Don't Automate Dysfunction

What the Quality Revolution Can Teach Us About Leading AI

I've been listening to the urgency around AI lately.

Move faster. Experiment. Find the use cases. Automate the routine work.

Much of that advice is right.

But it also gives me an uneasy feeling that I've seen this movie before.

Years ago, while working at General Motors' Rochester Products Division, I was leading a team trying to promote the thinking of Dr. W. Edwards Deming on the factory floor. At about the same time, GM was learning extraordinary lessons from NUMMI, its joint venture with Toyota.

NUMMI gave GM an opportunity to look inside the Toyota Production System.

We looked.

I'm not sure we always understood what we were looking at.

I remember a confidential GM document that came out of that work. We studied the practices. We learned the tools. And then, like many companies, we began cherry-picking them.

We saw the tools without fully appreciating the word system in Toyota Production System.

In Wisdom Orchard language, we were looking at the trees without paying enough attention to the soil.

The Soil Matters

Many of the practices we admired at Toyota rested upon assumptions about people that weren't nearly as visible as the tools themselves.

One was especially important:

The people who do the work know things about the work that management doesn't.

Improvement wasn't supposed to be something done to them. They were supposed to be deeply involved in doing it.

Yet as American companies embraced continuous improvement and Lean, we often professionalized improvement.

We created departments for it. We trained specialists. Engineers learned the tools and did much of the improvement work.

The people on the floor participated, certainly.

But too often they were implementing someone else's improvement.

They complied.

That isn't the same as commitment.

What Management Doesn't Know

Deming frequently reminded us of an observation he credited to his friend and fellow statistician Lloyd Nelson: some of the most important figures management needs are unknown or unknowable.

I remember Deming using customers to make the point.

A dealership can count the people who walk through the door. It can count cars sold, margins and customer satisfaction.

What it can't count very well are the people who never walked through the door because of something they heard from someone who had a bad experience.

They aren't in the data.

But they are certainly in the results.

Some things are unknowable. Other things are simply unknown because people don't feel safe telling us what they know.

The people doing the work may already know the new process doesn't work. They may know the metric being celebrated in the conference room bears little resemblance to what is happening on the floor.

The question is whether it is safe to say so.

The healthier the soil—the greater the trust, transparency and safety—the more of that knowledge becomes available to the organization.

Leaders cannot eliminate the unknowable.

But they can stop making the organization unnecessarily ignorant.

I've Seen This Movie Before

That's what worries me about some of today's conversations about AI.

Organizations understandably ask:

Where can we use AI?

What can we automate?

How much productivity can we gain?

I'd start somewhere else:

What problem are we trying to solve?

What do the people doing this work know that we don't?

What are they reluctant to tell us?

What should this work look like if we redesigned it together?

Then: Where can AI help?

Otherwise, we risk repeating an old mistake—cherry-picking the visible tools while missing the system that makes them effective.

Only this time we'll be able to make the mistake much faster.

Catch the Ball

There's an idea from Hoshin Kanri that I've always liked: catchball.

Direction doesn't simply travel downward as an instruction. The ball moves back and forth. People catch it, examine it, contribute to it and throw it back.

Eventually, the people doing the work aren't merely executing somebody else's plan.

They helped create it.

They've caught the ball.

That's how I'd approach AI.

Ask the people doing the work where the friction is. What wastes their time? What do customers complain about? What requires human judgment? What does management misunderstand?

And perhaps most importantly:

What do you know that you don't think we want to hear?

Then listen carefully to the answer.

New Technology. Old Leadership Truth.

AI is different from the quality revolution in enormously important ways. Its capabilities, reach and speed are unprecedented.

But human beings haven't changed nearly as quickly.

People still know things leaders don't.

They still want some voice in decisions affecting their work.

They still distinguish between something being done with them and something being done to them.

And compliance is still not commitment.

Before you automate the work, understand the work.

Before you redesign the process, listen to the people living inside it.

Before you celebrate what the dashboard tells you, ask what it cannot tell you.

And before you plant the newest management idea, take a good look at the soil.

Because even the best idea struggles in soil that cannot sustain it.