The question nobody in our industry wants to answer
Every AI automation agency will tell you what to automate. Very few will tell you what to leave alone.
We've walked away from engagements because the math didn't work. We've recommended hiring over automation. We've told clients to keep doing things manually because the ROI didn't clear our threshold.
This is the framework we use to make those decisions. It's the same one we apply to every diagnostic.
The 6-month payback rule
Every automation we build must pay for itself within six months. Not in theoretical value — in recovered hours multiplied by real blended hourly cost.
If the payback takes longer than six months, we don't build it. Here's why:
- Systems drift. After six months, the business processes you automated may have changed
- Tool changes. Vendors update APIs, pricing, and features faster than most automation can adapt
- Attention decay. Nobody will champion a project that hasn't proven itself in half a year
The math is simple:
Payback months = Build cost ÷ (Hours saved per week × Blended hourly cost × 4.33)
If the result is over 6, we have a conversation about whether to proceed. Usually, we recommend against it.
When manual is cheaper: the three tests
Test 1: Volume threshold
If the task happens fewer than 5 times per week, automation often costs more than just doing it. The build cost, testing, and ongoing monitoring eat the savings.
Example: A consulting firm wanted to automate their monthly board report assembly. It took one person 4 hours once a month. The automation would have cost $8,000 to build and $300/month to monitor. At 4 hours × $75/hour = $300/month of recovered value, the payback was 26 months. We said no.
Test 2: Variability threshold
If the process changes more than once per quarter, the automation will break before it pays for itself. Automation is good at doing the same thing repeatedly. It's bad at adapting to shifting requirements.
Example: An agency's client onboarding process changed every time they won a new type of client. The process was different for e-commerce vs. SaaS vs. professional services. We recommended systemizing the process with templates first, then revisiting automation in 6 months.
Test 3: Judgment threshold
If the task requires nuanced human judgment more than 30% of the time, the automation will generate more exceptions than it resolves.
Example: A wealth advisory firm wanted to automate portfolio rebalancing recommendations. But each client's risk tolerance, life stage, and tax situation required genuine advisor judgment. We recommended leaving it manual and automating the surrounding admin instead — the data prep, the compliance pre-check, the meeting scheduling. The judgment stays human; the busywork doesn't.
The framework: Automate, Systemize, or Leave It
For every process we evaluate, we assign one of three recommendations:
| Recommendation | When to use it |
|---|---|
| Automate | High volume, low variability, clear ROI within 6 months |
| Systemize | Medium volume, some variability, would benefit from templates/SOPs before automation |
| Leave it | Low volume, high variability, or requires judgment that doesn't scale |
Most processes fall into "Systemize" — they need documentation and templates before they need technology. We tell clients this even though it doesn't generate revenue for us.
What this looks like in practice
In a recent diagnostic for a professional services firm, we evaluated 8 processes:
- 3 were clear automation candidates (high volume, low variability, strong ROI)
- 2 were systemize-first candidates (needed SOPs before technology)
- 2 we recommended leaving manual (low volume, high judgment)
- 1 we recommended hiring for (40+ hours/week of work that needed a dedicated person)
The client expected us to automate everything. We built 3 systems and helped them hire for the fourth. They saved more money than if we'd automated all 8 — because the other 5 would have broken within 90 days.
The honest answer
If an automation agency can't tell you when NOT to automate, they're selling tools, not solutions. The discipline of knowing when to walk away is what separates systems that compound from automations that break.
Want to know which processes in your business clear the threshold? Book a diagnostic. If the math doesn't work, we'll tell you — and you keep the process map either way.