What is the 30% rule in AI?
There is no standard, official "30% rule" in AI. The phrase is used loosely online. The rule of thumb we use: start by automating the roughly 30% of a workflow that is repetitive and low-judgment, and leave the 70% that needs context, approval, or nuance to humans.
Treat it as a deliberately conservative starting point, not a research finding. It prevents the "automate everything" mistake that wrecks workflows, frustrates staff, and usually ends with a rollback.
In practice, this means looking at any process and splitting it into parts. The 30% that is templatized, repeatable, and low-risk is a good target for automation. The 70% that involves judgment, client-specific context, or accountability stays with people. Over time, as trust and data build up, that balance shifts.
The risk of ignoring this rule: teams automate tasks they did not fully understand, the AI makes confident errors on edge cases, and trust in the system collapses. Starting at 30% and growing slowly is almost always faster than starting at 80% and cleaning up.
Want to go deeper?
Signal & Form helps Canadian businesses move from curiosity to working AI systems.
Related questions
What are the three golden rules of AI?
Start from a business problem, not a tool. Respect your data (quality, security, consent). Keep a human in the loop for anything that matters.
Why do 85% of AI projects fail?
Not because the models do not work. Because objectives are fuzzy, data foundations are weak, nobody owns the project after the pilot, and change management gets skipped.
What are the 5 things AI cannot do?
AI cannot feel empathy, take responsibility, reason well in novel situations with no data, set strategy or values, or be held legally accountable for outcomes.
Have a specific question about your business?
Book a free 30-minute discovery call. We will answer your specific question and tell you whether we are the right fit, or not.