From Idea to Operating Case: Building the Business Case for Automation-to-Autonomy

Make the operating logic part of the investment decision

A rigorous business case already considers more than financial return. HM Treasury’s Green Book, updated February 5, 2026, describes a management case covering implementation, governance, change management, benefits realization, risk, and evaluation. Operating discipline belongs inside established investment practice. HM Treasury: The Green Book, Five Case Model.

I use operating case to describe the explicit connection between the investment forecast and how the business will work. It can sit within the existing approval document. Its purpose is to expose the assumptions that determine whether the return is achievable.

Start with the outcome and its current constraint. Faster document preparation has limited value if an unchanged approval queue still determines when customers receive a decision. Removing a handoff may create more value than accelerating it.

Compare credible alternatives against the same outcome, baseline, and time horizon. Include the current approach, process simplification, conventional automation, and AI where appropriate. This is where the problem-selection work in The Value Lens becomes a specific investment choice.

When proposing autonomy, explain what additional value comes from allowing the system to act. Then account for the additional assurance, monitoring, and recovery it requires. A useful recommendation tool may be the right destination for a particular decision.

Explain how the benefit will become real

Every benefit in the forecast needs an operating change that makes it possible and an owner who can realize it.

Released time creates usable capacity when it can be consolidated or reassigned. That capacity produces value when needed work gets done or service improves. Cash savings arise when an identifiable expense falls. Better service becomes measurable when customers receive a more timely or reliable outcome. These benefits deserve separate measures.

Research illustrates why a productivity finding cannot simply become a standard savings assumption. In a study published in February 2025, Brynjolfsson, Li, and Raymond examined 5,172 customer-support agents during an AI-assistance rollout concentrated in late 2020 and early 2021. Access to the assistant increased issues resolved per hour by 15% on average, with substantial differences across workers. This was evidence from one firm’s support operation and its contractors, where people retained authority over responses. Generative AI at Work.

My implication for investment reviews is straightforward: use external evidence to shape a local test. It cannot establish your realizable savings or justify autonomous action in a different workflow.

The Outcome owner commits to the accepted result, its service and quality thresholds, and the resources needed to achieve it. The Operating owner changes assignments, queue priorities, staffing allocation and handoffs to make that commitment workable. If a service promise or downstream team must change, the person with that authority must agree; the Sponsor resolves barriers beyond the owners’ remit. Redesigning Work Around Outcomes develops this connection between the result and the work.

If the aim is to absorb demand without additional hiring, identify the demand forecast, the staffing decision, and when the avoided expense would otherwise arise. If the aim is better service, track completed outcomes, waiting time, and rework. Finance challenges the benefit evidence and its financial interpretation; the Outcome owner remains accountable for realizing the result.

Workforce design is part of this logic. Involve the people doing the work in identifying hidden checks and difficult exceptions. Budget for learning and changes to responsibilities. If routine work falls away, the remaining workload may require more judgment per case; staffing it solely from average handling time can leave the operation exposed at peak demand.

Give those people time and authority to challenge outputs, resolve exceptions, and improve the process. A benefit forecast that assumes this contribution must also resource it.

Price the operation that will deliver the return

The relevant cost is the cost of delivering the outcome over the chosen period. Include preparation and integration, knowledge maintenance, user transition, human review, technology consumption, support, evaluation, and incident recovery.

Separate initial expenditure from recurring costs, and show when benefits are expected to begin. Test a downside case with slower adoption, fewer eligible cases, and more review effort. Those assumptions should change the forecast rather than sit in a risk appendix.

The control model also changes the economics. If a system needs authoritative policy content to work, maintaining that content is an operating expense. If it may take consequential action, monitoring and intervention capacity are part of the service being funded.

NIST’s voluntary AI Risk Management Framework includes post-deployment monitoring, user feedback, override, incident response, recovery, and decommissioning. These are useful reference points when estimating what continued operation requires. NIST AI RMF 1.0, MANAGE 4.1, January 26, 2023.

Shared foundations deserve explicit treatment too. Where several initiatives need the same knowledge service or monitoring capability, show the common investment and how its costs are allocated. Record each dependent initiative’s incremental costs and benefits. Count the shared cost once in the consolidated portfolio, and distinguish committed reuse from speculative future demand.

For the running operation, use Cost per accepted outcome:

Total recurring cost of running the capability in a period ÷ outcomes accepted by the owner in that period.

Include technology and usage, human review and correction, exception handling, support and recovery. Failed and repeated attempts count in costs, not as additional accepted outcomes. Keep one-time build costs in the investment case, outside this operating metric. Match scope and measurement periods, including the acceptance follow-up. With zero accepted outcomes, the metric is not calculable.

A worked example: invoice exceptions

Hypothetical illustration. All figures below are assumptions for testing, not measured results or a forecast for a particular organization.

