Automation isn't always better — but often it's obvious.
The right question isn't 'automate everything?'. It's: which processes cost the most time to run manually and are stable enough to hand to a system?
What each one actually is
A process where people move data between systems, run checks, and trigger next steps.
Manual processes rely on people doing the work — updating spreadsheets, moving records between systems, chasing approvals. They're flexible, adapt to edge cases, and cost time linearly.
A process where systems handle routine steps automatically, with humans looping in for exceptions.
Automation hands routine steps to systems. Cost is upfront; savings compound. Humans stay in the loop for judgement calls and edge cases.
From decision to five years in
- Stage 1 · Cost to startManual ProcessZeroAutomationUpfront investment
- Stage 2 · Cost to runManual ProcessLinear with volumeAutomationNear-zero at scale
- Stage 3 · Handles edge casesManual ProcessYes (humans adapt)AutomationYes if designed for
- Stage 4 · AuditabilityManual ProcessDepends on disciplineAutomationNative
- Stage 5 · SpeedManual ProcessHuman speedAutomationSub-second
- Stage 6 · Best forManual ProcessLow-frequency + high-judgementAutomationHigh-frequency + rule-based
A closer look at the features
How they compare on the things that matter
An honest look at both sides
- +Zero upfront investment
- +Fully adaptable
- +Great for edge cases
- +Right thing to do for rare or high-judgement work
- −Doesn't scale without headcount
- −Consistency depends on the person
- −Cost grows with volume
- +Compounds over time
- +Frees team for judgement work
- +Auditable + consistent
- +Runs 24/7
- −Upfront investment
- −Requires ongoing maintenance
- −Rigid without careful design
Nuanced recommendations, not blanket ones
Empathy, discretion, and creativity beat rules.
Automation ROI compounds fast.
Reclaims capacity without hiring.
You don't automate what you haven't stabilised yet.
Automation payback formula
The math is not "time saved × rate". This is the version we actually use to prioritise automations.
| Input | How to size it | Example |
|---|---|---|
| Frequency | Runs per month | 200 runs |
| Manual minutes / run | Timed, not estimated | 12 min |
| Hourly loaded cost | Salary × 1.4 / 160 | ₹700/hr |
| Error rate × cost per error | Historical data | 3% × ₹2,000 = ₹60/run |
| Build cost | Discovery + build + test | ₹1.5L |
| Payback months | Build / (monthly savings) | ≈ 3 months at 200 runs/mo |
Which processes should never be automated (yet)
- Runs fewer than 20 times a month
Rarely worth the build. Better to write a checklist and revisit at higher volume.
- Requires judgement more than steps
Automation encodes rules; judgement encodes context. Automate the rule-bound sub-steps only.
- Process itself is broken
Automating a bad process just makes it fail faster. Fix the process, then automate.
- Regulatory + reputational risk on failure
Financial reconciliations, legal filings — keep humans in the critical path with automated assist.
What we hear most often — and what's actually true
It replaces the boring parts of jobs; people move to judgement work.
Only high-frequency, stable processes usually earn back the investment.
Small teams often see faster ROI — they have less capacity to spare.
Frequently asked questions
Frequency × pain × reversibility. High-frequency, high-pain, easily reversible steps first.
n8n, Zapier, custom code — we pick per problem, not per preference.
Every workflow has retries, observability, and a human-in-the-loop for exceptions.
First automation ships in 2 weeks. Compounding savings become obvious around week 6.
Not sure which to pick? Let's decide together.
A 30-minute call is enough for us to give you a recommendation — no pitch.