Issue 076 - Workplace AI - Break-even time
How much useful time must AI save to pay for itself?
Companies are still trying to decide whether broad AI rollouts are productivity tools, expensive perks, or both. This scenario asks what happens if a company buys $30-per-month AI subscriptions for 10,000 employees.
Computing, media, and attentionAbout 1 minute
Sources checked October 13, 2026. Figures and circumstances may have changed.
The problem
If a company pays for AI subscriptions for 10,000 employees at $30 per person per month, about how many minutes per employee per working day must the tools save in the first year to justify the rollout after subscription cost, training, checking, and imperfect conversion of saved time into productive work?
The guess unit is minutes per employee per working day. The 10,000-person headcount mostly scales the total dollar exposure; the per-employee break-even math is the cleanest way to think about the problem.
Because Fermi problems target an order of magnitude, I normally use no more than two significant digits and write most calculations in scientific notation; the Fermi reference explains both conventions.
Grounding facts
BLS reported that private-industry employer compensation costs averaged $46.89 per hour in June 2026, and full-time private-industry workers averaged about $54 per hour including wages and benefits. So $50 per hour is a reasonable round loaded-cost peg for an average employee hour, with knowledge-worker roles often higher.
The strongest workplace studies do not show one universal AI effect. A large customer-support study found AI assistance increased issues resolved per hour by about 15% on average, with large variation across workers. A BCG/Harvard field experiment found consultants were about 25% faster on tasks inside the AI frontier, but performed worse on a task outside that frontier when they over-trusted AI output.
A 2026 survey study of Korean workers found reported AI time savings, but also emphasized that saved time does not automatically become measured output. That is exactly why this problem separates gross time saved from useful productive time realized.
After checking sources
Checked answer and calculation
The company-wide subscription spend is large:
monthly cost
= 10,000 employees x $30/month
= $300,000/month
annual subscription cost
= $3.6 million/yearBut the per-employee subscription-only break-even is tiny:
subscription cost per employee
= $30/month x 12
= $360/year
hours needed at $50/hour
= $360 / $50/hour
~= 7 hours/year
minutes per working day
~= 7 hours x 60 / 250 days
~= 2 minutes/daySo if you count only the subscription fee, even a few useful minutes per workday can justify the tool. The harder question is the fully loaded first year.
Using Matt's overhead assumptions of 5 hours of training and about 4 hours per week of checking or correcting AI work:
first-year overhead time
~= 5 training hours + 4 hours/week x 50 weeks
~= 205 hours/year
subscription-equivalent time
~= 7 hours/year
total useful time needed
~= 212 hours/year
~= 51 minutes/working dayThat gives the useful-time answer: under these assumptions, the tool needs to create about one hour per employee per day of value. If only half of gross saved time turns into actual extra productive work, the gross time saved needs to roughly double:
gross time saved needed
~= 51 useful minutes/day / 0.5
~= 100 gross minutes/dayThe sensitivity is the point. With only subscription cost, the break-even is 2 to 3 minutes/day. With light rollout friction, maybe 10 to 30 minutes/day is enough. With heavy checking and only partial realization, the required gross saving can rise to an hour or more per day.
That means a company-wide rollout can be financially sensible if the tools reliably touch common workflows. But it is not enough for employees to feel faster. The organization has to turn some of that freed time into better output, shorter cycle times, higher capacity, or avoided hiring.
Before checking sources
Matt's original estimate
This is the unverified estimate Matt wrote before checking sources, not the checked answer.
The 10,000 employees doesn't seem like important information for this calculation.
First thing I estimated was an employees hourly rate. If I started with a $70k salary and divided by 2000 annual work hours, that's a pretax hourly rate of about $35.
That means ignoring all other factors, about 55 or 60 employee minutes need to be saved by AI each month, or about 3 minutes daily, to break even on the subscription cost.
Then when we account for training and time spent checking AI work, I assumed 5 hours initial training and about 4 hours weekly spent reviewing/checking AI products. When you include the monthly hour necessary to justify the subscription cost, that adds to 217 hours annually that AI needs to offset to justify the spend. If there are 250 work days per year, that's about 0.87 hours daily that need to be offset by AI, or about 52 minutes daily.
Then when we account for how that saved time isn't always applied to productive tasks, and if I assume that only happens 50% of the time, then really the AI needs to offset 434 hours per employee annually, or about 104 minutes daily.
Reasoning score
Matt's reasoning score: 80 / 100
Higher is better: earn points for useful facts, a sound reasoning approach, correct math, and a final estimate close to the sourced answer. The owl meter shows percent full of it: 100 minus the reasoning score.
Useful facts: 20/30. The $35/hour salary peg was low versus loaded employer cost, but still within the right ballpark. The review-time assumption is plausible for some workflows but high as a general rollout average.
Reasoning approach: 30/30. Separating subscription-only break-even, training/review overhead, and realized productive value was exactly the right model.
Math: 10/10. The arithmetic followed cleanly from the assumptions.
Final estimate: 20/30. The subscription-only result was very close, and the full first-year answer is within the checked scenario range. The exact threshold depends heavily on review burden and how much saved time becomes output.
Post-check reflection
Matt's reflection
Nothing super off-base here, I thought I was starting from pretty defensible assumptions but the time spent checking AI work could vary considerably, and the multiplier for reclaimed time being applied to productive tasks is also questionable. Either way, I'm probably in the right ball-park, and it takes about an hour saved per day to justify a $30 subscription.
Recommended memory peg
For workplace software, remember: at about $50/hour loaded labor cost, every $100/year of per-employee software cost needs only about 1 minute per working day of useful saved time. Review overhead and realization rate can dominate the subscription fee.
Reader results
Bars show how submitted estimates sort into the answer choices from the gut-check prompt.