Is this job worth automating with AI? How to score it in five minutes

Why do most AI projects fail before they start?

Most AI projects don’t deliver. RAND’s analysis found more than 80% fail to deliver the value promised, about twice the failure rate of comparable IT projects. Gartner expects 60% of AI projects running on data that isn’t AI-ready to be abandoned through 2026.

The technology is rarely the cause. In RAND’s practitioner interviews, 84% cited leadership-related causes rather than model performance. In small businesses it looks like this: nobody owns the job, the process is a mess, the data is scattered, and nobody measured the before.

Scoring the work first catches all four.

What are the six questions?

What does the score mean?

• FIX (under 45): sort the process before you build. Write the steps down, pull the data into one place, name an owner.
• TEST (45 to 69): run a 4-week supervised pilot. AI drafts, a person approves, you measure.
• SCALE (70 and over): build it properly, with monitoring and an owner.

Two hard rules override the score. If nobody owns the job, or the inputs live in people’s heads, it’s FIX. If the work is mostly judgement, or a mistake causes legal or money damage, it can’t go higher than TEST.

How do you work out what a build is worth?

Multiply the hours a year by the loaded hourly cost of the person doing it. Then estimate the share AI can take on: about 80% for jobs with the same steps every time, 60% for mostly rules, 35% for mixed work and 10% for judgement calls. Half of that figure is a sensible build budget, because it pays back in six months.

Worked example: quote follow-ups in a trades business. 20 quotes a week, 10 minutes each to chase, admin time at $50 an hour. That’s 160 hours and $8,000 a year. The steps are mostly rules, the quotes live in one job system, mistakes are easy to fix, the owner is keen but the team is unsure, and the systems connect through Zapier. Score: 68, TEST. AI can take on about $4,800 a year of that work, so a sensible build costs about $2,500. Anyone quoting $15,000 for it is pricing their time. Your return doesn’t support it.

What does it look like inside a real business?

At a sports apparel brand selling across Australia and New Zealand (name withheld), two builds scored very differently:

• Weekly ads report: SCALE. Same steps every week, data in Meta and Shopify, low cost if wrong, an owner who wanted it gone. It now writes and emails itself every Monday. The full write-up is the next case file.
• Building Meta ad tests: TEST. Creative choices need judgement and mistakes cost money. So the AI builds every ad set paused, and nothing reaches the ad account until a person types YES.

Same business, same tools, opposite verdicts. That’s the point of scoring each job on its own.