Wealth Experiment

The $90 AI service that earned $21 an hour

Count the invisible work before deciding whether to offer a second batch.

By Ashley Kays Rabbitt · November 18, 2026

Imagine selling a small business a $90 update to its customer FAQ. The tool produces a decent first draft in twelve minutes. That number makes the job sound spectacular. The client buys a verified, approved update, and the clock runs during the parts around the draft too.

Keep a time log for one order. In this fictional example, outreach and scoping take 45 minutes, intake takes 30, drafting and editing take 45, fact-checking takes 60, and client review and delivery take another 60. That is four hours. After $6 in direct costs, the job leaves $84 before overhead, taxes and other expenses, or $21 for each hour worked.

Illustrative economics of one FAQ order
$90collected
$84after direct costs
$21per working hour

Run the second sale on a different assumption

Change to test What would make it worthwhile?
Ask for one approved source document at intake Fewer minutes spent reconciling contradictory answers.
Limit the order to five questions A predictable review cycle and an update the buyer still values.
Sell a monthly refresh to the same buyer Less acquisition work, with enough substantive changes to justify renewal.

Change one of those things for the next comparable order. Include time spent finding buyers who never purchased; allocate that selling time across the orders it helped produce. Record cash collected separately from proposals and invoices. An outcome-based check also applies to the work itself: ask whether the client accepted and used the delivered FAQ. Anthropic • Evaluating agent outcomes

If the second buyer supplies clean source material and earns you a better hourly return, test that intake condition again. If demand exists only when you absorb hours of research, you have discovered something important about the price or the customer. Either result belongs in the experiment log.