First Response Time: The KPI That Rewards the Wrong Behavior
A customer support organization weworked with had made First Response Time their flagship metric. They’d drivenmedian FRT down impressively over two quarters — from hours to minutes acrosstheir digital channels. Leadership celebrated. Then CSAT started declining inthe same period, which made no sense to anyone, because faster response issupposed to make customers happier.
We looked at what the fast firstresponses actually contained. The pattern explained the paradox. Agents,measured on FRT, had learned to send a fast first response that satisfied themetric without addressing the issue — “Thanks for reaching out, I’m lookinginto this for you” sent within 90 seconds, followed by the actual substantiveresponse hours later. The metric measured the acknowledgment. The customerexperienced the wait for the real answer, which hadn’t improved at all. FRT hadgone down. Time-to-actual-help had not. The metric was being satisfied withoutthe outcome it was supposed to represent.
This is the classic failure mode of FirstResponse Time, and it’s nearly universal. FRT measures speed of first contact,customers care about speed of resolution, and the gap between the two is whereagents learn to optimize the metric at the expense of the goal.
First Response Time is unusually easy to game because the behaviorthat satisfies it is trivially separable from the behavior that helps thecustomer.
A first response can be anything. An automated acknowledgment, aholding message, a “looking into it” reply — all of these satisfy FRT withoutdoing anything for the customer. The moment the metric exists, the easiest wayto hit it is to decouple the first response from any actual work, and rationalagents under measurement pressure do exactly that.
The metric rewards the wrong half of the interaction. Customersdon’t contact support to receive a fast acknowledgment. They contact support toget a problem solved. FRT measures the acknowledgment and stays silent on thesolution, which means it can improve dramatically while the thing customersactually care about doesn’t move at all.
The gaming is invisible in the metric itself. FRT looking greattells you nothing about whether resolution improved. You only see the problemwhen you look at a different metric — CSAT, resolution time, repeat contacts —and notice it diverging from the FRT trend.
FRT isn’t useless. A genuinely fast, substantive first response doescorrelate with satisfaction. The problem is FRT in isolation. It needscompanions that close the gaming gap.
Time to resolution. The metric customers actually care about. If FRTimproves while time-to-resolution doesn’t, you’re gaming, not improving.
First contact resolution. Whether the first response actuallyresolved the issue rather than just acknowledging it. A fast first responsethat resolves beats a fast first response that defers, and only FCR capturesthe difference.
Response substance. Whether the first response contained actual helpor just an acknowledgment. Conversation analytics can distinguish asubstantive first response from a holding message, which makes the gamingvisible in a way the raw FRT number never will.
Customer effort. How much work the customer had to do across thefull interaction. A fast first response followed by three rounds ofback-and-forth is worse than a slightly slower response that resolved in one.
FRT is a specific instance of a general problem: any KPI thatmeasures a proxy for the goal rather than the goal itself will eventually beoptimized in ways that satisfy the proxy and abandon the goal.
This is why single-metric management fails in contact centers. AHToptimized alone produces rushed calls and repeat contacts. FRT optimized aloneproduces fast acknowledgments and slow resolutions. Containment optimized aloneproduces deflection and frustration. Every proxy metric, pursued in isolation,drifts away from the outcome it was supposed to represent.
The defense is measuring the actual outcome alongside the proxy, sothe gaming becomes visible. When FRT and time-to-resolution are trackedtogether, the “fast acknowledgment, slow help” pattern can’t hide. The proxystays useful only as long as the outcome is watched beside it.
1. Compare your FRT trend to yourresolution-time trend. If FRT improved andresolution time didn’t, you’ve found the gaming. The gap is the problem.
2. Audit what your first responsesactually contain. Pull 20 fast first responses. Howmany resolved anything versus just acknowledged? The ratio tells you whetherFRT means what you think.
3. Cross-reference FRT against CSAT. If they’re diverging, the metric has decoupled from the experienceit’s supposed to represent.
4. Measure customer effort across thefull interaction. A fast first response meanslittle if it’s followed by a long, effortful resolution path. The wholeinteraction is what the customer experiences.
5. Pair every proxy metric with itsoutcome. FRT with resolution time. AHT with repeatcontacts. Containment with downstream contacts. The pairing is what keeps theproxy honest.
First Response Time was supposed to makecustomers feel attended to quickly. Pursued alone, it teaches agents to send afast nothing and call it a response. The metric improves, the celebrationhappens, and the customers wait exactly as long as they always did for the onlything they actually wanted — an answer. A KPI that can be satisfied withoutachieving its purpose will be, every time, unless you measure the purposebeside it.
The AI terminology chaos is real. Your "divide and conquer" framework is the clarity we needed.
Finally, a clear way to cut through the AI hype. It's not about the name, but the problem it solves.
