Calculate AI ROI by subtracting an initiative’s total costs from its financial benefits, dividing the result by those costs and multiplying by 100. Costs and benefits must cover the same period, and forecast returns should remain separate from realised results.
AI ROI (%) = [(financial benefits − total initiative costs) ÷ total initiative costs] × 100
The calculation is straightforward. Establishing reliable inputs requires more care. Time saved, for example, represents additional capacity; it does not automatically reduce spending or increase profit.
Consider a department using an AI assistant to draft customer responses. Drafting becomes faster, but employees still check every response. Before extending the investment, the department head needs to establish the net time saving, the full cost and whether the improvement produces measurable financial value.
For business leaders, department heads and transformation managers, a useful AI ROI assessment connects operational evidence to a funding decision. It should explain what has changed, how that change creates value and which assumptions remain uncertain.
Define the investment and measurement period
Start with a specific workflow and business outcome. “Improving productivity with AI” is too broad to support a dependable calculation.
Record:
- The process is being changed.
- The employees and work covered.
- The expected business outcome.
- The alternative against which the initiative will be assessed.
- The evaluation period.
- The person accountable for benefits and the finance reviewer.
The alternative might be the existing process or another proposed solution. Use a consistent comparison rather than switching between whichever baseline produces the most favourable result.
Also distinguish forecast ROI, based on estimated costs and benefits, from realised ROI, based on observed results and a documented valuation method.
For broader decisions about business goals, use-case selection and organisational preparation, refer to AI Adoption Strategy: A Practical Roadmap for Business Leaders.
Establish a baseline before measuring improvement
A baseline records how the process performs before the proposed change. Without it, you may observe better results without knowing how much improvement occurred.
Measure the work that the AI system will actually support. If it handles straightforward enquiries, do not apply the measured improvement to complex complaints it cannot process.
| Baseline measure | What it helps establish |
|---|---|
| Eligible work volume | How much work could benefit from the system |
| Labour time per completed item | Current effort for a comparable task |
| Review and correction time | Effort beyond initial production |
| Error or rejection rate | Whether completed work meets requirements |
| Relevant expenditure | Spending that might change |
| Work complexity | Whether comparisons involve similar tasks |
Use a period that reflects the relevant workload, and record unusual conditions. A quiet week may not represent the demand or task mix the system will encounter after deployment.
APQC’s guidance on measuring AI value and productivity recommends connecting a pre-AI baseline to attributable process changes and verified business outcomes, while considering quality and rework alongside productivity.
If the organisation cannot collect these measures, address that limitation before claiming precise returns. Data availability and measurement capacity are part of the preparation covered by the AI adoption strategy: A practical roadmap for business leaders
Include the full cost of delivering and operating AI
Licence fees are only one part of the investment. Implementation, internal effort, supervision and maintenance can materially affect the result.
Assess each relevant cost category:
| Cost category | Examples to include |
|---|---|
| Initial preparation | Process analysis, data preparation and procurement |
| Implementation | Integration, configuration, testing and deployment |
| People and adoption | Training, internal project effort and workflow changes |
| Technology operation | Licences, usage charges, storage and connected services |
| Oversight | Monitoring, evaluation, human review and maintenance |
| Failures and changes | Correction, incident handling and significant updates |
Distinguish additional cash expenditure from allocated internal resource costs. Existing employees’ project time has an economic cost, even where their salaries do not change. State whether your assessment uses a cash basis or includes the value of internal resources.
Apply that basis consistently. If review time is already included in the measured labour difference between the old and new processes, do not also deduct its value as a separate cost.
For shared infrastructure, document an allocation method that reflects use or another reasonable basis. Avoid assigning the whole platform cost to every initiative.
AWS’s guidance on calculating AI ROI discusses allocating expenditure to particular uses and accounting for supporting services. It also distinguishes cost per outcome from ROI: knowing the cost of an acceptable result does not establish what that result is worth.
Separate financial returns from released capacity
An initiative can improve work without immediately generating cash savings. Keep different types of benefit separate so that the financial calculation remains understandable.
Realised spending reductions
These are evidenced reductions in expenditure, such as lower paid overtime or reduced external processing fees.
Check that spending fell because of the initiative rather than a decline in demand or another change. Record the evidence and the period covered.
Supported cost avoidance
Cost avoidance concerns expenditure that would otherwise have occurred. For example, additional capacity might remove the need for a documented outsourcing arrangement.
The alternative must be credible. A possible future expense with no supporting plan is a weak basis for a realised benefit claim. Identify the evidence behind the expected spending and label uncertain estimates accordingly.
Incremental contribution
Where AI supports additional sales or output, estimate the attributable financial contribution after associated incremental costs.
Gross revenue alone can overstate the benefit. Additional sales may require fulfilment, support or other expenditure. Finance should agree which costs are included and how the resulting contribution is calculated.
Released capacity and non-financial improvements
Time saved can allow employees to complete more work, reduce delays or undertake other responsibilities. Multiplying saved hours by a labour rate estimates the value of that capacity; it does not demonstrate a payroll saving.
Record how the capacity is used. If it produces additional profitable work, value that outcome with supporting evidence. If the result is improved service that cannot yet be valued reliably, report it separately.
