Technical note 02
Costless Positions in Procurement: How Refusal Moves Through a Peer Network
JDM Inc
Proposition 3 holds that refusal spreads at a lower reinforcement threshold than adoption. In regulated industries, where peer networks are dense and professional reputation is the currency, that asymmetry shapes the buying process more than the product does.
Ask why a compliance team has not adopted a given tool and you will usually get a reason about the tool. Ask how they arrived at that reason and the history often runs through a conference panel, a peer call, an industry listserv, or a remark from a counterpart at a similar institution.
This note is about why the second history is load-bearing, and why it produces a systematic bias that has nothing to do with the merits.
Two different kinds of spread
Information spreads by simple contagion: one exposure can be enough. Behaviors that carry cost or risk are widely held to spread differently. Centola’s experimental test of this compared clustered networks providing redundant social reinforcement against random networks offering greater reach, and found a health behavior spread farther and faster through the clustered structure, with individual adoption much more likely when a participant received reinforcement from multiple neighbors (Centola, 2010).
The behavior in that experiment was itself low-cost, so applying the result to a behavior carrying genuine reputational risk is an inference rather than a demonstration. We flag that rather than lean on it. But it is the direction the mechanism predicts, and the motives are independently documented: people conform not only to be accurate but to affiliate and to maintain a coherent self-concept (Cialdini & Goldstein, 2004).
The asymmetry
Set the two behaviors side by side.
Adopting an AI system inside a regulated function is a complex contagion in the full sense. It requires repeated reinforcement from multiple trusted sources. It requires someone to run a real trial on real matters. It requires tolerating a period of visible early incompetence with a new tool. And it carries the specific reputational exposure of being the person who brought in the machine — an exposure that is asymmetric in its own right, because a bad outcome will be attributed to the decision to adopt in a way that a bad outcome from the manual process will not.
Declining the same system is a simple contagion. It requires one sentence. It costs no pilot budget, no hours, no exposure of anyone’s learning curve. And because effort carries moral weight — Celniker and colleagues found across eight studies that effortful individuals are judged more moral and rewarded accordingly, even when the effort produced no additional output (Celniker et al., 2023) — the sentence confers immediate standing on the person saying it. They are someone with standards.
So the skeptical position propagates under conditions where the adopting position cannot: through weak ties, through a single overheard remark at a conference, through professional cultures that reward visible discernment. The person repeating it collects the reputational benefit of rigor without paying the cost of investigation. Per transmission, skepticism has a lower threshold and a higher payoff than curiosity.
The sequencing error
The asymmetry produces a specific and consequential ordering problem.
Because the low-cost position arrives first and requires no trial, most buyers form a position before any encounter with the system. They then encounter it, if at all, as someone with a position to defend. This inverts the order the instrumental adoption model assumes, in which experience precedes evaluation (Davis, 1989).
The consequence for a pilot is direct: first use is rarely a test. It is an audit, conducted by a party with an interest in the outcome. That interest predicts the pattern every vendor in this space will recognize — a single early failure treated as confirmation, a string of successes treated as flukes. The laboratory version of that asymmetric updating is what Dietvorst and colleagues documented: people abandon algorithms after seeing them err, even when the algorithms outperform the human alternative (Dietvorst et al., 2015).
Conventional sales practice reads this as an objection-handling problem and responds with more evidence. If the framework is right, more evidence is the wrong instrument, because the position was not formed from evidence and is not being maintained by it.
Two testable predictions
The source paper’s Proposition 3 yields predictions that a vendor with enough pipeline could actually check, and we state them as predictions rather than observations because we have not run them.
The first: adoption should cluster by institution and peer network more sharply than task characteristics predict. If two institutions face substantially identical regulatory workloads and one has adopted while the other has not, the framework says look at who each of them talks to before looking at the workload.
The second: the most vocal refusal should concentrate in the roles whose status is most closely tied to cognitive effort. Within a single compliance function, that is a within-organization prediction, not just a between-organization one.
There is a third implication the paper raises under stratification, and it is the one we find least comfortable. If refusal travels through affinity networks, then who ends up with capability tracks network membership rather than need. The prediction is that the institutions with the least cover — the small shop, the solo practitioner, the community bank with three people in the function — may be talked out of tools they would benefit from most, by peers at larger institutions with more to lose from disruption. The paper calls this a hypothesis rather than a finding and flags it as among the more urgent items for empirical test. We agree on both counts.
What follows for how we sell
If the mechanism is reinforcement thresholds rather than argument quality, then the thing that moves a regulated buyer is not a better deck. It is a peer at a comparable institution who has run the thing on real matters and will say so — more than one, if Centola’s threshold result carries.
That is slower than the alternative and it is not a growth hack. It is also the only move the model actually endorses. We would rather say that plainly than pretend the asymmetry is something a demo can overcome.
References
Celniker, J. B., Gregory, A., Koo, H. J., Piff, P. K., Ditto, P. H., & Shariff, A. F. (2023). The moralization of effort. Journal of Experimental Psychology: General, 152(1), 60-79.
Centola, D. (2010). The spread of behavior in an online social network experiment. Science, 329(5996), 1194-1197.
Cialdini, R. B., & Goldstein, N. J. (2004). Social influence: Compliance and conformity. Annual Review of Psychology, 55, 591-621.
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319-340.
Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114-126.
Rogers, E. M. (2003). Diffusion of innovations (5th ed.). Free Press.