Who Adopts—and Why the Software Share Falls

The key question now is selection: why does the same non-rival software investment make sense for some firms but not for others, and why can its share of spending decline after adoption? The Census model answers by combining productivity, scope, and an adoption cost with a changing factor mix. This gives a more useful interpretation than treating adoption as a simple consequence of firm size or treating a falling software share as evidence that software has become less important.

Adoption is worth it only after scope makes it worthwhile

A firm does not adopt custom software merely because the technology exists. It compares two operating plans. Under one plan, it remains a non-adopter and produces with labor and capital. Under the other, it pays the cost of using the software technology and obtains a production process in which the software can be shared across its establishments. Adoption is worthwhile when the profit from the second plan, after the adoption cost, exceeds the profit from the first.

The crucial term is the number of establishments over which the investment can be spread. A low-productivity firm generally chooses a small scope. If it operates only one or a few establishments, the software investment has little room to reduce the average cost of the firm. The technology may improve the operation, but the benefit of reuse is too small to offset the cost of adopting it. A more productive firm can support a larger scope, giving the same non-rival input more places to work. As scope rises, the effective software cost falls, and the adoption plan becomes more attractive.

This creates a threshold. Below it, the firm remains a non-adopter because its expected gain is insufficient. Above it, adoption can raise profit because the firm can use the software across enough establishments. In the calibrated model, this threshold can arise from the increasing return to scope even without a fixed adoption cost. The authors nevertheless include a fixed cost of adoption so that the model matches the observed share of firms with positive custom-software investment.

The fixed cost should not be read too literally. It is a compact way to represent a barrier that is paid at the firm level rather than at every establishment. The data do not say that every firm faces one identical implementation bill, and the model does not require that real adoption costs be identical. The authors note that firm-specific adoption costs could rationalize why some small firms adopt even though the baseline version, with one common fixed cost, produces a clearer productivity threshold.

That qualification changes how the empirical relationship should be interpreted. The Census data show that adoption becomes more common with scope, but they do not imply that scope is the only determinant. Productivity, expected demand, management capacity, access to finance, and the fit between the software and the firm's operations can all affect the payoff. In the model, those differences are summarized mainly through heterogeneous productivity and the resulting scope choices. In reality, they can also appear as differences in adoption costs or in the ability to implement the same technology.

A threshold is not a moral ranking.