Ology

The Enterprise AI Return: From Useful Work to Durable Value

Frameworks for evaluating enterprise AI investments, measuring returns, costs, risks, and organizational readiness

5 sessions · 10 readings · 3 views

Curated by Ned Horowitz

The Outcome Ledgersession 1Proof Before Scalesession 2Context Is Capitalsession 3The Moving Cost Curvesession 4Productivity With a Half-Life?session 5

Opening

Enterprise AI is moving from chat assistants toward systems that act across tools, files, and multi-step workflows. That changes the investment question: a seat, prompt, or token is no longer a useful stand-in for work completed. The relevant unit may be a resolved case, a tested code change, a better decision—or a new service the organization could not previously offer.

But measuring those outcomes is difficult. A faster task may still need extensive review; time saved may turn into more output rather than lower payroll; and usage data can show where AI is being tried without proving that it improves the business. Even “successful task” depends on how success is defined and verified.

Meanwhile, the economics are moving too. Falling unit prices can make new uses practical, while agents’ longer runs and higher volumes can raise total spend. Organizational context, permissions, training, and workflow redesign may matter as much as model capability. The seminar follows those tensions from measurement to deployment and cost control, then asks whether short-term gains endure. What evidence should convince an organization that its AI investment is creating value—and that the value will last?

Five questions worth arguing about

  1. 1When saved time is reinvested rather than removed from payroll, which evidence should count as enterprise AI return?
  2. 2How much evidence is enough to scale a workflow when benchmark reliability and real-world value diverge?
  3. 3Are weak AI returns primarily a model problem, or a failure to assemble context, incentives, and operating change?
  4. 4How should leaders allocate capital when falling unit prices can coincide with rising usage and total spend?
  5. 5Can near-term output gains count as durable returns if AI changes expertise, judgment, and who captures the gains?

The Sessions

1Session 1start here

The Outcome Ledger

The problem

AI investments are often justified with time saved, adoption, or a projected financial return. But none of those measures alone establishes that an organization is getting useful work done. OpenAI’s scorecard argues for measuring completed outcomes and the full cost of successful tasks. Google Cloud’s framework takes a more conventional business-case approach: define value drivers, account for total cost of ownership, then calculate ROI.

The difference matters. Google Cloud’s customer-service example makes the business case tangible, but its value figures are estimates, not evidence of a controlled, realized return. The scorecard raises a harder accounting question: if AI frees people to do more valuable work rather than reducing headcount, what should count as the return? That ambiguity leads directly to the next session: how can teams tell whether the promised outcome is actually happening?

Readings

  1. A scorecard for the AI ageSarah Friar
    Open ↗
  2. Three-part framework to measure the impact of your AI use caseEva Dong and Pathik Sharma
    Open ↗
2Session 2Proof Before Scale2 readings3Session 3Context Is Capital2 readings4Session 4The Moving Cost Curve2 readings5Session 5Productivity With a Half-Life?2 readings

Next Up