Why generic feedback doesn't change behavior, and what does
Everyone agrees coaching matters. The research is clear: agents and sales reps who receive regular, specific coaching improve faster, stay longer, and produce better customer outcomes. What most organizations actually deliver is something else entirely. A supervisor pulls up a quality score, plays a 30-second call excerpt, and offers advice that could apply to anyone: "try to be more empathetic," "remember to set expectations," "take more ownership." The agent nods and nothing changes. The problem isn't the agent or sales rep. It's the coaching model. It's too vague to act on, too slow to feel relevant, and too disconnected from what actually happened in the conversation to earn trust. The gap between the coaching most organizations intend to deliver and what they actually deliver comes down to one thing: evidence, not opinions. Not a number on a scorecard. Evidence that is specific, traceable, and grounded in real moments from real conversations.
Coaching fails when the person leaves the session knowing their score but not what to do differently. This happens more often than most organizations admit, and the reasons are structural, not personal.
Vague feedback is unactionable. "Be more empathetic" is a personality aspiration, not a behavior an agent can practice on their next call. "Show more urgency" doesn't tell what urgency looks like at the 2:30 mark of a billing dispute. These aren't coaching instructions. They're wishes. Agents can't operationalize them because nobody has shown them what the behavior looks like in practice, in their own conversations, with their own customers.
Delayed feedback disconnects from the moment. Most agents and sales reps receive coaching on calls from one to three weeks ago. By then, the call is a distant memory. The person can't recall what they were thinking, what the customer's tone was, or why they made the choices they made. Coaching on a two-week-old call is like reviewing game film from last season. Technically useful. Practically disconnected from anything the person can change right now.
Score-centered sessions replace understanding with judgment. When a coaching session consists of reviewing a scorecard line by line, the dynamic is evaluative, not developmental. The agent is defending a number rather than exploring a skill. The scorecard might identify that empathy was marked low, but it doesn't explain what happened in the conversation that produced that score, what the agent could have done differently, or what a stronger approach would have sounded like. The score erases the story, and the story is where the coaching value lives.
Inconsistent coaching across supervisors. One supervisor coaches with detailed examples and collaborative problem-solving. Another emails the scorecard and schedules a five-minute check-in. A third skips coaching entirely when the queue is busy. Without a consistent evidence base, coaching quality depends on which supervisor the agent reports to, not on a systematic development process.
Evidence-based coaching starts from a different premise: every coaching recommendation should be grounded in something specific that happened in a specific conversation, traceable to a specific moment, and verifiable by both the coach and the agent.
The difference plays out in every part of the coaching workflow:
Identifying what to coach on. In traditional models, coaching priorities come from quality scores or supervisor instinct. Evidence-based coaching identifies priorities from behavioral patterns across every conversation an agent handles, not a sample of three to five. If an agent consistently moves to troubleshooting before confirming the customer's issue, that pattern shows up across dozens of interactions. The coaching target is specific, data-supported, and defensible, not based on whichever call the supervisor happened to pull.
Grounding the conversation in real moments. Instead of telling an agent their empathy score was low, a coach can point to a specific moment: "At 3:42 in this call, the customer described their frustration with the billing charge. Your next statement moved directly to the resolution without acknowledging what they said. Here's what happened next." The agent can listen to that moment. They can hear their own words. The coaching conversation starts from shared evidence rather than one person's recollection.
Adjusting for what the agent was actually facing. Not every conversation carries the same difficulty. A routine address change is different from a billing dispute with a customer who has called three times already. An agent who navigates a high-difficulty interaction to a reasonable outcome has done stronger work than one who smoothly handles a simple inquiry. Evidence-based coaching accounts for this. The difficulty of the conversation is part of the evaluation, which means agents are coached fairly rather than penalized for drawing harder calls.
