From First Online Order to Lasting Market Access

For the Georgian firms that responded to a conditional buyer, the important question was not only whether they could get online once. It was whether training made the demand opportunity easier to use, and whether an initial online transaction survived after the special order disappeared. This chapter follows that lifecycle: first the test of complementarity, then the movement from short-run activation to 2025 catch-up and possible backsliding. A separate Liberian experiment sharpens the comparison by showing that targeted market-access information can matter in a different setting, without making the two studies interchangeable.

Training did not clearly unlock the demand shock

The overlap between the two Georgian experiments made it possible to examine complementarity. Of the firms considered for the demand intervention, 205 had also been part of the earlier training experiment. This created a comparison among firms that received training, firms that faced the later demand offer, firms that experienced both, and firms that experienced neither. The structure resembles a two-by-two design, but it must be read with its timing intact: training and demand were delivered at different moments, and firms had to remain eligible for the later experiment.

The complementarity question is straightforward. If training lowered the technical cost of getting online, it might make a firm more responsive when a buyer later appeared. A firm that had already learned how to create a profile, understand customer information, or use a digital platform might be able to act on the order more easily. In that case, the combined effect would be larger than the effects of training and demand support considered separately. Training would not merely add its own small effect; it would unlock part of the demand shock that was otherwise inaccessible.

The opposite prediction is also coherent. If the main remaining obstacle was organizational inertia, training might not change the response to a deadline. A firm could understand the process and still fail to reserve time, coordinate fulfillment, or believe that the opportunity was worth disrupting routine work. The two interventions would then address different terms in the decision, with little interaction. The demand offer could trigger firms that were ready to act, while training alone remained insufficient and did not materially expand the set of firms able to respond.

The short-run interaction estimate was 0.018, with a p-value of 0.849. The training effect was small, the demand effect was strong, and the combined exposure did not show a detectable additional response attributable to their interaction. In practical terms, the evidence does not support a large claim that the training program made firms substantially more responsive to the conditional order. It is consistent with the idea that a technical lesson did not unlock the demand opportunity for the average firm in the overlap sample.

That is not the same as a precise zero. The overlap sample was limited, and breaking the firms into several experimental groups increases uncertainty. The paper notes that the interaction standard errors were large enough that economically meaningful complementarities could not be ruled out. A non-significant interaction therefore weakens the case for a large complementarity under this design; it does not establish that training and demand support are always independent, or that a more targeted form of assistance could never help a buyer’s offer succeed.

The distinction between an additive effect and a complementary effect is useful beyond this study.