The Cummulative Flow Chart That Was Still Telling the Truth

A CTO I’d coached years earlier invited me for coffee, mostly to catch up. Partway through, almost as an aside, he mentioned the team’s delivery metrics had stopped making sense to him. The process was unchanged, he said — same board, same chart, updated the same way it always had been. Lately, though, nobody on the team could use it to find where work was actually getting stuck.

I asked if I could see the chart. He pulled it up: a cumulative flow diagram, one column added per deployment, set up exactly the way we’d built it together years before.

At a glance, it looked fine to me too. The columns ran across the screen at even width, each one next to the last, and some part of my memory of this team — from when they deployed roughly every week — read that even spacing as even time.

It took longer than I’d like to admit to notice what the chart was actually counting. Not weeks. Deploys. The width of each column had never meant a fixed stretch of time; it meant one event. For years, that distinction hadn’t mattered, because the team deployed often enough, and regularly enough, that one event and one week were close to the same thing. Somewhere since I’d left, they’d stopped being the same thing, and the chart kept rendering exactly as it always had.

Once I asked the right question, the real change turned out to be the deploy decision itself. It used to happen whenever the work was ready. Now it happened when someone with the authority to say so said so — which meant finished work queued up behind that decision the way water queues behind a lock gate, and released in a lump whenever the gate opened. The chart didn’t show a lump. It showed one column, evenly spaced next to the one before it, because a lump is still just one event to a chart that only counts events.

The team hadn’t lost the ability to read their own metrics. I had, briefly, lost it on their behalf — carrying an assumption from years earlier into a chart that no longer supported it. And the actual problem was never the chart. Somewhere along the way, a gate had been added to a system that used to run without one, and the gate was doing exactly what gates do: holding inventory out of sight until someone chose to let it through.

💬 Want more conversations like this one? Conversations lands once a month, straight from the room. Real client and prospect exchanges, anonymised, each built around a moment where listening changed the outcome.