Most automation proposals talk about hours saved. Very few show the arithmetic behind that claim, and even fewer subtract the cost of maintaining bots, paying for AI usage and handling the items the automation cannot finish. When the numbers are vague, you either approve a project that never pays back or reject one that would have.
This guide gives you a worksheet method you can complete with your own team in about a week, before you speak to any vendor. You will measure the process as it runs today, estimate the value of automating it, count the full cost, and adjust for exceptions so the result holds up in a finance review.
Quick answer: To estimate business process automation ROI, multiply monthly volume by the minutes each item takes and the loaded cost of that time, add the cost of errors and delays, and apply a realistic straight-through rate. Then subtract the one-time build cost and yearly running costs such as licences, AI usage, maintenance and exception handling.
Step 1: Measure the process as it runs today
Every ROI figure is only as good as its baseline. Measure the current process before anyone redesigns it, because a project without a baseline can never prove what it saved.
| What to measure | Where to find it | Tip |
|---|---|---|
| Monthly volume | Counts from Tally, ERP, CRM or the shared inbox | Use 6 to 12 months to capture seasonal peaks such as year-end |
| Handling time per item | A timed sample of real items | Include checking, waiting and correcting, not just data entry |
| Loaded cost per hour | Salary plus PF, ESI, office and system costs | Ask finance for a standard figure so every project uses the same one |
| Error and rework rate | Credit notes, corrections, audit queries, complaints | Note what each type of error costs, not just how often it occurs |
| Turnaround time | Timestamps in email or the workflow tool | Measure waiting time separately from working time |
Write the numbers down per process, not per department. "Accounts payable" is too broad; "entering supplier invoices into Tally and matching them to purchase orders" is measurable.
Step 2: Work out the value side
Value from automation comes from four places. Treat each one separately so that a reviewer can challenge the assumptions one at a time.
- Time value: volume x minutes saved per item x cost per minute. Be clear whether this is a cash saving or released capacity, because finance will ask.
- Error value: the cost of errors avoided, such as wrong input tax credit claims, duplicate payments, late-payment interest or customer refunds.
- Speed value: what faster turnaround is worth, for example quotes sent the same hour, an earlier month-end close or supplier early-payment discounts you currently miss.
- Control value: audit trails, consistent approvals and fewer people handling sensitive data. These are hard to price, so list them in words rather than inventing a rupee figure.
Only the first three belong in the formula. The fourth strengthens the case but should never be the reason the numbers add up.
Step 3: Count the full cost, not just the build
A common mistake is comparing yearly savings with a one-time build quote. Automation has running costs that continue for as long as the process runs.
- Discovery and redesign: mapping the process and removing steps that should not exist at all.
- Build and testing: API integrations, RPA bots for screens with no API, AI steps for emails and documents, and a parallel run alongside the manual process.
- Tool licences and hosting: platforms such as n8n, Power Automate or UiPath are priced differently, some per user, some per bot, some per run.
- AI usage: language and document models are usually charged per request or per page, so this cost rises with volume.
- Maintenance: RPA bots break when a portal or desktop screen changes, and APIs get new versions. Budget for this every year.
- Exception handling: someone still has to clear the items the automation flags.
Step 4: Adjust for exceptions and adoption
The straight-through rate is the share of items the automation completes with no human touch. It is the single most important assumption in the worksheet, and the one most often left out.
- Apply time savings only to the automated share of the volume.
- Assume exceptions take as long as today, or longer, because they are the difficult cases.
- Allow a ramp-up period in the first months while rules are tuned and staff learn the new queue.
- Replace your assumption with the real rate measured during the proof of concept or parallel run, then recalculate before approving the production build.
The resulting worksheet is simple: yearly value on the automated share, minus yearly running costs, gives the net yearly benefit. Divide the one-time cost by that figure to get a payback period you can defend.
Step 5: Compare candidates and pick the first one
Run the worksheet for three to five processes, then score them on factors the ROI figure does not show. A process with good ROI but poor inputs is a risky first project.
| Factor | Easier to automate | Harder to automate |
|---|---|---|
| Volume | Hundreds of similar items a month | A few items with high variety |
| Rules | Written down and followed consistently | Depend on one person's judgement |
| Inputs | Structured data or standard documents | Free-text emails, poor scans, phone calls |
| System access | Systems with APIs or exports | Legacy desktop software or portals with captchas |
| Cost of a wrong result | Easy to catch and reverse | Statutory filings or payments to the wrong party |
If you would like an outside view on this ranking, our business process automation team runs exactly this kind of discovery before proposing any build.
Common mistakes in automation business cases
- Automating a broken process: remove unnecessary approvals and duplicate entry first, then automate what remains.
- Using ideal handling times: estimates from managers are usually lower than what a timed sample shows.
- Ignoring legal sign-off: GST returns and other government filings still need a responsible person to review and submit, so keep that step in the design and the cost.
- Counting every freed hour as cash: it weakens the case the moment finance questions it.
- No owner after go-live: someone must watch run logs, clear exceptions and request changes, or savings quietly fade.
How EthicsComputer can help
Our Business Process Automation service starts with process mapping and a fixed-price proof of concept on your own data, so the straight-through rate in your worksheet is measured, not assumed. A proof of concept typically takes 3 to 4 weeks and a production rollout typically 6 to 12 weeks. Where invoices or forms are the input, we combine it with intelligent document processing, and where the process lives inside your ERP we work alongside ERP software development.
Which process costs your team the most hours each month?
Share the process, rough volumes and the systems involved, and we will tell you whether automation is likely to pay back and how we would test it.
About the author
Vinay Grover founded EthicsComputer in Rohtak, Haryana in 2018 and leads its cybersecurity, software and AI work for clients across India and abroad. EthicsComputer is ISO 9001, ISO/IEC 27001 and ISO 20000-1 certified.