Suppose a finance operation handles 400 invoice exceptions each week, averaging 15 minutes of staff effort per case. The objective is to reduce the aging backlog while preserving payment controls.

The proposal would assemble relevant records and approved policy guidance for analysts. People would continue to decide resolutions; payment approval and supplier bank-detail changes would remain outside the system’s authority.

Assumption	Implication
Current work: 400 cases × 15 minutes	100 staff hours per week
240 cases are suitable for assistance and use it	60% of weekly volume
Assisted cases take 9 minutes, including checking and corrections	6 minutes potentially released per assisted case
Remaining 160 cases stay at 15 minutes	40 hours of unchanged work
Additional monitoring and knowledge maintenance require 4 hours weekly	Proposed total effort becomes 80 hours per week
Net reduction: 100 − 80 hours	20 hours of potential weekly capacity

The arithmetic is conditional on suitability, actual use, and review effort. If only 120 cases benefit, with the same time reduction and overhead, the potential gain falls to eight hours per week. Test those assumptions before annualizing them.

Suppose the analyst team remains the same size and its employment costs do not change. The table supports a capacity hypothesis. A cash-saving claim would need a separate change in expenditure.

The Operating owner for finance operations must decide how that capacity will reduce backlog, then measure whether it does. A trial should compare similar cases with and without assistance, including total handling time, resolution quality, repeat work, backlog age, and staff review effort. Match the comparison for case complexity and workload; include peak periods and difficult exceptions.

No payback claim follows from this illustration: implementation costs, ongoing technology costs, benefit timing, and any realizable cash benefits still need to be established.

The result may justify wider use of assistance. Delegating exception clearance would require additional evidence about decision quality, boundaries, and recovery, alongside a fresh assessment of value.

Following the capacity into an accepted outcome

Extend the illustration with a six-week reporting period. All costs and results that follow are hypothetical. Routine arrivals remain at 400 a week. Separately, 400 aged exceptions await resolution because purchasing confirmations arrive too slowly. The Outcome owner judges this overdue service a priority: suppliers need reliable decisions about their invoices. Before starting, the owner agrees to spend up to an additional $2,400 over six weeks to secure at least 120 accepted aged resolutions, while maintaining routine service and payment controls.

Assume local measurement confirms the net 20 hours released each week. The Operating owner assigns that capacity to aged cases and agrees a daily confirmation slot with the purchasing manager, who reallocates existing staff time. Assume no additional overtime or displaced purchasing commitments. An aged case requires 30 minutes of total staff effort, including confirmation, correction and acceptance follow-up. Four weeks’ released capacity therefore supports 20 × 4 ÷ 0.5 = 160 aged cases. That is calculated capacity; it establishes no accepted outcome.

Acceptance requires a documented, correct resolution, confirmation by the receiving team, and no reopening during ten business days of follow-up. The routine effort estimates also include case-level follow-up; the four weekly monitoring hours remain separate. The Operating owner must verify that these demands fit within funded capacity.

At week 6, suppose the register confirms 160 accepted overdue resolutions, with routine service protected and additional spending within the $2,400 limit. The owner’s minimum of 120 has been exceeded. That is the reason to fund this service improvement. There is no demonstrated cash saving or avoided hiring: better service can represent realized value without cash savings. Whether it remains worth funding depends on the next review and the alternatives.

Six-week case reconciliation — hypothetical

Pending cases earn no benefit credit. An equivalent comparison queue, with comparable demand and case mix, is assumed to maintain routine service but make no progress on its aged backlog.

Recurring cost assumption	Six-week cost
Existing operation: $6,000 weekly, including unchanged employment costs, allocated purchasing effort and existing support	$36,000
Additional technology and usage: $300 weekly	$1,800
Additional support and recovery: $100 weekly	$600
Total: $36,000 + $1,800 + $600	$38,400

These assumptions include all six weeks of review, correction, monitoring, knowledge maintenance and follow-up, including work on cases still pending. Remaining follow-up enters the next period’s operating cost; outcomes count only when accepted. Existing payroll already covers the reassigned hours; do not add them again as a new expense. One-time build costs remain outside this table.

Cost per accepted outcome = $38,400 ÷ (2,400 routine + 160 aged) = $15.00. This is a period measure across two case types, not evidence that routine and aged cases cost the same. Keep their volumes, effort, waiting time and quality visible separately.

The financial effect is $1,800 + $600 = $2,400 more recurring expenditure over six weeks. The accepted service improvement is the benefit being purchased.

If the same 20 hours remain unassigned, or purchasing confirmations still block resolution, the calculation produces no additional accepted outcomes. With service otherwise unchanged, the additional expense has bought no realized benefit. Faster preparation alone has not removed the remaining constraint.

