90% of US working hours are spent on tasks not covered well by AI benchmarks

What does it mean that 90% of US working hours fall outside AI benchmarks?

Mercor's "Benchmark Map" estimates that about 90% of American working hours go on tasks with little or no coverage from existing AI benchmarks. In plain terms: the tests used to claim AI is "ready for work" mostly do not measure the work most people actually do.

Some caveats matter. This is a company analysis, not a peer-reviewed study, and Mercor calls the result provisional. It mapped 29 benchmarks onto the US Department of Labor's O*NET job taxonomy. Only 924 of 18,797 tasks were covered, and only 208 of 923 occupations had at least one well-covered task.

Why are benchmarks so poor at measuring real work?

Benchmarks favour work that is easy to write down, score and automate. Mercor found that tasks rated as AI-exposed are about 6x more likely to appear on benchmarks than unexposed ones, and a job's most important tasks are 6 to 10x more likely to be covered than its least important.

Mercor also notes that about 63% of US jobs involve physical-world components, which benchmarks handle badly. Its sample is incomplete too, and it likely understates coverage in heavily benchmarked fields such as software engineering. So the gap is real, but the exact 90% figure should be held loosely.

Does a high benchmark score mean AI will save my business time?

No. A benchmark score tells you a model can do a narrow task in test conditions. It does not tell you what happens once a person has to check the output and fix mistakes. That checking is where most of the promised saving disappears.

OpenAI's GDPval tested AI on 1,320 real professional tasks across 44 occupations. The best model, Claude Opus 4.1, was graded equal to or better than human experts 47.6% of the time on the gold subset. Naive use looked about 90x faster and 474x cheaper. After expert review and redoing failed work, savings fell to about 1.12x in time and 1.18x in cost.

A 90x saving on paper became roughly a 12% saving in practice. Plan for the second number.

Where should a small business actually use AI first?

Start where the work is repetitive, the output is easy to check and a mistake is cheap. Drafting replies, summarising enquiries, sorting inbound messages and answering routine questions fit that description. Anything involving money, legal commitments or a customer relationship needs a person in the loop.

A simple test before you automate anything:

1. Can you check the result in under a minute? If not, review will eat the saving.
2. What does a wrong answer cost? A clumsy email is fine. A wrong quote is not.
3. Does it happen weekly or more? Rare tasks rarely repay the set-up time.
4. Time it for two weeks, including review and rework, then decide.

What should founders automate with plain software instead of AI?

Bookings, reminders, follow-ups and missed-call handling. These are rule-based jobs, so ordinary software does them reliably and cheaply, with no review step. Reaching for AI here adds error risk to problems that are already solved.

This is where we see the clearest money. A nail salon moved off Fresha, which was costing about 1,800 a month in commission, to its own booking system at 35 a month, saving roughly 21,000 a year. No model was involved. The gain came from owning the system instead of renting it.

What is Braynex Services' view on the AI hype gap?

At Braynex Services, we think the gap between benchmark headlines and daily work is a reason to be boring first. Fix the leaks you can already measure before chasing the tools that dominate the news.

Most small businesses lose money in places they cannot see: missed calls, platform commissions and old customer data that is never followed up. A plumber we worked with was paying 280 a month to Yell for 3 leads. A Google Business Profile plus a 120 a month ad campaign now brings 18 to 22 enquiries a month. That is a measurable result, not a provisional one.

There is one AI-related move worth making now: being recommended by assistants such as ChatGPT, Gemini and Google AI Overviews is becoming the new local SEO. That depends on owning a real website and an accurate Google presence, not on the latest model.

What should I do this month?

• Pick one repetitive task and time it honestly, including checking and fixing.
• Move one rented dependency (bookings, link page or directory listing) towards something you own.
• Check your Google Business Profile is verified, accurate and answering the questions customers ask.
• Treat any AI saving claim as unproven until it survives your own two-week test.

Want a second pair of eyes on where your business is leaking time and money? Book a free audit at braynexservices.com and we will show you what to fix first, with or without AI.