An AI ROI calculation compares the annual value an AI project creates, from cost reduction, faster revenue or risk avoided, with its full cost of ownership, including build, running costs, monitoring and staff review time. The simplest starting formula is hours saved per task times loaded hourly rate times annual frequency, minus total annual cost.
Key takeaways
- Start with cost reduction: it is the easiest value to measure and defend to a board.
- Revenue acceleration and risk avoidance can be larger but need more careful assumptions.
- Count every cost: build, model usage, hosting, monitoring, human review and training.
- Measure today's baseline before building anything. Without it, ROI is a guess.
- Phase the work so you can stop after a pilot if the numbers do not hold.
Why should CFOs ask what doing nothing costs?
"How much will this AI project cost?" is half the question. The other half is how much the current process costs every month it stays manual: staff time, errors, delays and the key-person risk when one person holds the process in their head.
AI is a P&L lever. If you cannot draw a line from the initiative to lower cost, faster revenue or reduced risk, do not build it. If the numbers do not work on paper, they rarely improve in production. Try our interactive ROI calculator alongside this framework.
What are the three sources of AI value?
1. Hard cost reduction
The safest place to start. It removes the person-as-connection work: staff moving data from one system to another.
Annual value = hours saved per task x loaded hourly rate x tasks per year- Invoice volume (assumption)500 a month
- Current staff time (assumption)83 hours a month
- Loaded hourly rate (assumption)$45
- Current annual costAbout $44,800
- Time remaining for review if 85% is automatedAbout 12 hours a month
- Annual staff time value releasedAbout $38,000
Subtract annual running costs and divide the build cost by the net annual value to get payback. Use your own volumes and rates.
2. Revenue acceleration
Harder to measure but often larger. It removes bottlenecks that slow sales.
Hypothetical example: an engineering consultancy. Tender responses take days because they need input from busy engineers. An AI system drafts much of each response from past bids, capability statements and project histories, and engineers review it. If the team can respond to more tenders at the same win rate, the extra wins are the value. Model it as: additional tenders per year x win rate x average project margin.
3. Risk mitigation
Critical in regulated sectors: financial services (ASIC, APRA), mining (WHS), healthcare (TGA, AHPRA).
Hypothetical example: a mining services contractor with thousands of active contracts wants to know which contain uncapped indemnities. A legal team can spot-check a small sample each year. An AI contract review can screen every contract and flag high-risk clauses for lawyers to review. The value is the probability-weighted cost of the disputes avoided, which your legal team can estimate.
What hidden costs of AI should you show your board?
The build is only part of total cost of ownership.
| Cost category | What it covers |
|---|---|
| Discovery and build | Process mapping, engineering, integration, testing |
| Model usage | Per-use fees for cloud models, or hardware for models on your own servers |
| Hosting and storage | Infrastructure, databases and document indexes |
| Monitoring and maintenance | Accuracy checks, model and system updates, prompt changes |
| Human review | Staff time checking outputs, highest in early months |
| Training and change | Onboarding staff and updating procedures |
- BuildOne-off discovery, build and integration
- RunUsage, hosting and monitoring
- ReviewStaff time on approvals
- Year 1 totalBuild plus first-year running costs
- Payback pointWhen cumulative value passes total cost
What should an AI ROI board paper include?
Use this structure for an investment committee paper. Replace the illustrative figures with your own measured baseline.
Hypothetical project: automated customer onboarding (illustrative)
| Metric | Current state | AI-enabled state | Change |
|---|---|---|---|
| Throughput | 20 applications a day | Measured in pilot | Capacity |
| Cost per application | Staff time per unit | Usage cost plus review time | Unit cost |
| Error rate | Measured data entry errors | Errors after AI plus validation | Accuracy |
| Turnaround | Measured hours | Measured after pilot | Customer experience |
| Staff time | Hours on data entry | Hours reallocated | Higher-value work |
Our AI business case template for board presentations has a full paper structure.
How do you calculate AI payback period?
- Measure the baseline: volume, time per task, error rate and rework cost.
- Estimate the automated share: the portion AI can draft, with a person reviewing.
- Value the released time: hours released x loaded rate.
- Add other value: revenue or risk, with conservative assumptions.
- Total the costs: build plus annual running and review costs.
- Payback: build cost divided by (annual value minus annual running cost).
What does your calculation show?
- Payback within 12 months on cost reduction aloneStrong case: proceed to a pilot
- Payback depends on revenue or risk assumptionsPilot first and measure before scaling
- Low volume, under a day a week of manual workUsually not worth a custom build; use built-in features
- No baseline measuredMeasure for two to four weeks before deciding
Why should AI projects be phased?
Avoid open-ended retainers with vague deliverables. AI work is modular, so structure it to produce evidence quickly:
- Assess one process and measure its baseline.
- Build a pilot that runs alongside your team.
- Measure the result against the baseline.
- Decide to proceed, adjust or stop.
A fixed fee per phase, agreed before work starts, keeps cost predictable and lets you stop after any phase with what you have built.
How Solve8 can help
Our consultants measure your baseline and build the business case as the first phase of our AI strategy service, and our engineers deliver the pilot through process automation, running on your infrastructure in Australia. Each phase has a fixed fee agreed before we start. Talk to us about your first AI business case.
