Why Does Dr. Jonah Tebaa Say an AI Business Case Needs a Half-Life?
Dr. Jonah Tebaa argues an AI business case needs a half-life because initial peak benefits inevitably decay as processes change and simple tasks are exhausted. In an illustrative composite distributor example, monthly savings dropped from $42,000 to $27,000 by month eleven. A static model obscures when run costs will overtake returns. He recommends quarterly re-measurement, pre-committing to refresh or retire when gross benefit reaches 1.5 times run cost, and decomposing drops into mix, usage, price, and quality.
Dr. Jonah Tebaa opens his argument with a composite scene most finance teams will recognise. A mid-size distributor automates order handling. In the second month, the system saves $42,000 a month, and that figure goes on a slide. By month eleven, measured in exactly the same way, the saving is $27,000. The year-two budget, however, still assumes $42,000.
Dr. Tebaa argues that this is not a story about a failing technology. It is a story about a business case written without a shelf life. His proposal is direct: every AI business case should state an expected half-life and a retire-or-refresh date, and the value should be re-measured every quarter against that curve.
The problem with booking the launch month
In his work with owners and executives, Dr. Tebaa sees the same habit repeatedly. The first clean measurement of benefit becomes the annual figure, and that annual figure is assumed to repeat. He points out that the early months are the most flattering to measure. The simplest work has been automated, attention is high, and the surrounding process has not yet had time to change.
The result is a year-one return that looks stronger than it is, and, more importantly, a case that never reveals when the system will stop covering its own running cost.
The composite example
Dr. Tebaa is careful to label the distributor a composite. It is illustrative and does not describe a single client. The numbers are chosen to show the mechanism.
- Build cost of $120,000 and a run cost of $9,000 a month.
- A static case of $42,000 a month gives about $276,000 net in year one.
- Re-measured benefit of $36,000, $31,000 and $27,000 at months five, eight and eleven, which is about 86% retained each quarter and a half-life near 14 months.
- A decay-adjusted year one of about $180,000, roughly 35% below the static case.
In his reading, the lower year-one figure matters less than what the curve exposes. Gross benefit reaches one and a half times the run cost around month 25 and drops below the run cost around month 33. A flat line shows neither date.
Why the value fades
Dr. Tebaa deliberately steers leadership teams away from blaming the model first. In the composite distributor, the causes he describes are all on the business side. The easiest order types were used up, the incoming mix moved toward complex orders, two teams worked around the tool on rush orders, and a supplier price change reduced the saving per order. A passing change to the model or prompts can contribute, but in his experience it is rarely the main driver.
The practical consequence is that a fall in benefit should be taken apart before anyone reacts. He suggests separating volume mix, usage, price and quality, because each calls for a different response.
Four steps he recommends
- State the half-life and the break-even date in the case itself, even as a rough estimate.
- Re-measure the same metric by the same method every quarter, keeping the launch baseline visible.
- Decompose any drop into mix, usage, price and quality before acting.
- Pre-commit the decision: refresh, reinvest or retire when gross benefit crosses 1.5 times run cost.
The 1.5 times line is chosen to leave room. Replacing or rebuilding a system takes months, and in the composite the threshold is crossed roughly eight months before the benefit falls under the run cost.
Dr. Jonah Tebaa sets out the full worked example, and the questions he puts to a team before approving a year-two budget, in his own words in Your AI Business Case Needs a Half-Life.
