AI EBITDA Is Not Free Money
Productivity gains become profit only when a company can prevent them from being competed away. Most cannot, and the value case rarely says so.
The arithmetic in the value-creation plan usually looks like this. Engineering costs X. AI makes engineering forty percent more productive. Therefore either X falls by forty percent, or output rises by forty percent at the same cost. Either way, EBITDA improves.
The arithmetic is correct. The economics are not, because it treats a cost reduction as though it were proprietary.
Four exits before the margin
A productivity gain has to survive four things to become profit. Each one takes a share, and in competitive markets they frequently take all of it.
Competitors receive the same gain
The tools are commercially available and roughly as effective in every organisation with competent engineers. A cost advantage that arrives simultaneously across an industry is not an advantage. It is a new baseline.
Prices decline
Where buyers have alternatives and any sense of what production costs, lower costs eventually reach the price. This does not require a price war. It requires one competitor to decide that market share is worth more than margin.
Expectations expand
The same budget buys more feature, faster support, better integration. The saving is spent inside the product to hold position, and it never appears as a line anyone can point to.
Barriers to entry fall
The clearest case, and the least discussed. If building a competing product cost fifteen million and now costs four, the number of credible competitors rises. That is a structural change in the market that shows up as pricing pressure years later, and it is caused by the same productivity gain that appeared in the value case as an upside.
A cost reduction becomes a margin only if something prevents it from being competed away. Naming that something is the whole exercise.
When it does become profit
The gain is retained where the company has a position that lets it hold price while its costs fall. That is the same short list as everywhere else in this argument, which is not a coincidence.
- Pricing set by value delivered rather than by cost of production — outcome-based, regulated, or embedded in a process where switching is expensive
- Buyers who cannot easily compare, because the purchase is bundled, infrequent or specified rather than shopped
- A market small or specialised enough that new entrants do not arrive despite lower build costs
- Cost structures where engineering was never the dominant line, so competitors gain less from the same tools
- A genuine head start in applying the gain, converted into something durable before it is matched
The last one is real and it is temporary. It is worth having, but it should be modelled as a two-year advantage rather than a permanent margin uplift, and the plan should say what the two years are used to buy.
Consequences for a value-creation plan
None of this argues against pursuing the productivity gain. Not pursuing it is worse: a company that declines a cost reduction available to all of its competitors is simply choosing to be expensive.
It argues against booking the gain as upside. The defensible treatment is to model it as a defensive requirement — the cost of staying at parity — and to model any margin expansion separately, with an explicit reason why this company can keep what others cannot.
The workforce version of the same error
The identical mistake appears in headcount planning. If AI makes a function more productive, the saving is real for the company that reduces headcount first and structural for the industry once everyone does. The difference between an early mover and a late one is usually one or two years of margin, not a permanent position.
What survives is the part of the workforce whose output was never mainly volume: the people who decide what should be produced, who carry accountability for it, and who hold the customer relationships. Which returns the argument to the thing that does not compress — judgment.