TL;DR
- A national operation doubled its revenue. My team ran the supply chain underneath it, and it had to carry twice the volume without breaking.
- Everyone assumes the win was the software. The win was the process. I mapped it first, fixed it, and automated it last.
- The real result was a team that learned to think like operators and could run the whole thing without me.
My first week, I sat in a status meeting and watched two managers argue for twenty minutes about whose numbers were right. Same company, same week, two different truths. Nobody could agree on what was actually happening, so nobody could fix it. That meeting told me exactly where to start.
When people hear that a national operation went from $150M to $300M, they ask about the technology first. Which platform did you put in? What did you automate? They want the shortcut, the tool that did the work.
The honest answer disappoints them. The automation was the last thing I did, not the first.
I have a name for this way of working now. I’m a forward deployed operator: I go inside an operation, find what is actually broken, and build the fix so it stays. Back then it was just the only way I knew. You automate a process after you understand it. And you understand it by watching how the work really moves, not by reading the binder.
What I walked into
Every report was built by hand. Someone pulled a set of SAP transactions, dumped them into a spreadsheet, and stitched a report together. Different people pulled different transactions, so they got different numbers. That is why the meetings started with a fight about whose figures were right. There was no single source of truth. There were as many truths as there were analysts.
That is exactly when people reach for a tool. A dashboard would have looked like progress. But a dashboard on top of a broken process just makes the broken process faster and prettier.
Step one: map it before you touch anything
I call this Process Archaeology now. Before I changed one policy, I mapped how the work actually happened, not how the binder said it happened. Who touched what. Where things sat and waited. Which “standard” steps everyone skipped because they didn’t work. Where the real decisions got made, and where the org chart only pretended they did.
Most people skip this step, because it shows no quick wins. Nobody claps for a map. But the map is what tells you which fixes matter and which ones are just theater.
Step two: fix the process
Only once I could see the real system did I start changing it. Inventory came down 10%, about $2.5M, because I fixed the ordering and stocking rules that were creating the excess in the first place, not because I squeezed anyone. On-time delivery climbed past 90%, because the process finally made the promise realistic and the misses visible.
A $100M procurement budget got one standard policy across Canada, instead of every region making up its own. Supplier vetting got real guardrails, so we stopped taking on risk we couldn’t see. And we moved a warehouse with more than 10,000 SKUs without the whole operation stalling.
None of that is glamorous. All of it is the actual work.
Step three: then automate
This is where estable platforms comes in, the part everyone assumes was the whole story. We moved off the manual, transaction-by-transaction reporting and onto one central data model. One source of truth, finally.
But look at the order. By the time we automated, the process underneath already worked. We weren’t building the mess into a database. We were making a working system faster, and giving everyone the same numbers to start from. It turns out that ends most of the arguing.
Automate a good process and you get more out of it. Automate a bad one and you just get a bad process, faster.
What I actually left behind
The part most operators don’t talk about, because it doesn’t flatter the ego. The goal was to make myself replaceable. On purpose, from day one.
By the time I left, the team ran without me. The reports ran themselves. The procurement policy held whether I was in the room or not. Suppliers got vetted by a system, not by my gut.
But honestly, the real handoff was the people.
I hadn’t just built processes for them to follow. I had trained them to think the way I think. Process first. Map how the work really moves. Find the root cause, not the loudest symptom. Then fix that. So when a new problem showed up after I was gone, they didn’t wait for me to fly back in. They ran the same play I would have run.
That is the whole point of a forward deployed operator. You embed, you map, you fix, you automate, and then you leave. But you do it right by making more operators, not more dependence. My team didn’t just keep the machine running after I left. They became forward deployed operators themselves.
The revenue number makes the better headline. But a team that doesn’t need you, because they have learned to see what you see, that is the real result.
What you can do this week
Pick one thing your team does over and over. Don’t write down how it is supposed to work. Pull the last five times you actually did it, and lay them side by side. The steps that show up every time are your real process. Name them in plain words your whole team will use, and let the next person follow that instead of guessing. You just did an hour of Process Archaeology on your own operation. Do that before you point any tool at it.
PS. Yes, I still map the first version by hand, on paper. It is slower, and it makes me actually see the work instead of skimming past it. 🙂
-Ysi
Vantelira
info@vantelira.com
⚡ Ops Intel — AI & Process
The same signal ran under this week’s AI news, and it all points back to process:
↑ AI agents cleared only about half of real business workflows in a new benchmark (Artificial Analysis, July 6). The best models finished about half of 657 real SaaS tasks, finance was the hardest domain at roughly a third, and every model broke at least some business rules along the way. An agent is exactly as safe as the process and the guardrails you hand it.
👀 Anthropic redeployed its most powerful model with new usage limits (Anthropic, June 30). Fable 5 came back July 1 on a capped weekly allowance with metered pricing behind it. Reach for the strongest model on your hardest slice of work and keep cheaper tools for the rest, and never wire a workflow to a model whose access can change under you.
↓ Gartner: companies abandon 60% of AI projects that lack AI-ready data (Gartner). Most AI failures trace back to messy inputs and undefined process, not weak models. Define the process and clean the data first, or the tool just automates the mess faster.