Given the progress of AI, asking “What’s your AI roadmap?” is standard in vendor reviews. But buyers won’t get the answers they need unless they can evaluate whether a provider can stand behind a defined outcome, document how it will be delivered, and accept accountability when it falls short. That’s a different approach from assessing vendors based on the capabilities they show you.

The gap is that most existing procurement processes aren’t yet structured to probe at that level. Service-level agreements still describe system availability rather than the delivery of results. The spotlight shifts to finding evidence of result delivery and outcome reliability at comparable firms.

Consider how a portfolio manager gets answers today. When a strategy underperforms its benchmark, finding out why — pulling attribution data, reconciling across systems, cross-referencing position changes — can take hours of manual work. In a service-as-a-software model, the platform doesn’t surface the data and wait for a human to analyze it. It delivers the explanation: what drove the underperformance, what the firm’s historical response has been in similar situations, and what options are available. The provider is accountable for the quality of that output, not just for the uptime of the system that produced it. That accountability becomes the focal point for how you evaluate vendor fit. The workflow alone is just table stakes.

What changes inside the firm

If you’re buying outcomes rather than tools, something has to change inside your firm as well. Outcomes shed light on new decisions your organization could make about where to direct its operational investment and its people.

The clearest implication is for technical resources. Post-trade operations are defined and measurable — which means they’re also the area where the efficiency gains from agentic AI are most legible and most auditable. Firms that redirect technology investment toward front-office AI — research, portfolio construction, alpha generation — are finding that the operational foundation underneath can carry more of its own weight than it used to.

When agents handle more defined, repeatable operational work, operational roles can change as well. As high-volume, rules-based processing becomes more automated, the operational professionals who understood that work become more consequential. The judgment about whether the model is right, whether an exception is truly an exception, and whether the operational model is keeping pace with portfolio strategy requires someone who has done the work. That expertise moves up.

The big-picture result here is that end-user expertise becomes more consequential when you move from users looking at data to agents recommending or carrying out specific operational actions. It’s true for most roles in your organization. We wouldn’t bet on AI taking those skills out of the equation.

The domain depth question

There’s a reason why the recent memo from Citrini Research on AI-driven software displacement created such a flurry and moved markets. It raises a fundamental question: Which software is actually at risk?