The short answer
Estimate value per completed task, including human review and failures. Compare that value with implementation, integration, security and ongoing operating costs. Time released is capacity, not automatically cash saved. The best first project is one with a measurable bottleneck, usable data and a clear owner for exceptions.
Measure the existing workflow first
Choose one task, such as preparing an internal support response or classifying an incoming request. Record monthly volume, handling time, error rate and rework. Use observed samples across normal and difficult cases rather than the fastest demonstration. Separate waiting time from hands-on work so you do not count the same benefit twice.
Our planning worksheet records the baseline, proposed workflow, acceptance criteria and person who can redeploy any released capacity. If the organisation cannot use the time saved, the financial case should not treat it as an immediate reduction in payroll.
Include the costs that arrive after the demo
Budget for integration, data preparation, identity and access, testing, training and handover. For ongoing operation, include model and cloud charges, human review, support, monitoring and changes to connected systems. Ask what happens to the quote if a supplier API changes or a new approval step becomes necessary.
Microsoft’s cost-optimisation guidance recommends a maintained cost model aligned with business value. NIST’s Manage guidance calls for weighing AI benefits against negative risks. [1][2] Our commercial application of those ideas is to make security and exception handling explicit cost items before comparing proposals.
Worked example: capacity is not the same as savings
This is a fictional planning example, not a client result or a SundAI price. Assume 1,200 tasks a month, six minutes of manual work per task and two minutes after automation, including review. That releases 4,800 minutes, or 80 hours. At an illustrative loaded labour value of €35 per hour, capacity value is €2,800 per month.
If incremental operating costs are €900 per month and implementation costs €12,000, the simple net capacity-value estimate is €1,900 per month. Dividing €12,000 by €1,900 gives about 6.3 months. This is a capacity-value payback illustration, not a cash-payback forecast. It excludes financing, tax and ramp-up, and assumes every released hour is productively used.
If review takes four minutes instead, only 40 hours are released. Their illustrative value is €1,400; after €900 operating cost, the difference is €500 and the same calculation becomes 24 months. Small changes in real review effort can matter more than a cheap model subscription.
Illustrative sensitivity check
| Assumption | Base case / slower review |
|---|---|
| Monthly task volume | 1,200 / 1,200 |
| Minutes after automation, including review | 2 / 4 |
| Monthly capacity released | 80 hours / 40 hours |
| Capacity value less €900 operating cost | €1,900 / €500 |
| €12,000 divided by monthly net capacity value | About 6.3 months / 24 months; not a cash forecast |
Agree a pilot decision before buying a rollout
Set a minimum usable output rate, maximum review effort and acceptable exception path for your actual workflow. Assign somebody to inspect failures and record costs during the pilot. Avoid expanding permissions just to reach a speed target; that changes the risk as well as the economics.
At the decision meeting, choose to proceed, narrow the scope or stop. A useful pilot can show that ordinary rules-based automation is sufficient, that data needs preparation, or that human handling remains preferable. Capture the reasoning so the next project starts with evidence instead of another optimistic demo.
Bring these inputs to a scoping call
- Task volume and measured handling time, including difficult cases.
- Systems and data the workflow needs to access.
- Actions that need approval and people who handle exceptions.
- Implementation budget and ongoing cost assumptions.
- How released capacity will be used and how success will be measured.
Questions buyers ask
Can you quote an automation project from the task name alone?
A useful quote needs the systems, data, permissions, integrations and acceptance criteria. Two workflows with the same name can require very different implementation and assurance work.
What does the €490 AI Decision Review cover?
One AI decision, a 45-minute discussion and a 3–5-page memo. Scope, timing and applicable VAT are confirmed before acceptance. It is separate from implementation or a technical security assessment.
Unsure whether the project earns its place?
Start with one decision and a defined brief. The €490 AI Decision Review can help organise the trade-offs before a larger commitment; applicable VAT and scope are confirmed first.
Explore the €490 Decision ReviewKeep exploring
Sources and editorial method
SundAI editorial guidance, prepared with AI assistance. Source claims are linked; worksheets and scenarios are our suggested planning tools, not observed client outcomes. This article is not a certification or a legal opinion. Sources checked on 16 October 2026.