APM in Practice: How an Automation Project Is Executed From Start to Finish

Deciding to automate is rarely the problem. The problem is what happens between the decision and the result: how much is invested, what's expected, and who responds when something doesn't go as planned.

To show it, let's follow an illustrative case built with typical figures from the diagnostics we run: a services company in Colombia, between 51 and 200 people, that wants to automate purchase order approvals.

The process: 120 requests a month. Each one takes an average of 90 minutes of work spread across six people, from the requester to the finance manager. Emails, reminders, a shared spreadsheet, and approvals that sit waiting when the approver is traveling.

This is what that same process looks like as an automation project with ROI, from start to finish.

Before: Guesswork

Before APM, the conversation sounded like this: "this takes up a lot of our time, let's automate it." Nobody knew how much time, how much it cost, or which part of the process was worth touching.

The first step was turning the feeling into numbers. 120 cases at 1.5 hours each is 180 hours a month, the equivalent of 1.07 full-time people. With an average salary of COP 3.5 million plus benefits, an hour of work costs about COP 31,900. The process consumes roughly COP 5.7 million a month.

But that wasn't the most useful number. The most useful finding was where the time went: 60% of the waiting wasn't in reviewing orders, it was in chasing approvals. And only one person, the purchasing coordinator, knew the status of each request. If she was out, the process stopped.

With that, the business case stopped being a hunch: 70% of the process automatable, projected savings of COP 4 million a month, a first-year investment of COP 30 million (implementation plus license), and an estimated payback of 6 months. Projected first-year ROI: 61%.

What would not be automated was also written down: orders above a certain amount and purchases from new suppliers would still go through human judgment. That decision is worth as much as the previous one, and it answers the same questions that define ROI before automating.

During: Control

In a traditional project, implementation is a black box: requirements are handed over and everyone waits for the go-live date. With APM, execution has review points with business criteria, not just technical ones.

Three decisions shaped this phase.

→ The order. The team didn't start with the request form, which was the most visible part. It started with the approval routes, which is where the process stalled. It's the logic of deciding what to automate first, and what not to.

→ The exceptions. Every case that didn't fit a rule got an owner and a criterion: who decides, with what information, and how fast.

→ The role of people. The purchasing coordinator stopped being the place where every request's status lived. She moved on to leading the process: she monitors the dashboard, decides the exceptions, and proposes adjustments. The process no longer stops when she's out, and her judgment carries more weight than before, because now it's applied where it matters.

⚠️ In week five, a wrong assumption surfaced: almost a third of orders arrived with incomplete data, and the flow kept sending them back. With an improvised method, that would have been fixed with a patch. With APM, it was logged as a finding, the form's entry rule was adjusted, and the business case was updated with the real data.

That kind of control is what makes it possible to run the automation without depending on whoever built it.

After: ROI Measured

After three months of operation, the first measurement was done with real data, not the business case figures.

Time per request dropped from 90 to 34 minutes. The projection said 27. The process automated 62% of the work, not 70%: orders with incomplete data and some supplier exceptions still required intervention.

With that real figure, monthly savings came to COP 3.56 million, first-year ROI to 43%, and payback to just over 7 months.

💰 Less than projected, and still an investment that pays for itself before the year is out. The gap between the promised 61% and the real 43% isn't a failure. It's the information that makes the next calculation more precise, and it's exactly the point of measuring the real ROI, not the promised one.

The measurement also showed something no projection included: approvals that used to take up to five days are now resolved in less than one. Suppliers get answers on time, and the purchasing team stopped spending its mornings chasing signatures. That time now goes into negotiating better terms, something there was never room for before.

With the real data, the second cycle already has a clear goal: fix data capture at the source to bring the process closer to 70% automatable. That turns the project into a system that learns, not a one-time delivery.

Why Decisions Are No Longer Made Blind

What changed at this company wasn't just the purchasing process. It was the way decisions get made.

Before, every automation was justified by urgency or enthusiasm. Now there's a method: measure before, control during, verify after. Every project leaves real data that improves the estimate for the next one.

🌀 And people didn't leave the process. They changed their place in it: they stopped holding it up by hand and moved on to managing it. Technology runs the repetitive work. People think, decide, and improve.

That's the fundamental difference between APM and traditional automation: one delivers tools, the other builds the capacity to keep deciding well.

An automation that isn't measured before, during, and after isn't a decision, it's a bet with a budget.

If you can't say today how much your last automation returned, start there.

📌 Write to me and let's talk about a process.