The Human Bit method
Count the checking cost
Judge value after reviewing, correcting and governing the output, not from the speed of the first generated draft.
How it works
AI cost includes more than a subscription or API bill. It includes preparing inputs, reviewing output, correcting failures, managing privacy and maintaining the workflow when products change.
Checking should be proportionate to consequence. If review is difficult or expensive, reduce the scope, improve the evidence available to the reviewer or choose a more reliable intervention.
See the difference
A consultant saves forty minutes drafting a workshop summary and spends ten minutes checking it against their notes. The result remains useful after review.
An analyst generates a complex financial workbook quickly, then spends longer tracing invented assumptions and repairing formulas than building the model conventionally.
Watch for this failure
Counting generated volume or time-to-first-draft as productivity while ignoring correction work and the cost of plausible errors.
Questions to ask
- How will a person check the result?
- How much time does reliable checking take?
- What does one important failure cost?
- Would a simpler tool reduce the review burden?