Reversing a churn spike

In August 2023 monthly order turn hit 10% - one order in ten, gone.

A retention program started that September, and nobody in the room needed convincing that it mattered - the argument was never about whether to act, only about what was actually wrong. The theories on offer were the usual ones, service quality and account manager workload and price, and each of them had somebody senior behind it.

Everybody had one.

I interviewed every account manager on the team, one at a time, asking which accounts they'd lost and why they thought they'd lost them. Then I checked what they told me against what they'd actually done: task completion, comments left on accounts, campaign management activity. All of it was already sitting in systems nobody had read next to an interview.

The third source was the customers themselves, by way of the cancellation reasons the partner collected from the people who were leaving.

Any one of those three would have been wrong on its own. An account manager's memory is shaped by the accounts that hurt, and activity data says what happened without ever saying why. A stated cancellation reason is a sentence written by someone who has already decided to go. It's often not the real reason either.

Put all three beside each other and the disagreements do the work. That's where the finding was.

What came back wasn't a service problem, and that was the surprising part: the work was getting done, mostly on time and mostly to standard.

The problem was the distance a message had to travel. This was a white-label channel, so a seller at the partner talked to the client and the client's expectations reached an account manager on my team. That account manager briefed a fulfilment team who never spoke to anyone outside the building, which is four layers, spanning departments and in some cases separate companies. Nothing governed how a message moved between them: not a document, not a template, no agreed shape for what a handoff had to contain.

So each layer restated the last one in its own words. By the fourth, the client's actual expectation had been through four translations with nobody checking the result against the original. Nobody was doing their job badly, and a chain with no standard degrades a message by default, which makes this structural rather than personal.

Structural problems are fixable.

Three things came out of it, and all three were about making a message survive the trip.

An expiring-campaign report put every renewal date in front of leadership on a fixed schedule. That sounds administrative. What it actually changed was the conversation: account reviews were now structured around a shared list instead of around whoever happened to remember what. That's a different meeting with the same people sitting in it.

Renewal presentations got built per account, from that account's own performance data rather than pulled from a general deck. There was documented proof that the accounts receiving one retained at a higher rate. The third piece was call support, sitting in on renewal conversations and running role-plays before the hard ones, which gave account managers a rehearsed shape for a conversation they'd otherwise improvise.

We hit the 4% target in January 2024, hit it again in May, and again in July, with quarter-over-quarter order turn improving underneath the whole time. The book more than doubled across the same stretch, and that's the part that matters. Holding a rate steady is easy when volume is flat.

Underneath the renewal presentations was a library of templates I'd built to produce them: proof-of-performance layouts, report shells, one-pagers, each one sourced from a template marketplace and restyled so an account manager could customize it without design help. I spent weeks on that, and months if you count the upkeep.

I'd cut all of it today, because a language model assembles that material in minutes now and it was never what made a renewal presentation work. What made it work was the analysis behind it: which numbers this client would care about, and what there was to say about them.

I spent my time on the container.

There's a second cost I'd weigh differently, and it came out of other people's time rather than mine. Reviewing accounts with account managers pulled them off live account work, and I ran most of those sessions in a group, on the theory that problems recur and one discussion would arm everyone for the next one. Some of it held up, and some was five people hearing about an account that wasn't theirs, which I'd now run one to one.

The tracker and the presentations earned their time. The template library didn't.