A Report Is Not a Response

I have known a secret for years. This month, Medallia told it to the entire market.

The secret is this: Voice of Customer (VoC) insights and reporting was never the job. It was a ruse, a mechanism to get someone to care, and to act. I learned that running VoC reporting cycles for years. We would spend weeks aggregating data, synthesizing themes, producing a defensible report or a clean stack-ranked list. That was never the job. The report was the admission ticket into the room. The job in the room was to get someone important to care, and to act.

Then it got written down on a balance sheet. Thoma Bravo lost effectively all five billion dollars it put into Medallia, the category's establishment leader, as the Financial Times reported. Not because people stopped wanting to hear from customers. Because the value of customer feedback was never in collecting it or writing it up. It was in what you do about it, and Medallia was built to collect and report.

A report is not a response

Almost everyone is about to draw the wrong lesson from this. They will say the platforms died because AI made synthesis cheap. The work those companies sold can now be done for less, the price collapsed, and the category went with it. A tidy story that treats this as a cost problem.

I have sat on both sides of that report, and it was never a cost problem. It was a layer problem. VoC sold you the collecting and the reporting: the survey, the readout, the quarterly deck nobody acted on. But think about the last time a customer took the time to tell you something real. They were not waiting for you to write it down nicely. They were waiting for you to do something. Those are not the same layer of the stack.

A report is not a response.

The people closest to the money are saying it out loud. CMSWire, covering the wipeout, put the mechanism plainly:

AI is lowering the cost of generating customer insights, which puts pressure on platforms that built their value proposition around collecting feedback and generating reports rather than driving measurable business action.

The pressure is not on insight. It is on platforms that focused purely on the report, rather than the response.

Bill Staikos, who has spent his career inside this discipline, named the thing under the thing:

We have never had a measurement problem. We have an ownership problem.

Sit with that, because I felt it land. The admission ticket gets you into the room. Ownership is what happens once you are there. Every seasoned VoC practitioner has lived the version underneath that word. It has nothing to do with whether you heard the customer. We heard them fine. We always have. What was missing was never the signal. It was the empathy to feel why it mattered, the care to make it someone's job, and the will to act before the moment closed. The feedback came in, the report went out, and the silence on the other end was a small betrayal the customer felt every time.

Inaction compounds

Inaction does not just fail to add value. It subtracts it, one cycle at a time. Every survey that lands on nothing teaches the customer the same small lesson: my voice does not move anything here. So they tell you less, and then less. Silence is not the absence of a signal. It is the interest compounding on every signal you ignored.

Here is the symmetry that matters. Inaction compounds against you. A loop that acts compounds for you. The same force that drains a feedback program when nothing happens is the one that makes it valuable the moment something does.

Look at what Medallia is building now: conversational feedback, action orchestration, agentic automation. The incumbent that defined collect-and-report is rebuilding itself around action. When the company that wrote the playbook quietly rewrites it, that is the market telling you where the value moved.

So when was the last time a customer told you something and actually felt you do something about it?

We have been excellent listeners for twenty years, and it cost five billion dollars to prove listening was the easy part. The job itself changed. The new unit of the work is a loop that ends in action, not a slide.

A system of loops

If the new unit is a loop, the next question is what kind. Loops are having their moment right now, and for good reason. The people building agents have spent the last couple of years naming a handful of them, and the vocabulary is useful far beyond the people who write the code.

There are four worth knowing. The agent loop does the work: a model using tools, over and over, until the task is done. The verification loop makes it trustworthy: a grader, a rubric, or a second model as judge that checks the output and sends it back when it fails. The event-driven loop makes it proactive: a schedule or a signal fires it, so it runs in the background instead of waiting to be asked. This is where it stops being a chat and becomes part of the system. The hill-climbing loop makes it improve: an analysis pass reads what happened across many runs and rewrites the loop itself, better prompts, better checks, better routing. This is the one that compounds.

That is the whole architecture underneath this piece. The agent loop is the "it has to act." The verification loop is how you trust it. The event-driven loop is how it keeps pace without being asked. The hill-climbing loop is the "it has to compound."

