Users value the system they have above its measured worth. That claim is used to explain resistance to replacement, and it is doing more work than the evidence supports.
The evidence is contested
Laboratory markets reported less trading than predicted and higher selling than buying valuations for consumption goods. Later work reported no willingness-to-accept versus willingness-to-pay gap under modified elicitation procedures.
Loss aversion has the same shape of dispute. The original framing describes losses and gains relative to a reference point with losses weighted more heavily. A 2024 meta-analysis reports an aggregate estimate above the neutral value. A reanalysis reports weaker loss aversion for designs using symmetric gains and losses with unordered presentation.
None of this is settled, and none of it was measured on people using workplace software. Two propositions stay separate: that a valuation gap exists, and that it explains a particular objection to a particular replacement.
The practical consequence is that the behavioural explanation is the expensive one to act on and the cheap one to assume. Invoking it converts a solvable workflow problem into a communications exercise.
Check the workflow first
The diagnostic order is functional before psychological.
Record the objection in the user’s own terms and identify the task or consequence at issue, separating a stated preference from an observed inability to complete work. Compare old and proposed workflows on the same representative tasks — including the exceptions, the saved work, the personal arrangements, and the people downstream who receive the output.
Then list the rival explanations and the observation that would distinguish each. Restoring a missing correction path tests a functional explanation. It does not isolate an ownership effect, and it need not.
Dispatchers objecting to a new workboard look like resistance until a realistic task shows that handover notes disappear between shifts. The response is to restore or redesign the handover and observe again. Continued objection after that leaves questions open; it does not prove an ownership effect by elimination.
And the asymmetry does not license a conversion rate. A scheduling replacement that improves lookup while making recurring appointments harder to amend has two outcomes to record separately and one amendment path to fix. It does not have an exchange rate under which two new features compensate for one lost one.
Parity-plus is the acceptance shape
Parity-plus agrees the outcomes that must be preserved for the chosen scope, and the additional benefit that justifies the increment. It is not feature-count equality, and it is not a rule that every incumbent behaviour survives.
The sequence matters. Observe the incumbent on representative work before setting pass criteria, and record broken or unwanted behaviour separately from necessary outcomes. Then specify the preservation criteria and the intended improvement together, with agreed handling for exceptions, deliberate retirements, and the groups affected by each.
Run comparable tasks and relevant adverse cases, examine outcomes by affected group, and record the exclusions alongside the successes. Review operational support and recovery, including changed data. Then proceed, revise, defer, or stop — keeping unmet conditions visible after the decision rather than closing them with it.
A permit-renewal slice that makes status checking simpler must still support correcting a submitted address and the downstream inspection handoff. Where ordinary renewal succeeds and the correction route fails, that condition stays open. Removing an unused legacy screen is evaluated by its consequences rather than counted as lost parity.
Service guidance supports this shape directly: solve a whole problem for users, while explicitly permitting incremental improvement. Complete for the work it claims, not complete relative to the old system.
Adoption is not agreement
Where the old route is removed, usage measures compliance.
An assistant used by nearly everyone because the previous form was withdrawn, whose users copy each answer into a second system and correct the missing fields, has complete adoption and an unmeasured burden. The review records completion quality, the copying, the correction work, the affected groups, and the available alternatives before deciding whether to redesign or expand.
Consultation has a similar gap between form and substance. Survey evidence associates worker consultation with more positive reported experiences of AI, and says explicitly that causality cannot be established — from 2022 surveys in finance and manufacturing across seven countries, with workers recruited through online panels. Historical association, not an effectiveness estimate.
Two further distinctions keep the evidence honest. Commitment to a specific change is different from shared belief in the collective capability to implement it, and a team can have one without the other. And consultation records should show influence and unresolved concerns rather than invitations sent.
The rule
What stays fixed is that a captive population converts every gap into someone’s unpaid workaround, and the workaround is observable before release. What changes is whether the objection is functional or attitudinal, and that is answered by testing the workflow rather than by naming a bias.
Not to be confused with
Irrationality. A user objecting to a replacement that removed a step they need is correct. Labelling that as resistance ends the investigation at the point where it would have been useful.
A settled coefficient. No behavioural multiplier applies to feature counts, satisfaction scores, or projected benefits. Where an average improvement conceals a failing workflow, the average is the misleading number.