Killing our own project

As she pulled the file up on her screen and started scrolling, I couldn’t believe what I was seeing. Hundreds of comments, then thousands, maybe tens of thousands, each one written by a patient, in their own words, about the experience we had been brought in to improve. The list seemed to go on forever. I was looking at a treasure trove of rich data, and the more I heard her talk about it, the more I realized it wasn’t being used to its fullest potential.

She was one of the chief experience officers we were interviewing across the National Capital Region, eleven military treatment facilities in and around Washington, D.C. I had arrived mid-project. The work I inherited was already in flight: a printed passport booklet, with phone numbers, directions and a map to help patients find their way around the hospital.

The booklet had its reasons for being. But my intuition kept telling me it would never change the patient experience in any systemic way. It treated a symptom and left the body as it was. Patients arrived in pain or distress, physically or mentally, and the booklet asked them to keep hold of it, read it, find their glasses, and navigate a large building from a map on paper. It added load to people who had come to us to have some taken away.

The file on her screen asked nothing more of them. The patients had already done their talking: over a hundred thousand open-ended comments and growing, delivered by a third-party survey agency as a raw Excel export. People were listening, but they could only hear a few handpicked voices. They couldn’t make sense of what the entire crowd was saying. Somewhere in all those rows was the whole story.

What I saw in that file was the most honest account of patient experience the network had, and it kept arriving whether anyone could use it or not. The network was making a trade-off it never chose: acting on the comments someone had time to read, and missing what the rest were saying. My job became removing that trade-off.

The question changed with it. Instead of guiding one patient down one hallway, how could the network hear the whole story, see which problems touched the most people, and know where to start?

Answering it meant stepping off the project I had inherited and bringing our client with us. I built the case, in collaboration with my design strategy partner. What won it was a prototype. It turned their own patients’ raw comments into something they could act on, and nobody could deny it. My bet was that fixing the system patients moved through would do far more for them over time than any booklet could. It meant walking away from a deliverable the client had already bought, mid-project, and taking on a much heavier build. They said yes. Now we had to build it for real, before large language models were everywhere.

We brought on two data scientists, one with the engineering background to build the pipeline. But first the machine had to learn to read, and someone had to teach it. That was me. I went through thousands and thousands of comments, one at a time, in the same file she had scrolled for us, tagging the sentiment of each so the natural language processing had something to learn from.

What came out the other end did what no amount of scrolling could. Every new batch of comments arrived sorted by topic, sentiment and critical issue. You could see what came up over and over, where the biggest pain points were, and what needed attention now. Much of it had been heard before, one patient at a time. Now those voices carried weight, because they could be counted, and the count showed people where acting would matter most.

The last layer came from something we learned while testing the prototype, and it became one of the things that drove adoption. The facilities were intensely competitive with one another, and they were already tracking their benchmark scores against each other. So I decided to put each facility’s performance beside the others’. No facility wanted to be the one at the bottom of that list.

The first sign came from our deputy director. We walked through the top ten comments in the dashboard together, and one of them was about the cleanliness of the bathrooms, something the deputy director hadn’t known. It came as a surprise, and an upsetting one, and it was acted on right away.

I knew this was going to be something big when the chief experience officer who had first scrolled through that file for us told us she had taken the dashboard into a planning and leadership meeting with her executive peers. They were thrilled with what they saw, and the conversation in that room turned to what the patients were telling them.

It was adopted and scaled out across the facilities, and it outlived our engagement. After I moved on to another project, it was escalated one level up, to the region’s parent, the Defense Health Agency, as something to fund and scale across all of its facilities.

I had pushed us to set aside the project I inherited to build it. When I first saw that file, the only way to hear those patients was to scroll and hope the right comment caught your eye. In the end, the whole story, told in their own words, finally had somewhere to go, and the people who could act on it knew where to start.

We gave the booklet our full effort, and the research we did for it is what led us to the survey. What I took from it is to do the primary work as well as it can be done, without becoming so focused on it that you miss a bigger opportunity beside it. Research is abductive: new evidence connects to what you already know, sometimes pointing down a road you hadn’t planned, and a systems lens shows where those connections lead. When one of them leads somewhere bigger, it is worth having the nerve to leave the path you were already on.