Jeff's Daily Dose: Following Your Employees Home.

My first job out of grad school was at Intuit ... makers of QuickBooks and TurboTax. I had no idea that what I learned there would be the single most useful lesson for leading through AI ... 30 years later.

Here's what happened.

Intuit's co-founders, Scott Cook and Tom Proulx, had a problem.

They'd built Quicken, personal finance software they thought was intuitive. Customers said it was great in surveys. But support calls kept flooding in.

Surveys lied. Focus groups lied. People told Intuit what they thought Intuit wanted to hear.

So Scott & Tom invented something radical. They called it "Follow Me Home" research. The concept was dead simple: Ask a customer buying the software at a retail store if an Intuit employee could literally follow them home and watch them use it. No helping. No guiding. Just observing.

(People said yes, kinda bizarre.)

What they discovered was brutal.

People didn't read the instruction manual. They didn't follow the Setup Wizard. They invented bizarre workarounds that no engineer would have predicted. The gap between how Intuit designed the experience and how people actually experienced it was ginormous.

That gap changed everything about how Intuit built products. We adopted the assumption that nobody would ever read the instruction manual. And this should change everything about how you approach AI in your organization.

Here's the connection:

Right now, most leaders I talk to are doing one of two things with AI. They're either (1) dabbling with tools themselves & assuming their experience reflects their team's reality. Or they're (2) surveying their people ... asking "Are you using AI?" and "Do you find it helpful?" ... and getting polished, useless answers.

Both approaches have the same flaw Scott & Tom discovered in the early '90s.

People don't do what they say they do.

Your marketing director will tell you she's "using AI for content creation." What does that actually mean? Is she pasting entire strategy docs into ChatGPT and blindly publishing the output? Is she using it to brainstorm headlines and then rewriting every one? You have no idea ... and neither does she, really, until someone watches. Literally.

So here's what to do this week.

Pick 2 or 3 people on your team. Not your most technical people. Not your most resistant people. Pick people in the messy middle ... the ones who say they're "figuring it out."

Ask them this: "I'd love to spend 30 minutes watching how you work. Not to evaluate you. I just want to understand what your day actually looks like right now."

Then shut up.

Watch them move between tabs. Watch where they copy and paste. Watch where they pause, think, and do something manually that makes you wince. Watch where they don't use AI but probably could. Watch where they do use AI but probably shouldn't.

Don't coach. Don't suggest. Don't say "Oh, you know what you should try..."

Just observe.

(And ask them to say outload what they're thinking. You're no mind-reader.)

When you see something interesting, don't ask "Why do you do it that way?" ... that sounds like a challenge. Instead, ask them to narrate: "Talk me through what you're thinking right now." That's the gold. That's where you hear things like "I don't trust the AI output so I rewrite it anyway" or "I didn't know I could upload a file" or "I spent 40 minutes on this prompt and gave up."

What you'll find will surprise you.

You'll discover that your "AI Ready" people have massive gaps. You'll discover that your "resistant" people have legitimate workflow concerns you never considered. You'll discover that the tools you invested in aren't being used the way the vendor demo promised.

Most importantly, you'll discover the real barriers to adoption. Not the ones people report in surveys. The actual, observable, in-the-moment friction points.

Now do something with it.

After your observations, sort what you saw into 3 buckets:

(1) quick wins ... things you can fix with a 5-minute explanation or a prompt template.

(2) training gaps ... real skill deficits that need structured learning.

(3) process problems ... workflows that need to be redesigned before AI can actually help.

That third bucket is the one everyone misses. You can't AI-enable a broken process. You'll just get a faster broken process.

Scott Cook built Intuit into a $110 Billion company on a deceptively simple insight: Go watch people do the thing. Don't survey them. Don't assume. Don't guess.

The leaders who will win the AI transition aren't the ones with the best tools.

They're the ones who truly understand how their people work today.

Go follow someone home this week (Metaphorically, of course. No need for legal issues.)

1 comment