Every enterprise executive is sitting with the same decision right now. Here's a framework for thinking through it honestly.

Somewhere in the last six months, the question stopped being "should we invest in AI" and became "how do we know if our AI investment is working."

That shift matters. It means the early-adopter phase is over. Boards are asking. CFOs are modeling it. And the honest answer — the one most AI vendors won't give you — is that for a large portion of enterprise AI spending right now, the ROI is genuinely unclear.

This isn't a cynical take on the technology. The technology is real. The productivity gains at the individual level are documented and significant. But there's a difference between a technology being powerful and a technology being deployed in a way that produces measurable business outcomes at scale. Most enterprise AI rollouts are doing the first thing while hoping for the second.

What You're Actually Buying

When a large organization purchases enterprise access to a tool like Claude, what lands in the hands of employees is a capable, general-purpose AI assistant. It can draft communications, summarize documents, analyze data, prepare briefings, and assist with dozens of other tasks that previously required more time or more people.

That's genuinely valuable. The problem is that it's valuable in a way that's almost impossible to aggregate into a business outcome.

The value is created inside individual conversations, by individual employees, applied to individual tasks. It doesn't flow into a system. It doesn't accumulate. It doesn't produce an output that the next stage of a workflow can act on automatically. Each employee's use of the tool is essentially a private productivity gain — real, but contained.

The Measurement Problem

Individual productivity gains from AI tools are notoriously hard to measure at the organizational level. You can survey employees and ask if they feel more productive. You can track license utilization. What you generally cannot do is draw a clean line from "we gave everyone Claude" to "revenue increased" or "operating costs decreased by a specific amount."

The investments that produce clear, measurable ROI are the ones that target a specific workflow, eliminate a specific cost, or accelerate a specific outcome. Those are not general-purpose deployments. Those are designed systems.

Two Different Bets

There are really two different ways a large organization can invest in AI, and they're not mutually exclusive — but they're also not the same thing.

The first bet is on the individual. You equip your workforce with AI tools and trust that capable people will find valuable ways to use them. This bet pays off in ways that are real but diffuse. It raises the floor on individual output.

The second bet is on the system. You identify specific workflows where the cost of the current process is measurable. You build AI into the architecture of that process so that the output is consistent, the handoffs are automated, and the improvement compounds over time. This bet is harder to make and slower to deploy, but the ROI is traceable because you designed it to be.

Most enterprise AI spending right now is the first bet, framed as if it were the second.

What the Honest Question Actually Is

For an executive signing off on a significant AI investment, the question worth sitting with isn't "can we afford this" or even "is this the right tool." It's a more fundamental one:

Are we buying capability, or are we buying outcomes?

The companies that will look back on 2025 and 2026 as the years they got ahead of this won't be the ones that gave everyone a license. They'll be the ones that used this window to redesign the workflows that were already costing them the most.

Startedby helps organizations understand where the real workflow gaps are and builds the infrastructure to close them. If you're working through this decision for your own organization, [start with a diagnostic](/diagnostic).