Allen Vong
6 min
Abstract
I study how a firm uses mediated communication with a worker and its clients to maximize worker performance over time. I find that optimal mediation involves occasional randomizations, secret from clients, between two continuations. In one, the worker cuts corner and then retains his current continuation utility. In the other, the worker exerts effort and then receives the highest continuation equilibrium utility less a minimal penalty for underperformance. These randomizations eventually disappear, replaced by canonical carrot-and-stick incentives. Optimal mediation Pareto-improves upon no mediation for both the worker and the average client if and only if the worker is sufficiently patient.
Sam: In the high-utility phase—when the worker's promised future payoff sits near its peak—the firm sends private messages only to him. It secretly mixes between telling him to shirk with no future change in his payoff, or to exert effort followed by a big reward on good output but a sharp drop on bad. Clients hear nothing private, so they accept based on overall low shirking odds. This keeps the worker's payoff steady or pushes it higher, unlike public plans that force quicker drops.
Alex: Okay, so private there creates that asymmetry... what happens when utility dips a bit?
Sam: Then it shifts to what acts like public communication in an intermediate zone. The firm demands full effort every time, with acceptance guaranteed. Good output bumps utility up slightly; bad drops it by more, nudging back toward high or down to low. This reward-punishment spread enforces effort without rejections yet.
Alex: And the low zone?
Sam: There, the firm mixes publicly between full effort with acceptance or outright rejection paired with shirking. Utility stays trapped low if punishment fits public equilibria without mediation, or hits zero unless a lucky good effort output escapes. Eventually, after finite bad luck, it converges to pure public randomization matching no-mediation benchmarks.
Alex: Huh... so the private start builds a history-dependent path—past secrets lock in different futures—making incentives stronger than static public mixes where players just feel indifferent.
Sam: Precisely. This dynamic correlated structure—where private history shapes continuations—beats unmediated public strategies, as static mixes can't induce strict effort preferences. The paper frames secret randomizations as the firm's "inattention" to some paths, known to the worker but hidden from clients, turning attention into a correlation tool over mere monitoring. Worker always gains more payoff than unmediated top; clients do too if patience exceeds a threshold.
Alex: That welfare split makes sense... mediation delays waste but favors the worker early on. But even with that dynamic path from private to public, does mediation wipe out all the inefficiencies from the worker's temptation to shirk?
Sam: No, it doesn't. The paper shows through Proposition 4 that a fixed gap remains between the firm's best mediated payoff and its ideal first-best outcome, no matter how patient everyone is. Even in full communication equilibria, each period clients accept only if there's some chance the worker exerts effort, which always carries that minimum moral hazard cost—the hit needed to deter shirking.
Alex: Huh... so the short-lived clients prevent a full folk theorem, where payoffs could approach anything feasible over time?
Sam: Exactly. In games with long-lived players on both sides, patience lets them sustain nearly efficient outcomes via future promises. Here, short-lived clients can't carry favors across periods or punish far ahead, so mediation's dynamic correlation helps but can't replicate that long-lived client setup.
Alex: So gains but limits... any real-world angle for firms facing this?
Sam: In gig or service economies with quick client turnover, firms could use AI mediators for those early personalized secret signals—randomizing privately to boost worker effort and cut rejection-driven churn, before shifting to public tracking. It offers a practical tool beyond fixed wages.
Alex: Makes sense... mediation delays waste via private history, improves both sides meaningfully, but that gap reminds us of the core shirking tension.
Sam: Exactly—the paper's contribution is this precise dynamic plan from private to public, strictly expanding feasible payoffs in a setting public strategies can't. That's the key insight from Vong's work.
Alex: Thanks, Sam. That's it for this look at dynamic mediation and moral hazard. Thanks for listening.