I Forgot to Invite the Guy Who Built the Spreadsheet

The first thing I did in my first real meeting with the man whose spreadsheet we were replacing was apologize, because by then the replacement had been in his coworkers’ hands for a couple of weeks and he’d heard about it from one of them.
Some background on how I got there. A manufacturing client asked me in August to help their operations team, and the obvious place to start was the production schedule, which lived in one enormous Excel file that the whole team worked in at once. The scheduler told me he spends 30 to 40 percent of a 50-to-70-hour week on spreadsheet work, most of it copying numbers from one file into another and then checking them two or three times because he doesn’t trust the copying. People on the plant floor needed edit access just to print the thing, and once somebody trying to get a printable view deleted every formula in the file while leaving all the data in place, which I think is the most polite way a spreadsheet has ever been destroyed.
So we built software, and I do mean we. The VP who runs the operations team had only recently learned to use Claude, and within a few weeks he had built working prototypes of his own, including a dashboard that catches orders that have stalled and a tracker for inventory. The scheduler supplied the line speeds, the changeover times and every rule about what can run after what. My part was to put a real database underneath their ideas and turn the prototypes into one tool, with Claude writing the code. By late September there was a working schedule you could drag jobs around in, with capacity warnings and a queue of orders waiting for a slot. When we first showed it around, about $250 had been spent on it, and I said that number out loud to more than one person, usually followed by what a software firm would have charged. I was quite pleased with myself, and most pleased that the team had built it with me. The scheduler said the first prototype would cut his week in half, and we set up a standing Friday session to keep building.
Then we kept going, which is where I got it wrong. The schedule needs to know what’s about to be ordered, so we added a page for restocking and a page for custom orders, and I filled both with data from a tracker that the number two on the fulfillment team had built and kept up by hand for about three years. I took his file and left him out. I had been telling myself that this was software the team was building together, and the team I had in mind did not include the person whose work we were building on. He’d been on a few group calls with me, but we had never had an hour together, and nobody had asked him what the tracker was for. Around the same time the scheduler, who was excited, told the wider team to start using the new tool before everyone had a login. So the man who knew the most about how orders move through that company found out that his work was being replaced by a tool he couldn’t open, built by a group he wasn’t part of, with help from a consultant who hadn’t called.
He was upset, and he was right to be. We finally sat down in early November, eleven weeks after I started with his team, and what I said was, “Man, I’m so sorry you weren’t included earlier in the process. I had no idea you worked across the entire production process.” The second sentence is the embarrassing one, because eleven weeks is plenty of time to find out what somebody does. He was gracious about it, and then he opened his spreadsheet next to the new tool and started comparing the two.
It took him a few minutes to find something the rest of us had missed for weeks. He pulled up one product, and the inventory page said there were about 121,000 on hand, when he expected roughly double that. The page was counting what sat at the plant and ignoring what was already in Amazon’s warehouses, which meant the days-of-supply figure was wrong and so was the size of the production run the tool was suggesting. His spreadsheet had a column for Amazon stock, naturally, because at some point in those three years he had learned exactly what happens when you leave it out.
That hour changed how I think about these projects. I had been treating his spreadsheet as the old system, the thing to be moved off of. But every column in it was there because something had gone wrong once, and the orange cells and the red cells and the notes to the scheduler were all decisions he had made about what deserves attention. The file was a record of what he knows. I had copied the record and skipped the person, which is a bit like photocopying someone’s notebook and assuming you’ve taken the class.
He also told me, kindly, what it would take to win him over. A new tool that is about as good as his spreadsheet is of no interest to him, because his spreadsheet already works and he’s a creature of habit. It isn’t pretty, he said, but it works. Where the new thing makes the team better he’s all in, and where it only trades one system for another he’s going to be slow to move. By his reckoning it has to be about ten times better. I think that’s the most useful product requirement anyone has given me this year, and it came from the person I had left out of the requirements.
Since then I’ve noticed how much of my Friday bug list has nothing to do with code. One customer showed up as two different customers because the name had been typed once with an apostrophe and once without, and I had loaded the typed names instead of the customer numbers. Someone changed the weekly hours setting from 84 to 80, and every utilization figure went up without a single extra unit being made. I also tried to speed up adoption with a leaderboard and a weekly $50 gift card, and in the first week the scheduler, the person the whole thing was built for, came in last with one point, while someone in finance asked the fair question of why we were paying people to do their jobs.
I tell clients all the time that building has become cheap enough that you can just try things, and I still believe it. But I drew the wrong conclusion from my own advice. When the software costs $250, the software is no longer where the cost is. The cost has moved to the hours you spend with the people who know why the old way works, and I skipped those hours because the building was so fast that stopping to talk felt like the slow part.
If you’re about to replace a spreadsheet that runs part of your company, find out who built it and book an hour with them before anyone writes any code. Ask them to open it and tell you what each color means and why each odd column exists. Then, when you have something to show, put it next to their file and ask them to find what’s wrong with yours. It will take them a few minutes, and it costs a lot less than an apology.
We’ve met again since, and he’s now the one telling me what the custom orders page should do. The old Excel schedule is due to be retired after Thanksgiving, although the plan is to lock it and keep it around, which seems right to me.
