The moment you think AI will do everything, improvement stops
At the end of this article, there is information about a free self-diagnosis tool I created.
There was a workplace where transcription errors were occurring in Excel.
The countermeasure was 'try harder to check visually.' This is a common story. However, this only tires out the person doing the checking, and the system itself does not change at all.
However, it is not that simple to say that if you leave all the transcription work to Copilot, it will be solved.
In this case, you can use Excel's conditional formatting to color suspicious cells. You can use functions to pick up input omissions or numerical discrepancies. First, you can create a state where errors are visible using standard Excel functions.
Copilot's turn comes when you have it think about those settings with you.
'I want to color cells if this column and that column are different'
'I want to be able to see only the cells that have not been filled in'
You should have it build the necessary functions and setting procedures from such consultations. Instead of handing over the entire job to AI, you use AI to create tools to fix the job.
Please check the area where you are having the most trouble right now first.
→ Free '7 Classifications Self-Diagnosis for Why AI Utilization in the Workplace Stops'
I used to be a race mechanic. What came back from the riders were sensory words like 'it feels jerky' or 'it slips smoothly.' You cannot give orders to parts using those words as they are. You have to separate when it happened and how the vehicle body moved to finally connect it to adjustments.
Actually, something like this happened the other day. This is one example.
I created a workflow table and Mermaid diagram in 30 minutes from a one-hour meeting record. Then, I deliberately sent it without checking it myself. The next day, redlines were returned, and I made corrections based on that. The entire workflow was documented in three days, the test, which was originally scheduled for one week, started on the third day, and it proceeded to operational judgment by the fifth day.
'Sending without checking' does not mean I threw it away sloppily. It is faster for someone who understands the work to fix it than for me to fix it. Identifying unclear points is the job of a draft created by AI. If you are going to remake it anyway, there are times when inserting a process where you fix it yourself becomes a waste.
Of course, this is a way of proceeding with someone with whom there is a relationship of trust within the company. For the first delivery to an external party, it goes through human eyes.
Consulting with AI is similar.
'It doesn't work as I thought' is not enough to know where to fix it. Is it the way of asking? Is it how internal documents are placed? Is it the structure of the Excel file? Is it authority or licensing?
If you get the location wrong, you won't move forward no matter how many times you rewrite just the prompt.
AI is not omnipotent. But it is a capable tool.
Before automating everything, choose one current stumbling block. If it can be done with a normal function, a function is fine. If organizing materials comes first, organize them first. It is better to start from having it help with that judgment, as that is how things move in the field.
Therefore, I have categorized the reasons why AI utilization stalls into seven areas and created a self-assessment to first verify the assumption that "AI equals full automation." You can identify where to look first by answering 16 questions, without needing to input company names or internal documents.
Please start by checking only the area where you are currently having the most trouble.
→ Free "7-Category Self-Assessment for Why AI Utilization Stalls in the Workplace"
If you can only access Copilot
For those who find that document creation or agents are the bottlenecks in the assessment, and who can only use Microsoft 365 Copilot within their company, I have prepared a 5-piece set tailored for that environment. It includes 54 pages of instructional material, 8 prompts, and a tool to verify what you can do in your own environment.
Think of this as a way to implement the concept written in this article—"don't hand over the entire job, but create tools to fix the work"—specifically for document creation. It is ¥4,980 until September 30th.