Start with the cost of a completed job
An automation is financially useful when the value of the work it completes exceeds the cost of building, running and supervising it. Count a completed business task, such as an accurately reconciled report or a qualified call held. Messages generated and steps executed are activity measures.
Before buying, write down the current process, its monthly workload, the people involved and the evidence that it is a bottleneck. If no one can explain what happens after the automated step, a cheaper step may simply move the queue elsewhere.
Measure the current process before estimating savings
Sample normal work and exceptions. Record hands-on minutes, waiting time, corrections and completion quality separately. An email waiting overnight has long elapsed time but may require only two minutes of staff work. Multiplying the overnight delay by an hourly wage would invent savings.
Use paid working cost when valuing time, with a clear currency and period. For a Kenyan team, prepare the budget in KES; for another market, use its actual costs. Convert foreign software invoices consistently and record exchange-rate and tax assumptions rather than presenting a universal price.
Include the costs that disappear from the sales demo
Ask for setup and ongoing costs separately. A low model bill does not tell you what the complete service will cost.
| Cost | What to include | Evidence to ask for |
|---|---|---|
| One-time setup | Discovery, data cleanup, integration, training and acceptance testing | Scope, exclusions and payment milestones |
| Recurring software | Workflow platform, models, messaging, CRM and hosting | Usage units and estimated normal/peak workload |
| Human work | Review, exceptions, support and process ownership | Expected minutes per task and named owner |
| Maintenance | Broken integrations, changing business rules and regression checks | Support coverage and change fees |
| Failure and exit | Corrections, interrupted work, export and handover | Recovery procedure and ownership of records |
Worked example: time recovered is not automatically cash saved
This is an illustrative calculation, not a Rainlight client result or a market price. Suppose a monthly reporting task uses 40 staff hours. After automation it still needs eight review hours and two hours of exception handling. The recovered capacity is 30 hours. At an assumed KES 500 per hour, that capacity has a planning value of KES 15,000.
Suppose software and support add KES 6,000 per month and setup costs KES 45,000. Net monthly capacity value is KES 9,000, so a simple capacity-value payback is five months. Over 12 months, the illustrative net value is KES 63,000: KES 180,000 capacity value minus KES 72,000 recurring cost minus KES 45,000 setup.
Using total investment as the denominator, the illustrative 12-month ROI is KES 63,000 / KES 117,000, approximately 54%. This excludes financing, taxes and the timing of benefits. If the staff payroll stays unchanged and the recovered time is not put to useful work, that is not KES 63,000 deposited into the business. Keep a cash-savings scenario separate.
Treat extra sales as a separate hypothesis
If faster follow-up creates more attended calls, count the additional profitable customers after the actual sales cycle. Use contribution after delivery costs, not the entire invoice. Subtract added acquisition and operating costs, and avoid counting the same staff savings twice.
A before-and-after increase can also reflect seasonality, a different offer, more ad spend or a stronger salesperson. Keep those changes visible. Where practical, compare similar eligible work during the same period; do not turn a small uncontrolled pilot into a precise causal claim.
Test the assumptions that could reverse the decision
In the example, if review and exceptions consume 30 hours instead of ten, only ten hours are recovered. Their KES 5,000 capacity value is less than KES 6,000 recurring cost, before setup. The same proposed system now has a negative monthly capacity balance.
Ask what happens at lower workload, higher error rates and peak usage. A fixed monthly fee can look attractive at high volume and poor at low volume. A per-task fee can behave the opposite way. Compare both at your real workload instead of comparing subscription prices alone.
Approve a bounded pilot with a business owner
Agree the task, expected output, comparison period, spending limit and stop conditions before launch. The pilot owner should be able to inspect completed work and corrections, not just an attractive dashboard. Use the worksheet below to make assumptions visible.
Current discussions about AI automation are useful for finding candidate workflows. They do not establish your savings. Rainlight can review the relevant sales and operations handoffs with you; bring a sample process, workload and current costs so the conversation starts with your business rather than a tool list.