
Automation is worthwhile when it reduces the cost, elapsed time, error, or risk of a defined process by more than the cost of building and owning it. “A person no longer copies rows” is not a complete outcome: exceptions, supervision, integration downtime, and the new system dependency still exist.
Measure the current process, remove unnecessary steps, and automate a stable rule. Otherwise the business accelerates waste.
| Measure | Current state | Evidence |
|---|---|---|
| Transactions per month | System or observation | |
| Human touch time | Work sample | |
| End-to-end lead time | Event timestamps | |
| Exception rate | Queue and manual review | |
| Error/rework rate | Quality and incidents | |
| Financial exposure | Finance | |
| Peak workload | Campaign/season | |
| Required control | Risk/compliance |
Do not estimate the entire baseline from interviews. Observe a representative sample of real cases.
TCO = discovery + build/configuration + licences + integrations + data + training + oversight + support + incidents + upgrades + exit
Include process-owner time, exception testing, infrastructure, vendor costs, and decommissioning. A free script still needs ownership.
Annual benefit = usable capacity + prevented error loss + faster cash cycle + added throughput − new operating cost
Released minutes become financial value only when cost is removed, a growth constraint is lifted, or people redirect time to valuable work. Do not book every saved minute as cash saving.
| Process | Frequency | Standardisation | Error cost | Exceptions | Candidate |
|---|---|---|---|---|---|
| Copy order status | High | High | Medium | Low | Strong |
| Decide a disputed claim | Low | Low | High | High | Assist, do not auto-decide |
| Match IDs and amounts | High | High | High | Medium | Strong with exception queue |
The best first candidate is often frequent, observable, reversible, and governed by a rule that does not change every week.
Before release, define scope, decision metric, quality and risk guardrails, exception taxonomy, manual fallback, action limit, audit record, stop condition, and observation period.
Compare the tail as well as the average. Automation that improves median time but leaves unresolved exceptions for days can make the system worse.
After release it needs:
ROI = (cumulative benefit − cumulative TCO) / cumulative TCO. State the period, range, and assumptions. Show expected avoided loss separately when risk reduction is material.
A frequent, standardised, measurable process with costly manual error and a clear exception path: reconciliation, state transfer, feed checks, or routing.
When it is rare, unstable, judgment-heavy, or cheaper to execute reliably by a trained person. A checklist and better input form may be sufficient.
Reviewed: 3 September 2026.
Continue with AI-agent guardrails, technical-debt prioritisation, and the operations maturity model.
Pingvera can automate recurring checks and retain outcome evidence; the process owner still governs thresholds, exceptions, and response.
Pingvera watches whether an online business actually works — uptime, checkout, orders, domain, SSL and server — and alerts you in Telegram, email or a webhook before a customer has to tell you.
Start freeRead next: Do You Need Headless Commerce? · AI Agents in Ecommerce: Safe Guardrails.