Beyond the AI Use Case

An organization can identify fifty promising AI use cases and still have no clear picture of how its business will operate once those applications are in place.

For leadership, AI value realization means establishing how AI use cases change the operating model and produce measurable business value.

Customer Service, Finance, Sales, and Operations may each identify useful AI applications. The executive question is what the business must deliver when those capabilities work together and how it will measure the result.

AI changes how work is performed, not the business obligation. Even when AI drafts a response or prepares a report, the business still needs reliable information, sound judgment, appropriate authority, and an accountable owner.

Capacity released by automation does not automatically become financial value.

In October 2026, Gartner reported that AI use cases increased by 400% between 2024 and 2025. In a survey of more than 2,000 technology leaders, 13% reported significant or transformational enterprise value from AI. These are different measures, not a use-case conversion rate, but they illustrate the distance that can exist between adopting applications and realizing business value.^[1]

We can investigate that distance by asking not only what AI can do, but what the business must be capable of delivering because of it.

Start With the Business Performance, Not the Technology

Consider an organization introducing AI to automate customer-service responses.

Customers may receive faster answers, employees may spend less time drafting messages, and the organization may handle more inquiries without a proportional increase in manual effort. Those are worthwhile possibilities. The business, however, is responsible for resolving the customer's need accurately and within an acceptable service and cost standard.

Generating a response is only one part of that obligation. If a payment must be corrected, a policy exception authorized, or incomplete information investigated, the operation must still get that work done. Customers experience the result of the complete service process, not just the performance of the AI application.

Start with the performance the business needs to improve:

• How long does resolution take from beginning to end?
• How much employee effort does the complete process require?
• How often is the need resolved correctly the first time?
• How often does the customer return because the response did not resolve it?
• What does an accepted resolution cost?

These measures exist whether or not AI is in the operation. Establishing them gives leadership a baseline and an outcome to pursue. The objective might be shorter resolution times, lower cost per customer, better service quality, or more inquiries resolved with the same capacity.

Then investigate what achieving that outcome requires: which activities AI can perform, what information it depends on, which employees and functions contribute, and who remains accountable for the result. That is how the organization defines its AI operating model around the business requirement.

AI may assume responsibility for an activity. The business must still account for everything required to deliver an acceptable outcome.

Give Me the Improvement in Measurable Terms

Consider a hypothetical organization handling 10,000 eligible customer inquiries each month. It currently requires an average of twelve minutes of employee effort to deliver an accepted resolution. After introducing AI, that falls to seven minutes, including human review, corrections, transfers, and exception handling.

That is a five-minute reduction per accepted resolution:

This calculation illustrates the capacity that could be created under these assumptions; it is not a forecast. Actual results would depend on observed volume, case mix, acceptance standards, and the full effort required to deliver an accepted resolution.

The 833 hours represent capacity created. They do not automatically represent 833 hours of financial savings.

Leadership must decide what to do with the capacity. If the organization has a backlog, it might redirect capacity to resolving more requests, reducing outstanding work and customer wait times. If the capacity removes a substantiated need for planned hiring, that may be cost avoidance. If the business reduces overtime or an existing expense, it may recognize a different financial benefit. Each outcome needs evidence.

The hours may be spread across employees, shifts, and types of work rather than forming a block that can readily be reassigned. The new operation also requires funding and capacity for information maintenance, exceptions, performance validation, improved instructions, technology, and integration.

The measure should cover the complete operation: work reduced, work changed, work added, capacity available, the management decision that converts it into value, and evidence that the value continues.

A productivity improvement tells us what became possible. The operating decision determines what the business actually gains.

The Human Capability Behind AI Performance

Who has the knowledge needed to make AI perform to the standard the business requires?

An experienced customer-service professional contributes more than the ability to compose a response. They understand policies, recognize exceptions, know which records to consult, and can tell when a decision requires additional authority. That experience is part of the business's operating capability.

Consider a simple instruction:

Prepare a response to this customer's refund request.