Tracking whether coaching actually worked. The most important question in any coaching program is whether agents change their behavior after receiving feedback. Evidence-based systems can track this directly: did the agent apply the coached behavior in subsequent conversations? Did the pattern shift? If the agent was coached on acknowledging customer concerns before troubleshooting, the system can monitor whether that specific behavior improved in the following weeks. This closes the loop between coaching investment and behavior change in a way that scorecard reviews never could.
Evidence-based coaching is only as good as the evidence behind it. If the evaluation methodology is subjective, inconsistent, or disconnected from what actually happened in the conversation, the coaching built on it inherits those same weaknesses.
Full coverage, not sampling. Coaching from a sample of three to five calls per agent per month is coaching from incomplete information. An agent's strongest moment and their biggest development opportunity may both live in the 97% of conversations nobody reviewed. Full-coverage analysis ensures coaching priorities reflect the agent's actual performance, not a random draw.
Observable behaviors, not interpretive judgments. If the evaluation criteria include items like "showed empathy" or "used appropriate tone," two reviewers will score them differently. The coaching that follows is grounded in an opinion, not a fact. Criteria based on observable actions — did the agent acknowledge the concern before moving to resolution, did they set expectations for next steps — produce consistent findings that agents can trust.
Traceable findings, not aggregate scores. A quality score of 78 tells the agent nothing about what to change. A finding that links to a specific moment in a specific call gives them something concrete to work with. Every coaching recommendation should trace back to the evidence that generated it.
Difficulty adjustment. Agents who handle hard calls well should not receive the same coaching as those who handle easy calls adequately. Without adjusting for conversation difficulty, the evaluation systematically undervalues the best agents and overvalues those who happen to get simpler interactions. Fair coaching requires fair measurement.
Consistency across the team. When every agent is evaluated against the same criteria on every conversation, the coaching process is perceived as fair. Agents are more likely to trust and act on feedback when they know the same standard applies to everyone, not just those whose calls happened to get pulled for review.
When coaching resources are limited, and they always are, they tend to flow to the lowest performers. It is also the lowest-return investment a coaching program can make.
The agents performing in the middle of the distribution, adequate but not exceptional, typically have the highest development potential. They have the foundational skills. They know the product, the systems, and the process. What they lack are the specific behavioral adjustments that separate good from great: the pause before troubleshooting, the expectation-setting during a complex resolution, the proactive next step that prevents a repeat contact.
These are exactly the kinds of behaviors that evidence-based coaching identifies and develops. And because mid-level agents handle the largest share of conversations, small improvements in their performance compound across the operation faster than dramatic turnarounds at the bottom.
High performers also benefit from coaching, though the focus shifts. For top agents, evidence-based coaching reveals the specific behaviors that make them effective, which can be documented, shared, and used to develop others. It also identifies the edge cases where even strong agents have room to grow, keeping development continuous rather than stalling once an agent reaches "good enough."
Organizations deploying AI agents alongside human reps face a new coaching question: how do you evaluate and improve both with a single framework?
AI agents don't need the same kind of coaching humans do. They don't forget steps or have bad days. But they do need evaluation against the same criteria, resolution quality, compliance adherence, accuracy of information, and handling of edge cases. When something goes wrong with an AI agent, the "coaching" is a configuration change, a routing adjustment, or a model update. But identifying what went wrong, and how often, requires the same evidence-based evaluation applied to every human conversation.
When human and AI conversations are evaluated with a single framework, coaching becomes comparative. Which interactions does the AI handle better? Where does it consistently underperform? Which scenarios should be routed to humans? These questions inform both human coaching priorities and AI agent improvement, and they are only answerable when both sides are measured the same way.
There is a simple test for whether a coaching program is working: can an agent leave every coaching session knowing exactly what to do differently on their next call, and exactly what that looks like in practice?
If the answer is "try harder" or "be more empathetic," the program is failing. If the answer is "at this moment in this call, here's what happened, here's why it mattered, and here's an alternative approach," the program is coaching.
The difference is evidence. Everything else follows from that.
