What does The $740,000 Feature: Dr. Jonah Tebaa on AI Roadmap Timing mean in practice?
Dr. Jonah Tebaa argues that AI roadmap timing fails when companies waste scarce senior engineering attention building arriving generic capabilities instead of truly proprietary ones. To correct this, he introduces a four-question framework and a board instrument requiring a column classifying every initiative as BUILD, ADOPT-WHEN-AVAILABLE, or WATCH with a named owner and review date. This mechanism ensures organizations only fund proprietary builds while purchasing earliness exclusively against specific, dated competitive deadlines.
In month eleven of a sixteen-month build, the vendor whose claims platform the company already licensed shipped the exact thing the company was building. Not a rival product — an included feature, inside software already on the invoice. The in-house engine went live on schedule anyway, and it worked. By then the same function cost about $34,000 a year to anyone holding that licence, and a competitor had switched it on in six weeks.
Dr. Jonah Tebaa uses this composite to make an argument that is not really about that build. A regional insurance group of roughly 1,900 staff. A three-year AI roadmap carrying eleven initiatives. Item three — an in-house document-extraction engine for claims intake — at $740,000, sixteen months, and a team of six drawn from the strongest engineers in the business.
Nothing in that sequence was mismanaged, and Tebaa is careful on the point. Every approval was defensible on the day it was cast: the need was genuine, the number sat inside tolerance, and on the date of the vote nothing on the market performed the function. The project was delivered. The decision was still wrong, and the wrongness lived where the pack had no column.
It lived on item seven — a pricing model built on twelve years of the group's own claims history, the single initiative no vendor could ever package and sell to anybody else. It never started. The six people capable of building it were busy constructing something the market delivered on its own.
Two species of line, printed as one
Tebaa's first claim is structural. An AI roadmap, he argues, always contains two categorically different kinds of item, and the standard board pack renders them identically — same row, same cost, same timeline, same style of ROI figure beside each.
The first kind is an arriving capability. It is generic across industries, requires no proprietary input, and is already shipping from two or more major platforms whose published roadmaps state where it heads next. Document extraction, translation, meeting summarisation, first-line triage. They travel toward every licence holder whether or not a single line of budget is committed.
The second kind is a proprietary capability, existing only because of something the organisation alone holds — its historical data, its regulated process, its licence, its customer relationships. Item seven was of this kind. Twelve years of one market's claims outcomes cannot be built once and resold to eleven competitors.
Because both appear in the same table with the same arithmetic beside them, the difference is invisible at the vote. The pack says whether each line is worth doing, not whether it is theirs to do.
Earliness is a separate purchase
Funding an arriving capability early does not buy the capability, which was coming regardless. It buys earliness — a real product, carrying a real price, that boards almost never price.
Earliness is occasionally worth a great deal, but only against a specific dated competitive consequence somebody can name aloud: a tender closing in March, a regulator's deadline, a contract that renews once in five years. Where no such date exists, Tebaa is blunt. Earliness is not a benefit, and funding it as though it were turns a purchase into a project and staffs it like one.
Four questions, asked line by line
Tebaa puts these not to the roadmap as a whole, but to each initiative, one at a time.
- If nothing is funded here, does this capability arrive inside software already licensed within twenty-four months? Tebaa treats this as evidence, not opinion. Admissible: a published vendor roadmap entry, two or more major platforms already shipping it, and no proprietary input required.
- Does the value come from having it before competitors, or from having it at all? The first answer requires a named date. The second usually means the item can wait without loss.
- What in this build belongs to the organisation in a way a vendor could not sell to a competitor next quarter? The asset must be named — the data, the process, the licence, the relationship. A sponsor unable to name it is describing a purchase.
- If the organisation waits and is wrong, how long is the catch-up? Six weeks behind a rival on a vendor feature is a cheap error worth risking often. Two years behind on a data-and-model programme started from nothing is not recoverable inside a strategy cycle; build that one first.
Notice what these questions ration. Not capital: the insurance group could have found a second $740,000. No budget could produce another six engineers of that calibre. Tebaa's third claim is that the binding constraint is senior delivery attention, not money. Capital spent badly is embarrassing and recoverable. A year of the only team able to build the proprietary item is gone at any price.
What a decision to wait has to carry
The obvious objection is that waiting is how organisations fall behind. Tebaa accepts it, then raises the bar for what counts as a wait decision. Three things must be attached: a named owner, a review booked in a specific quarter, and an obligation to stay adoptable — data kept structured and exportable, portability and exit terms secured at the next routine renewal, not during an adoption negotiation where the leverage is gone. Absent those three, he calls it drift wearing the word patience.
His instrument is modest enough to survive a real board: one added column carrying one of three words beside every initiative — BUILD, ADOPT-WHEN-AVAILABLE, or WATCH — each with an owner and a review date. Then minute the timing decision, not only the funding decision. Boards record what was approved and for how much; very few record why now rather than in eighteen months, so nobody can be held to that reasoning later.
Item seven, in the composite, remains unbuilt. That is the sentence Tebaa wants directors to sit with. The measure of a roadmap is not how much of it moves, but whether the part only that company could have built is the part that got built.