AI may produce a polished message in seconds. But does it know whether the transaction has settled, whether a refund was already issued, which policy applies, or when the decision needs an authorized person? Without the relevant business knowledge, a convincing response can still fail the customer and the operation.

The technology may execute the activity. The business expertise determines what successful execution requires.

The expertise behind the prompt

Prompt development means translating business knowledge into instructions that support an acceptable result. Someone must establish authoritative information, action conditions, acceptable outcomes, and how to handle incomplete or conflicting information.

That work depends on expertise, information quality, and an understanding of the operation. Owners must maintain and approve instructions, assess results, investigate repeated errors, and intervene when performance falls below the standard.

As products, policies, customer expectations, and exceptions evolve, the information, prompts, and controls supporting AI must evolve too.

What happens to the people who previously performed the work?

An employee may prepare fewer customer responses while remaining essential to establishing the requirements for acceptable responses at scale. The manual activity can decline even as the expertise remains necessary.

Research on AI assistance among 5,172 customer-support agents found an average productivity improvement of 15%, with meaningful differences by experience and skill. Less experienced employees generally benefited more; results for highly experienced employees were more mixed. The study does not prescribe an operating model for every organization, but it reinforces the need to examine how expertise and technology work together.^[2]

Not every AI-generated action requires manual review. Some activities can rely on established instructions, automated validation, quality testing, and exception monitoring; others require direct human approval. The controls should reflect the business obligation, and someone must remain accountable for the result.

Time spent maintaining information, testing results, investigating exceptions, and overseeing performance is part of the cost of sustaining the AI capability.

The Economic Questions Behind AI Value Realization

Measuring AI ROI requires more than counting hours saved. It requires evidence of the financial benefits realized and the full cost of sustaining the capability.

Separate productivity estimates can obscure shared dependencies: information specialists, employees who maintain instructions, technology, oversight, and support funded outside the function reporting the gain.

Leadership needs to see three connected effects:

1. What operating effort is no longer necessary to deliver the required result?
2. What capabilities, expertise, and operating capacity are now needed to sustain that result?
3. What management decision converts the improvement into measurable value?

The capacity released, the capacity required, and the value created

Productivity is not the same as reassigned capacity, avoided cost, or realized savings.

The next three questions come from Torie T. Thomas's 100 Questions Every Business Outcome Owner Must Ask. Questions 78–80 connect the economics of an operating change to the value actually realized and the work or cost that can safely be retired.

Question 78: Establish what changes in the economics of the business

Will this outcome reshape the business model, margins, cash, capacity, or cost structure?

The executive needs to establish which economic mechanism changes and which function or capability owns each driver. The Business-Model and Enterprise Economics Case connects the units of work, economic drivers, accountable functions, and conditions needed to convert an improvement into value.

AI performing work equivalent to hundreds of employees may be a substantial operating improvement. It does not prove equivalent positions or payroll were eliminated, or that the full cost of the new capability was counted. Each conclusion needs evidence.

The Business-Model, Unit-Economics, and Capacity-Conversion Case helps leadership assess whether to invest further, change the design, or expand the capability.

Question 79: Establish which value has actually materialized

What value-tracking method will distinguish realization, savings, avoidance, protection, and recovery?

Suppose the organization has demonstrated the 833-hour monthly capacity improvement. If leadership expects that capacity to avoid future hiring, the financial benefit still depends on whether the hiring need would otherwise have occurred and whether the capacity can absorb the work.

The Governed Value-State Ledger records the type of value, supporting evidence, accountable owner, and conditions for recognition. Operating improvement and potential economic benefit are not the same. A reduction may await a contract ending, or an improvement may erode as costs or exceptions increase.

The organization needs one defensible account of what has been created and what has actually been realized.

Question 80: Establish what can safely be removed

How will redundant systems, roles, or processes be sunset without weakening operations?

Automating an activity does not mean the previous arrangement can immediately be retired. Employees may still handle exceptions or maintain information, and a system may remain necessary for some transactions. Keeping old capacity indefinitely can also prevent savings.