Run one account through all four. The event-driven loop fires when an account that pays you real money goes quiet and a routine ticket turns sharp. The agent loop drafts the save and routes it to whoever owns the relationship. The verification loop checks the read is grounded before it goes out, and escalates to a human when it is unsure. The hill-climbing loop watches which saves actually worked across many accounts and gets better at catching the drift earlier. That is the system. What follows is what it takes to commit to it, starting with the loop that acts.

It has to act

A smoke detector that emails you a quarterly summary of the fire is not a smoke detector. It is a historian.

That sounds absurd, and yet it describes how most customer intelligence actually works. The smoke is detected. The reading is logged. The summary is beautifully formatted and distributed on a predictable cadence. And the house still burns down, because nobody confused the summary with the act of putting out the fire. A detector that only describes the fire has become a very thorough way of being too late.

That is the failure under the failure. Not that the report was wrong. The report was usually right. It just arrived as a description of a moment that had already closed, to people who then had to start the actual work from scratch. A description, no matter how accurate, is not the same as something happening.

Two of those four loops are the commitments that matter most, the kind you make before you write a line of anything. The first is the agent loop. It has to act.

Not surface the insight. Not route the ticket to the person who surfaces the insight. The loop has to close on an outcome you can point to, and carry the proof that the outcome happened. Design the thing backward from the result, not forward from the data, and you stop optimizing for coverage and start optimizing for what moved.

There is a cost to getting this wrong, and it has a name. Call it ownership latency. Every day the gap stays open between a customer saying something and the company doing something about it, the function that owns that gap depreciates a little. It is trust leaking out of a relationship in real time, and the slow conversion of a team that was supposed to own outcomes into a team that produces summaries of outcomes it did not influence. You do not feel ownership latency in a single quarter. You feel it the way you feel deferred maintenance, all at once, much later, as a write-down.

That raises the obvious question: who owns it? My rule is to never invent new accountability, only borrow the accountability that already exists. Map each action to a metric someone already answers for. Whoever owns platform reliability, enterprise churn, product adoption, or return rates should own the loop acting on the feedback behind those numbers. You are not creating a new job. You are pointing the feedback at the accountability that is already there.

It has to compound

The second commitment is the hill-climbing loop, and it is harder, because it has to survive the obvious objection. The models keep getting better, exponentially, and they are not going to stop. So won't the model eventually do this part too?

The trap is in the question itself. The model improves on a curve everyone rides. Every leap arrives for you and your competitor on the same morning, so it can never be the thing that sets you apart. When models improve for everyone, your durable advantage comes from one place: a loop that learns and acts on your own customer feedback, and compounds with every turn. So you stop spending yourself on the curve you cannot win, and pour everything into the turns of the loop you can.

But compounding is not automatic, and this is where most people go wrong. Most companies do not have a compounding loop. They have a hoard. Ten years of tickets, a data lake, a graveyard of dashboards, and none of it climbs, because a pile is not a loop. It sits there and rots while the customers it describes move on. A loop compounds only when it acts on what it learns, remembers the correction, and turns faster than the problem it is chasing. Collection dressed up as context gathered everything and learned nothing.

Which means the question was never whether the work survives the machine. The question is what you have been building underneath it the whole time: a loop that compounds, or a pile that rots.

Build the loop

Do not build a Voice of the Customer program that stops at the report. Build loops that carry the feedback to the people who already care, get them to act, and then close back to the customer who raised it.

That last step is the one everyone forgets, and it matters most. A loop that acts but never reports back to the customer is only half a loop. The person who spoke up is still waiting to learn whether it mattered, and when they never hear, that silence is what teaches them to stop speaking up.

A program that depends on one person carrying every report into every room is not a function. It is a bottleneck wearing a lanyard. That is the part that quietly breaks the people who run these programs: not the volume of feedback, the begging. For a long time there was no alternative, because closing the loop took a person doing all of it by hand. That is what changed. The loops that work now are designed to the strengths of both: the agent carries the volume, the monitoring, and the drafting, the parts that never should have been a person's job, while the human keeps the judgment, the care, and the hard tradeoff call a model should not make alone. Build it that way, and the work stops waiting on any one person. The feedback reaches the owner, the action happens, and the customer hears back, whether or not you are in the room.

So here is the whole job, the one underneath all the slides: shrink the distance and the time between a customer saying something, the company doing something, and the customer hearing back. From said, to done, to heard.

The report was never the work. The report was the ticket. The work was always the loop, and now you get to build it.