Avoid counting both the value of released hours and the resulting business benefit when they describe the same improvement.
Calculate AI ROI using a worked example
The following figures are hypothetical. They illustrate the calculation and are not a benchmark or a GCI customer result.
Assume an organisation evaluates a defined administrative workflow over its first year. Its assessment identifies separate, attributable reductions in paid overtime and external processing expenditure.
| First-year item | Illustrative amount |
|---|---|
| Implementation and preparation | €12,000 |
| Operating costs, including review and maintenance | €18,000 |
| Total initiative costs | €30,000 |
| Verified reduction in paid overtime | €24,000 |
| Supported avoidance of external processing expenditure | €18,000 |
| Total financial benefits | €42,000 |
| Net financial benefit | €12,000 |
| First-year ROI | 40% |
The calculation is:
(€42,000 − €30,000) ÷ €30,000 × 100 = 40%
The result means that, on the stated basis, the initiative produces €12,000 of net financial benefit relative to €30,000 of costs.
This interpretation depends on several conditions: the two benefit categories do not overlap, expenditure changes are attributable to the initiative, all relevant costs are included and the figures cover the same period.
Calculate payback separately.
Payback measures how long it takes for cumulative net cash benefits to recover the upfront outlay.
For a simplified calculation, assume the example produces €3,500 in gross cash benefits each month and incurs €1,500 in monthly operating cash costs from launch. The monthly net cash benefit is €2,000.
€12,000 upfront outlay ÷ €2,000 monthly net cash benefit = six months
That estimate assumes benefits begin immediately and remain constant. Where adoption takes time or benefits fluctuate, use a monthly cash-flow schedule.
Simple ROI also does not account for when cash flows occur. Material multi-year investments may require discounted cash-flow analysis alongside ROI.
Check how much improvement is attributable to AI
A before-and-after comparison can identify a change. By itself, it does not establish that AI caused the whole change.
Demand, staffing, process redesign or other software may affect the same results. Record these factors rather than attributing every improvement to the initiative.
Where practical, compare similar groups or introduce the system in phases. Use consistent measurement definitions and account for differences in task complexity.
For example, comparing AI-assisted routine enquiries with manually handled complex cases would give a misleading impression of productivity. A more useful comparison examines similar work and includes review and correction effort.
When stronger comparisons are unavailable, explain the limitation. A measured improvement with uncertain attribution should not be presented as a precisely established AI benefit.
Pilot findings also need qualification. A small group of experienced users may not represent the workforce, workload or support requirements of a wider rollout.
Measure adoption, quality and costs after deployment
The business case provides an initial expectation. Ongoing measurement shows whether the initiative delivers it.
Use a small set of measures with named owners and accessible evidence:
| Measure | Evidence source | Suggested owner |
|---|---|---|
| Share of eligible work using AI | Workflow records | Process owner |
| Net handling time, including review | Task measurement | Operations |
| Error and rework rate | Quality checks | Quality lead |
| Initiative expenditure | Invoices and internal cost records | Finance |
| Realised financial benefits | Spending and contribution evidence | Business owner and finance |
Usage alone does not demonstrate value. Employees may use a system frequently without reducing effort or improving outcomes. Equally, limited use may explain why forecast benefits have not materialised.
Track quality with productivity. Faster initial work can create more correction, escalation or downstream errors. Review responsibilities and acceptable operating limits belong within the arrangements discussed in AI Governance Framework: What Business Leaders Need to Put in Place.
Agree on review dates and triggers. Significant changes to usage, pricing, the system or its purpose may require an updated calculation.
Use the results to continue, adjust or stop
The assessment should support a decision, rather than end with a percentage.
An initiative may justify continued investment when evidence supports its value, quality and operating requirements. Expansion still needs a cost and benefit assessment appropriate to the larger scope.
Where results fall short, identify the cause. Low adoption, excessive review effort and rising operating costs require different responses.
Test how the calculation changes under plausible conditions:
- Lower eligible work volume or usage.
- Higher correction and review effort.
- Delayed benefits.
- Increased operating charges.
- Lower attributable financial contribution.
These scenarios reveal which assumptions matter most. They also help distinguish an initiative that needs adjustment from one whose proposed value is unsupported.
There is no universal percentage that defines a good AI ROI. Consider organisational investment criteria, alternatives, uncertainty and the evaluation period.
A positive return does not establish lawful or acceptable operation. A negative first-year result may warrant a longer-term assessment, but future benefits still need evidence and clear assumptions.
Prepare leaders to evaluate AI investment decisions
Reliable investment decisions require coordination between business owners, finance, technology teams and those responsible for oversight.
GCI’s Digital Transformation & AI Adoption for Leaders course covers related leadership topics, including business value, AI strategy and roadmapping, implementation practices, governance, workforce preparation and responsible scaling.
Those topics connect to the organisational work behind an AI business case. The published curriculum does not explicitly list a dedicated ROI calculation module, so the course should not be treated as specialist financial modelling training.
Review the course syllabus against your team’s preparation needs in AI strategy, implementation, governance and responsible scaling before deciding whether to enrol.