Quick answerShould the agent explain why it's offering the gesture, or just offer it? State the gesture plainly as a goodwill offer tied to the specific issue raised; avoid implying it's compensation for a company failure unless that's actually confirmed, since an overstated apology can create expectations for a bigger remedy later. What if a customer asks for a specific gesture not on the approved menu? Route it to a human. The value of a fixed menu is that every automated gesture has already been pre-approved; a custom, off-menu request is by definition outside that pre-approval and needs a person to say yes.
Quick answer
Yes, and it is one of the lowest-risk autonomous authorities you can grant an AI agent, since a service-recovery gesture, a shipping upgrade, a waived fee, a small account credit toward a future purchase, returns no money and replaces no product; it only changes a cost the company was already going to absorb in goodwill. Set the authority as a fixed list of eligible gestures with a hard per-gesture and per-customer-per-period cap, log every instance the same way you would a discretionary spend, and reserve human review only for a request that falls outside the pre-approved list or exceeds the cap.
Why this sits below a refund and below a warranty replacement in risk
A refund moves real money out of the company. A warranty replacement commits inventory, shipping cost, and a second physical unit. A service-recovery gesture, in contrast, is usually a marginal cost the company is already absorbing somewhere, an upgraded shipping tier that costs a few extra dollars, a waived rush fee, a small credit that only has value if the customer transacts again. That lower financial exposure is exactly why the authority to grant it belongs at a different, lower threshold than the no-human-review refund threshold or the warranty-claim replacement decision, both of which involve returning money or committing physical goods the gesture never touches.
Building the eligible-gesture list
Do not give the agent open-ended discretion to invent a gesture; give it a fixed, pre-approved menu: free shipping upgrade on the current or next order, a waived rush or convenience fee, a small percentage credit toward a future purchase capped in absolute dollar terms, or an extended return window. Each entry on the menu should have its own cap and its own eligibility rule (first-time issue only, not available to an account already at its goodwill limit for the period). A fixed menu is what makes this safe to fully automate; an open-ended "use your judgment" instruction is what turns a low-risk gesture into an unbounded discretionary spend.
Setting the caps and the tracking
Cap both the per-instance value and the cumulative value per customer per rolling period, the same discipline used for a discretionary goodwill-credit budget, and log every gesture granted with the triggering complaint, the gesture chosen, and its dollar-equivalent cost, so finance can roll it up the same way they roll up refund spend. A customer who hits the cap should route to a human, not silently receive a second gesture that pushes total goodwill spend past what the budget was designed to absorb.
When to escalate instead of granting the gesture automatically
Escalate to a human, rather than defaulting to the largest item on the menu, whenever the underlying complaint suggests a pattern (the same customer requesting gestures repeatedly), whenever the gesture would exceed the period cap, or whenever the complaint itself signals something more serious than routine service friction, in which case it may belong on a dedicated VIP or high-value escalation path rather than the standard automated gesture flow.