Don't Let AI Adoption End in Local Optimization | Visualize Operations from Logs and Redesign the Whole

Why Adding AI Tools Doesn't Change the Whole Company

If you hand writing and summarizing over to AI, a single task may get faster. But if the same information is re-entered in another department, and approval waits and send-backs remain, the flow of the work as a whole does not change. Partial time savings and results for the company as a whole must be considered separately.

I want you to resolve to reset all of your existing systems. The reset I mean here is not about destroying running systems or data all at once. It means looking at departments, the division of work, and approval flows afresh, without taking them for granted.

Separate Human Adaptation from the Structure of Headcount Needed

There is a problem in which people cannot adapt to work that assumes AI. Even when they understand in their heads the direction of rebuilding the work they have carried out until now on the assumption of AI and systems, their feelings may not keep up.

At the same time, if you change how work is built, the same number of people as before is not necessarily needed. The problem of psychological adaptation, and the structural problem in which the roles and headcount needed change through business redesign, need to be considered separately.

My position is that we should think about this including hiring restraint and headcount reduction. I do not believe that reassignment will solve it. This is not about making assumptions about the abilities or aptitudes of individual employees, but about separating the design of the company's work as a whole from judgments that take into account each person's contract and rights.

First, Visualize the Current State from Logs

The first thing to tackle is creating a business flowchart and a business description document from logs such as the company's chats, past meeting minutes, and manuals. Diagram the work from start to finish, across departments.

The diagram built from logs must not be treated as the correct process as it is. This is because there may be work that is not recorded, or outdated manuals. Have the people involved look at the diagram, and correct omissions and differences from reality. Internal information should be handled only within environments the company has approved for use, and only to the necessary extent. Collecting logs and being free to send them anywhere are different things.

Diagram Even the Exceptions and Rework

Take a fictional order-processing process as an example. Sales writes what they received in a chat, the administration department transcribes it into a spreadsheet, the person in charge requests approval, and if information is missing, it goes back to sales.

Even if you hand only the writing to AI, the transcription, approval waits, and send-backs remain. Confirm who receives what, what judgment they make, and what they pass on to the next step. Also diagram where things return to when an exception occurs.

Eliminate Unnecessary Steps, Then Divide the Roles

Before trying to automate every existing step, consider whether there are any unnecessary steps, or any steps that can be combined.

• Can duplicate entry of the same information be eliminated?
• Can approvals with clear conditions be turned into rules?
• Can the necessary information be gathered at the start to reduce send-backs?

After reviewing the business flow across a wide scope, decide the roles of AI and systems. Processing such as reading necessary information from text and organizing vague requests is a candidate for AI. Checking required fields, matching against fixed conditions, verifying permissions, and saving records can be designed as a system that runs on defined rules.

Rather than executing AI output unconditionally, decide what to check and which exceptions to return to a person. It is important to separate the parts where people make judgments from the parts where the system processes deterministically and records and audits.

Confirm the Effect by Results for the Whole

Do not judge the effect only by how many minutes faster a single task became. Look at the time from the start to the completion of the work, the rate of send-backs and exceptions, quality, and the total cost. Even if sales entry gets faster, if the administration department's checking increases, you may have only shifted the burden.

Taking a broad scope for the design and switching everything over at once are different things. After drawing the whole picture, verify it in a limited area of work, and move forward in a way that lets you revert if problems arise. Decide in advance what to compare against the conventional method and what kinds of problems will make you stop. Partial trials too can serve as verification for rebuilding the whole.

What to Start Doing Tomorrow

Choose one piece of work that spans departments, and create the current flowchart and business description document from chats, meeting minutes, and manuals. With that diagram in front of you, set aside for the moment the current assignments and headcount, and think about how you would rebuild it if AI and systems were available.

Think of decisions about hiring and staffing as a result of that business design. Do not just increase small uses of AI, but change how the whole company's work is built. I want to ask executives for that resolve.