The Safe-Retirement Gate and its Legacy Retirement and Value-Release Plan establish which activities, capabilities, and expenses can responsibly end, what responsibilities must transfer, who can accept the replacement, and what evidence supports the release. Eliminating a task does not necessarily eliminate its expertise, responsibility, or position.

Together, Questions 78, 79, and 80 address the economics of the operation, the value actually realized, and what the organization can safely stop doing.

The Opportunity Beyond the Use Case

Now return to the organization with fifty promising AI use cases. Imagine that they all perform as intended: service is faster, reports are prepared sooner, proposals require less manual effort, and analysis takes less time.

The business has achieved useful results. But have the applications changed the enterprise's operating capacity, responsibilities, and measurable performance in a way leadership can demonstrate?

One function's gain may depend on approval elsewhere; released capacity may require shared specialists; and two teams may report the same improvement. The enterprise needs a view of these relationships, not just a list of technologies.

S.C.A.L.E. supports this enterprise value-realization responsibility by connecting expected value with the performance, capabilities, commitments, and evidence required to realize it.

An AI Value Office can coordinate AI value operations by maintaining portfolio visibility, reconciling overlapping claims, and setting evidence standards. It cannot own business results for the functions producing them. Business Outcome Owners remain accountable for measurable contributions and the value they sustain.

Bringing It Together: A Retrospective Look at Klarna

Klarna's public reporting illustrates the difference between reported operating performance and the full economic case.

In February 2024, the company announced that its AI assistant handled approximately two-thirds of customer-service chats during its first month, about 2.3 million conversations. Klarna reported that customers resolved inquiries in under two minutes, compared with approximately eleven minutes previously, and that repeat inquiries fell by 25%. It described the assistant's workload as equivalent to approximately 700 full-time agents.^[3]

In its 2025 Form 20-F, Klarna reported that AI handled approximately 80% of customer-service chats during the year and performed work it estimated as equivalent to more than 850 full-time agents. The company also reported about $59 million in AI-related customer-service cost savings for 2025, comparable customer satisfaction based on internal measures, and continued availability of human customer support.^[4]

These are company-reported operating results and financial estimates. The 2024 and 2025 workload-equivalent figures use different reporting periods and calculation bases. They do not independently establish corresponding employee reductions or provide all the internal records needed to reconstruct the complete economic case.

What changed in the business operation?

The assistant changed how a substantial portion of service work was performed, but the obligation to resolve customers' needs accurately remained. Reported resolution times, repeat inquiries, and customer satisfaction offer evidence of performance. An executive investigation would also compare customer populations, total service effort, and the combined AI and human operation.

What happened to the knowledge and responsibilities behind the work?

At millions of interactions, maintaining accurate information, policies, instructions, and decision boundaries is a significant operating responsibility. A retrospective would investigate how those requirements were assigned and sustained. This is an illustrative line of inquiry, not a claim that Klarna failed to address them.^[4]

What did the operating improvement create, and how did it become value?

Workload-equivalent estimates and reported cost savings describe different things. An executive would examine the human effort changed, the decisions that converted it into measurable benefit, and the technology, information, specialist support, and ongoing costs needed to sustain the capability.

Questions 79 and 80 would then examine the evidence for each value claim and which previous activities or expenses could responsibly be retired without weakening customer service. Public reporting does not provide all the internal evidence to complete that assessment independently; this retrospective does not suggest that Klarna did not conduct one.

What does this mean for the larger enterprise?

Across reporting, sales, financial analysis, and operations, AI applications may share information, technology, expertise, and governance. They can release capacity in one function while increasing requirements in another. An enterprise view helps leadership resolve dependencies, direct investment, and support value claims.

Beyond the AI Use Case

A use case tells us what AI can do. A capability investigation tells us what the business must be able to deliver because of it.

The executive responsibility is to establish the operating requirements, define expected performance, assign accountability, and determine how results will be sustained and measured. If every AI use case succeeds, leadership should be able to explain more than what the technology accomplished: what the business can now deliver, how it will sustain that performance, and what value the enterprise gained.

The AI use case identifies an opportunity. The executive responsibility is to establish how the business will deliver, evidence, and sustain the result.