Verify the benefit after deployment

An accepted outcome shows what happened. Attribution asks how much of that change resulted from the intervention. HM Treasury’s Magenta Book, updated May 2026, distinguishes tracking benefits from evaluating their causes, unintended effects and additional impact against a counterfactual. That distinction matters in an operating review.

Agree the baseline and comparison before the change. Keep the same scope, acceptance rules and observation window; record arrivals, case complexity, opening backlog and pending acceptance at both ends. Compare similar cohorts or use a staged rollout where feasible. Examine staffing, overtime, seasonality, policy changes and other improvements that could explain the result. Report uncertainty when a credible comparison is unavailable.

In the invoice example, the claim concerns AI assistance combined with reassignment and faster purchasing confirmations. It does not isolate AI’s contribution. Testing the same operating redesign without AI would help establish whether the technology earns its additional cost.

Finance should reconcile the claim to operating records and expenditure, challenge the counterfactual, and distinguish service improvement from cost avoidance and realized cash savings. Cost avoidance needs evidence of spending that would otherwise have occurred; cash savings need an actual expense reduction, net of additional costs. The Government Efficiency Framework provides a useful public-sector reference for evidence, sustainability and avoiding duplicate savings claims. The Outcome owner still decides whether the accepted service result warrants the resources.

Keep one claim linked to unique case records. The hours reassigned to backlog reduction cannot also become salary savings or support another allocation claim. Where several teams contribute to one improvement, reconcile their records before consolidating the benefit.

Check retention after normal operations resume—for example, at weeks 8 and 12. Examine reopenings, new aging, routine service, downstream waiting, receiving-team effort, support costs and control failures. A gain maintained through hidden overtime or transferred delay needs reassessment. Clearing a finite backlog is not a benefit to multiply indefinitely: once it is cleared, decide how the capacity will be used next.

Match the next commitment to the evidence

An early idea does not need a fully proven operating model to earn discovery funding. It needs a worthwhile question, a bounded cost, and a test capable of changing the next decision.

Use the operating case to state exactly what is being approved.

Commitment	Evidence the next review should examine
Fund discovery	Is the problem material, what is the baseline, and which alternatives merit testing?
Fund a bounded trial	Does the proposed workflow work with real users, representative cases, explicit boundaries, and accounted-for operating effort?
Fund expansion	Do measured outcomes, costs, operating ownership, and tested controls support the proposed scope?

For each commitment, specify the spending limit, permitted scope, Outcome owner, review date, and conditions for stopping or changing course. Set success and failure criteria before the results arrive.

Funding and delegated authority should be separate decisions. Expanding an assistant to more users does not automatically justify letting it approve transactions. Equally, an operation can create material value while keeping consequential decisions with people.

After deployment, use the benefit evidence and retention checks at scheduled reviews, and revisit the case when costs, demand, policy, system behavior or scope changes enough to affect its assumptions. Choose Expand · Continue · Narrow · Redesign · Stop, and record the evidence needed for the next decision.

Keep one concise operating-case canvas

Retain a one-page executive view, supported by the evidence needed for the decision. At the top, name the approval sought, funding requested, scope, and next review date.

Canvas element	What the case should make explicit
Strategic outcome	The enterprise priority, current baseline, target, and time horizon.
Operating problem	The constraint being addressed and the cost or consequence of leaving it unresolved.
Work and decision scope	What changes, affected handoffs, and decisions that remain human-owned.
Solution choice	Alternatives considered and why the proposed level of automation or autonomy is justified.
Knowledge and dependencies	Required sources, their owners, readiness gaps, and funded enabling work.
Governance boundaries	Permitted actions, exclusions, escalation, intervention, and evidence of control.
People and ownership	Outcome owner, changed roles, transition effort, support capacity, and running budget.
Value and learning	How benefits will be realized, full costs, uncertainty, evaluation method, and conditions for expanding or stopping.

Mark assumptions as assumptions and measured results as results. Give each material uncertainty an owner and a means of resolving it.

Carry each material benefit into review through the existing case, using a compact record linked to its evidence:

Benefit-review record	What to retain
Outcome and acceptance	Intended result; owner’s acceptance criteria and service safeguards.
Comparison	Baseline, scope, cohort, period, demand, case mix and counterfactual.
Capacity and decision	Net capacity verified; authorized allocation and remaining constraint.
Responsibilities	Outcome owner; Operating owner and receiving-team commitments; Finance validation.
Operating cost	Full recurring cost, allocation basis and matched accepted count; separate build investment.
Benefit and evidence	Service improvement, cost avoidance or cash savings; supporting records.
Claim limits	Attribution uncertainty; unique case and benefit references; duplicate-claim checks.
Retention and decision	Follow-up dates, persistence and displaced-burden checks; next formal review decision.

Within the Transformation Operating Map, the initial operating case remains a Frame the Opportunity discipline. Carrying its commitments into review connects the original investment decision to evidence from the running operation.