AI Adoption and ROI
How to Measure AI Adoption and ROI Beyond Usage Metrics
The Adoption Illusion examines the work behind the AI success story, and why that work matters to business leaders, customers and investors.
By Christine Barnett. Published .
There is a peculiar accounting problem at the heart of an AI success story. The system produces something quickly, so the business records a saving. The people who check it, correct it and make it usable are harder to track, so their contribution may never reach the same calculation. By the time the result arrives in a management presentation, the machine appears rather more self sufficient than it did on Tuesday afternoon.
Measuring AI adoption means examining how the technology changes a real workflow. Measuring AI return on investment means establishing whether the benefits of that change justify its full costs. Usage figures can help with both investigations, but they cannot settle either one.
This is the problem I explore in my forthcoming book, The Adoption Illusion, and through the Autonomy Index™ framework.
Why the AI investment story needs better adoption evidence
In Private equity and the illusion of diversification, published on 1 June 2026 as Future Standard partner content on FT.com, the argument turns from the concentration of investment in large businesses towards the potential of the middle market. Its AI section suggests that opportunities could extend beyond infrastructure providers to companies applying AI in their own operations.
The distinction matters. A business does not have to build an AI model to benefit from one. However, identifying a potential beneficiary requires evidence of what happens after the software arrives.
My argument is that this creates an adoption measurement problem. Two businesses might buy similar tools and report similar levels of activity while achieving very different results. One might redesign an entire process around a reliable system. The other might add generated output to an already congested chain of approvals. Their purchasing decisions could look alike; their operating economics would not.
What is the adoption illusion?
The adoption illusion is the appearance of AI progress without sufficient evidence that the underlying work has improved. It develops when access, experimentation or output volume is treated as proof of business value.
A team generating more material may be more productive, or it may have created a larger review queue. A customer using a feature regularly may depend on it, or may still be trying to make it useful. The distinction becomes visible when someone follows the work beyond the point at which the AI produces its answer.
Illustrative example: hypothetical data
AI reduces the time needed to prepare a first draft of a customer report, but employees must still reconcile the source data, correct explanations and move the report between systems. The first draft is faster. To establish whether the full process has improved, the organisation must include those remaining steps, the quality of the final report and the time taken to deliver it.
Assistance can still justify the investment. The calculation simply needs to describe what the company is actually receiving.
What Autonomy Index™ measures
I developed Autonomy Index™ to make the relationship between AI activity and operational change easier to examine. In The Adoption Illusion, four reader facing questions organise this explanation:
Adoption
Is AI use consistent and embedded across the relevant work and teams?
Capability Enablement
Which of the technology’s capabilities are actually active in the workflow?
Workflow Ownership
How much of the process does AI handle, and where does human intervention remain necessary?
Strategic Impact
What evidence connects the change to capacity, customer value, revenue, margin or reduced risk?
These are the book’s explanatory pillars. The formal Autonomy Index™ assessment uses the four weighted pillars and eight dimensions set out in the published methodology.
The framework supports a more precise conversation about adoption. A workflow ownership assessment is not, by itself, a financial return calculation. Greater autonomy also needs to be considered alongside output quality and appropriate oversight. In sensitive work, a carefully designed assistive system may be the right outcome.
The accompanying Autonomy Quality Matrix™addresses that relationship between autonomy and quality. Giving a system more responsibility only makes sense when the work it produces is reliable enough for that responsibility.
How to measure AI ROI at workflow level
Begin with a process that has a recognisable beginning and end, such as resolving a customer request or producing an approved report. Establish its baseline: the time involved, cost, error rate, review effort and quality of the completed result. Then compare the AI enabled process against that baseline using comparable work.
Include implementation, integration, software, training and ongoing oversight in the cost assessment. Review and exception handling belong there too. Where AI releases staff time, record what happens to that capacity. It may support more work, better service or less overtime; it does not automatically become a cash saving.
Attribution also matters. A result that improves after an AI rollout may have benefited from cleaner data, a simpler approval process or additional staffing. Those changes can be valuable parts of the programme, but the explanation should show how they contributed rather than allocating every improvement to the model.
Five questions for an initial review
- What can now be completed more effectively, from start to finish?
- How much human effort remains, including correction and supervision?
- Does output quality hold across ordinary cases and exceptions?
- Who is accountable when the system needs help or gets something wrong?
- What measurable benefit remains after the full costs are included?
These questions turn a claim about transformation into something a team can inspect and improve.
Why this matters for customer retention
For an AI vendor, adoption measurement also shapes the renewal conversation. A customer needs a credible account of what the product contributes, especially when budgets are being reviewed or an internal champion leaves.
Customer Success teams are well placed to uncover the gap between an attractive feature and an established way of working. That work becomes more valuable when it leads to clearer responsibilities, better handoffs and evidence the customer can use internally. In the book, I call this discipline Adoption Engineering.
It gives founders and delivery teams a practical starting point: choose one workflow, establish what is happening, redesign what needs to change and observe the result before expanding the programme.
About The Adoption Illusion
The Adoption Illusion draws on my work across Customer Success, SaaS adoption and AI operations. Through anonymised and composite examples, it examines why promising tools can struggle to become dependable parts of a business, and introduces frameworks for making adoption more measurable and deliberate.
The book is written for founders, Customer Success leaders and teams responsible for AI implementation and operational change. Its central concern is what happens after the purchase, when the organisation has to turn a product’s capabilities into work people can trust.
As attention broadens towards the businesses using AI, that question becomes commercially significant. An impressive demonstration can earn a meeting. A process that consistently delivers a better result gives the business something to keep.
The Adoption Illusion by Christine Barnett will soon be available to buy.
Last reviewed .
Ready to measure what AI actually changed?
Take the Autonomy Index™ Assessment and receive an executive report covering workflow ownership, capacity released, AI debt and prioritised recommendations.