The Real Problem Is Not a Lack of Ideas
When we speak with COOs and CFOs at industrial companies with between 50 and 500 employees, the conversation rarely starts with "I don't know what to automate." It starts with the opposite: there are too many things that could be automated, and it's not clear where to begin.
The production team wants to automate shift reports. The procurement team wants to eliminate manual reconciliation with suppliers. Finance has been asking for months that the month-end close stop depending on five people exporting Excel files at eleven at night.
Everything feels urgent. Everything has an internal advocate. And the usual result is that nothing gets started — or something gets started that doesn't move the needle.
This article describes the criteria we use with manufacturing and industrial companies to sort that list and choose the first process to work on.
Why the Criteria Matter More Than the Technology
Before discussing AI agents or workflows, there is a more important question: which process, if automated first, produces the greatest return with the lowest adoption risk?
The answer is not always the most complex process or the most visible one. In manufacturing, the processes that generate the most value when automated first tend to share three characteristics:
- They run at high frequency — daily, weekly, or per shift.
- The cost of error is measurable — an incorrect data point produces an incorrect decision, a delay, or a rework with a real cost.
- They require manual data preparation before anyone can act on them — someone consolidates, cleans, or formats information before another person can make a decision.
When a process meets all three conditions, automation has an immediate impact and the team notices it quickly. That is what makes adoption work.
The Criteria in Practice: Three Questions to Rank the List
When we work with an industrial company, we ask the team to list between eight and twelve candidate processes. We then apply three questions to each one:
How many times does this process run per month?
A process that runs 20 times a month has more cumulative savings potential than one that runs twice, even if the second one seems more "important."
What happens when there is an error in this process?
If the answer is "it gets corrected in ten minutes," the impact of automating it is low. If the answer is "it delays the financial close," "it generates an incorrect purchase order," or "production stops for an hour," the impact is high.
How much time does the team spend preparing data before executing the process?
This is the most underestimated indicator. In manufacturing, it is common for 60–70% of the time spent on an "analysis" process to be, in reality, data consolidation time. That is the time AI recovers most easily and most quickly.
A Concrete Example: Production Reporting at a Mid-Size Industrial Company
A manufacturing company with three plants and approximately 180 employees had a daily production reporting process that worked as follows: each shift supervisor completed an Excel spreadsheet, a coordinator manually consolidated them every morning, and the COO received the consolidated report around 10:00 AM with data from the previous day.
The process ran every business day. Consolidation errors produced incorrect planning decisions at an estimated frequency of two to three times per week. And the coordinator spent between 45 and 60 minutes each day solely on consolidating and formatting.
Applying the three-question criteria, this process scored high on all three dimensions: daily frequency, measurable cost of error in planning, and significant data preparation time.
We implemented an agent that pulls data from each plant directly from existing sources, consolidates it without manual intervention, and generates the report in the format the COO already used. The coordinator stopped spending those hours on a consolidation task, and the report became available before 7:30 AM.
In terms of estimated impact: between 15 and 20 hours recovered per month on the coordination team, consolidation errors reduced to virtually zero, and planning decisions based on same-day data rather than the previous day's. For a company of that size, the cost of planning errors avoided can represent between 8.000 and 20.000 euros annually depending on production volume, though that range varies by sector and operating margin.
What Doesn't Work: Automating What Is Visible, Not What Is Costly
The most common mistake we see in industrial companies is prioritizing automation by internal political visibility rather than by real impact. The process chosen is the one the general manager mentioned in the last meeting, or the one with the most internal advocates, or the one that seems the most "technological."
The usual result is a project that takes longer than expected, that the team doesn't adopt with conviction, and that doesn't produce the return that justified the investment.
The criteria of frequency, cost of error, and data preparation are not sophisticated. They are deliberately simple. But in practice, they rank the list in a way that gives the first project a real chance of producing a visible result in six to eight weeks.
How to Apply This at Your Company
The prioritization exercise described here takes between two and four hours with the right team. It does not require an extensive diagnostic or a complete process map. It requires that the people who operate the processes can answer the three questions honestly.
If your company operates in manufacturing, food production, industry, or any sector with repetitive operations and dependence on manual data, there are likely two or three processes on your list that meet all three criteria. Identifying them is the first step.
The second step is to build the first agent on that process, measure the impact in the first few weeks, and use that result to justify the next one.
That is the model that works: not an eighteen-month transformation roadmap, but a concrete first result in six to eight weeks that the team can see and the business can measure.
Conclusion
When everything feels urgent, the right criteria is not the one that generates the most internal consensus. It is the one that maximizes real impact in the shortest time possible. In manufacturing and industry, that almost always points to high-frequency processes with a measurable cost of error and manual data preparation as the bottleneck.
If you want to apply these criteria to your operation with someone who has done it before, you can request a free diagnostic through the form at the bottom of this page. No commitment required, no prior call necessary.
