Not every AI automation is worth building.
Some are clever but pointless. Some save three minutes once. Some create more review work than they remove. Some automate a process that should have been deleted.
That's why every automation needs a test.
At The SAGE Stack, the practical question is simple:
Can this recover meaningful time every week?
Call it the 10-hour test.
What the 10-hour test means
The target belongs to the overall build, so a single small automation can still earn its place.
The test is whether the work is tied to a real recoverable-hours number. If it doesn't give time back, reduce dropped balls, improve turnaround, or make the business easier to run, why build it?
AI has to change the work, or it's decoration.
Question 1: how often does this happen?
Frequency matters.
A task that takes 30 minutes once a year isn't a priority.
A task that takes 15 minutes and happens 40 times a month is a different story.
Ask:
- How often does this workflow happen?
- How long does it take each time?
- Who does it?
- What else could they be doing?
- Does it create delays for clients or the team?
Repeated work is where automation earns its keep.
Question 2: is the process clear?
Automation needs a path.
If no one can explain what should happen, the automation won't fix the confusion. It will just move the confusion faster.
Ask:
- What starts the workflow?
- What information is needed?
- What decision happens next?
- Who approves it?
- Where does the output go?
- What counts as done?
If those answers are unclear, map the process before building.
Question 3: what needs human judgment?
This is where a lot of AI projects go wrong.
They try to automate the judgment instead of the handling around the judgment.
Human judgment should stay in places like:
- final client communication
- sensitive decisions
- strategy
- pricing
- exceptions
- legal or compliance interpretation
- anything where being wrong would damage trust
AI can still help prepare the work. It can summarize, draft, classify, or flag. But the human owns the decision.
Question 4: what is the measurable win?
"It feels easier" is nice, but it isn't enough.
Define the win before building:
- reduce intake setup from 45 minutes to 10
- cut weekly report drafting by 2 hours
- reduce missed follow-ups
- shorten document collection time
- eliminate duplicate data entry
- reduce owner interruptions
The clearer the win, the easier it is to judge whether the automation worked.
Question 5: what could go wrong?
Every automation needs a failure plan.
Ask:
- What happens if the AI is wrong?
- What happens if the data is incomplete?
- What happens if the client replies unexpectedly?
- What needs review before sending?
- Where is the undo path?
- Who gets notified if something fails?
Good systems fail visibly and recover cleanly.
A simple scoring method
Score each opportunity from 1 to 5:
- time saved
- frequency
- pain level
- ease of implementation
- risk level
- client experience impact
Then start with the high-payoff, low-risk workflow. Start where time comes back, even if the impressive-sounding build has to wait.
Example: client document follow-up
Pain:
Clients forget documents. Team members chase manually. Projects stall.
Automation:
When required files are missing after three days, draft a friendly reminder, list the missing items, create a review task, and update the client checklist.
Human judgment:
Review before sending if the client is high-touch or the message needs nuance.
Measurable win:
Fewer stalled projects, less manual chasing, faster onboarding.
This is a good candidate because the work is frequent, annoying, rules-based, and easy to review.
Example: pricing strategy
Pain:
Pricing takes thought and context.
Automation:
AI can summarize scope, compare similar past projects, and list risk factors.
Human judgment:
Final pricing stays human.
Measurable win:
Better decision support, not automated pricing.
This is a support workflow, not a full automation.
The point
The best AI automation makes next week easier.
If it gives time back, reduces repeated handling, improves follow-through, and keeps humans in the right decisions, it's worth considering.
If it only adds complexity, leave it alone.