The Inversion
Why the Western cost-displacement case for AI does not transfer to Indian enterprises.
When labour is not the largest line on the sheet, displacing it is not an argument that survives a capital allocation meeting. What replaces it, and why knowledge retention is the underweighted case.

Radexus. Institutional memory for revenue. radexus.com


The prevailing case for enterprise AI was constructed in markets where commercial and administrative labour is the dominant controllable cost. In that setting, cost displacement is a clean argument: expensive hours become fewer hours, and the arithmetic is straightforward.

Transplant that argument into a mid-market Indian manufacturer and it produces a number that is real, modest and insufficient. Commercial headcount is a smaller share of the cost base. The saving competes for capital against a machine that will demonstrably add capacity, and it loses.


## Three arguments that do transfer

### One. Revenue already yours, leaking

Quotes that went quiet without being marked dead. Installed bases nobody is re-selling to. Dealers who stopped ordering and were never called. Escalation clauses never claimed. This case requires no new customer, no new capacity and no new headcount, which makes it the easiest thing a promoter has ever been asked to approve.

### Two. Working capital

Days sales outstanding, dead stock, retention money held by clients. The cost of capital makes this arithmetic materially more compelling in India than the identical model in a low-rate market, and a promoter feels working capital physically in a way he does not feel a productivity percentage.

### Three. Knowledge retention

The underweighted case, and structurally the most acute. Tenure in Indian commercial roles is shorter, attrition higher, and the institutional memory of how a territory actually buys leaves the building more frequently than in the markets where these systems were designed. A company that loses its best regional manager loses the map, and rebuilds it over three years at full cost.

The promoter has watched this happen twice. He has no mechanism to prevent the third time. That is the sale.


## Implications for how these systems should be built here

  - Design for retention, not displacement. The measure is what the company still knows after a departure, not hours saved.
  - Assume a shorter tenure in every role that feeds the system. If capture depends on a person being diligent for five years, it will fail in eighteen months.
  - Price against value released rather than headcount reduced, because the headcount argument is small and the value argument is large.
  - Expect the systems to be older, more fragmented and less documented than the Western reference architecture assumes. This is a build constraint, not a reason to migrate.


(c) 2026 Radexus, a Global AI Forum company. San Francisco and Chennai.