Where AI Actually Fits in Your Business: The 3-Bucket Test

Most AI projects fail before anyone writes a line of code. Not because the technology is weak, and not because the team is junior. They fail because nobody asked the only question that matters first: does AI actually belong here?

After 35 years building and shipping technology, I've watched this pattern repeat in every wave of hype—cloud, mobile, big data, and now AI. The tools change. The mistake doesn't. People start with "let's use AI" instead of "what are we actually trying to fix." So before you spend a dollar, run every task in your business through a simple test.

The 3-bucket test

Take any task your business does repeatedly and drop it into one of three buckets.

1. SHOULD automate — boring and repetitive

These are the high-volume, rules-light, low-judgment tasks: sorting inbound leads, transcribing calls, drafting first-pass replies, tagging records, reconciling data between systems. The work is tedious, the cost of a small error is low, and a human reviewing the output is faster than a human doing it from scratch. This is where AI pays for itself quickly and quietly. Start here.

2. COULD automate — promising, but the ROI isn't ready

These tasks are technically possible but not yet worth it. Maybe the data is messy, the volume is too low to justify the build, or the model still needs a human in the loop on every output. "Could" tasks are worth a small experiment, not a budget. Run a cheap pilot, measure honestly, and revisit in a quarter. Most failed AI initiatives are "could" tasks that got funded like "should" tasks.

3. SHOULDN'T automate — judgment and relationships

Anything that hinges on trust, nuance, or accountability stays human. Closing a key client. Handling a sensitive complaint. Making the final call on a hire. You can bolt AI onto these, but you'll erode the exact thing that makes them valuable. Knowing what to leave alone is a competitive advantage, not a limitation.

Why experience is the lens

The buckets look obvious on paper. The hard part is putting each task in the right one, and that takes pattern recognition you only get from having shipped real systems. I once had a face-recognition system confidently tag a live footballer as having been dead for 20 years. The fix wasn't more AI—it was constraining the problem to the 22 players actually on the pitch. Accuracy went through the roof. That's the job: not believing in AI, but building around what it can and can't do.

The same instinct turns a "shouldn't" into a "should" when you find the right constraint—and, just as often, stops you from pouring money into a "could" that was never going to clear the bar.

Run the test this week

You don't need a strategy offsite. List the ten tasks that eat the most time in your business. Put each one in a bucket. Then pick a single "should," automate it, and measure the time you get back. That one honest loop will teach you more about where AI fits than any vendor demo.

AI is a swappable tool, not a religion. Build your business around the system, not the model—and you'll still be standing when the model you picked is outdated in a year.

Where do yours fall?

If you want a second pair of eyes on which tasks belong in which bucket, that's exactly what I do on a free 15-minute AI Fit Call. We'll find the one place AI actually fits in your business—and the places it doesn't. Book your AI Fit Call →