions give organizations direct visibility into several dimensions that RFP responses cannot: the depth of a provider’s AI architecture, how they think about automation governance, how they approach transition risk, and how their commercial model is structured to evolve over time.
Cultural fit and governance alignment also surface naturally in this process — in ways that proposal documents rarely reveal.
What Strong AI Delivery Capability Actually Looks Like
Through the design process, meaningful differences emerge quickly between providers who have genuinely built for AI-centric delivery and those who have not.
Providers with mature AI capability tend to design with specificity. They can articulate where automation applies, how governance works, what the transition architecture looks like, and how their commercial model reflects automation economics over time. They treat AI not as a feature to highlight but as the foundational operating logic of the delivery model.
Providers with less mature capability tend to design with generality. They reference AI in the context of specific tools or initiatives, often without demonstrating how those capabilities integrate into the overall delivery architecture or evolve across a multi-year engagement.
For an AI sourcing strategy to work effectively, the evaluation process has to be designed to surface that difference, not assume it will appear on its own.
The Evaluation Standard Has Changed
For enterprises pursuing infrastructure sourcing in today’s market, an IT vendor evaluation that doesn’t specifically assess AI delivery maturity is an incomplete process. The cost of selecting a provider without genuine visibility into that dimension isn’t just measured in the first year of the contract. It compounds across the full term, in missed automation opportunities, in pricing structures that don’t reflect market economics, and in governance models that aren’t built for how AI-driven operations actually work.
The standard for evaluation has changed. The methodology used to execute it needs to change with it.