Most companies are sprinkling AI onto processes that should be thrown in the garbage. I got into why on a recent episode of Content Amplified, and I wanted to pull the key ideas together here for anyone who missed it.
Working with more than 1,000 companies through SmartBug, I've started to see AI adoption as a bell curve. A small group of early adopters figured out how to run entire marketing campaigns with AI on their own. A chunky middle relies on embedded AI, the kind that shows up inside tools they already own, like a feature that auto-writes an email subject line. And laggards, concentrated in litigious industries like finance and healthcare, are still in the education phase. They know what ChatGPT is. They use it to write an email. That's about it.
Almost none of our work is about jumping a company from zero to a hundred. It's about moving them one notch: a one to a two, a two to a three.
I recommend a lightweight audit before anything else. Interview department leaders, or whoever's most plugged in on each team, for about 30 minutes. Ask where the bottlenecks are. Ask what's the most mundane process in the org, the thing that if you waved a magic wand, would run twice as fast or handle twice the volume. That conversation surfaces exactly where AI can help, and it kicks off a habit you'll need going forward: process mapping. You're going to need a process for mapping processes, because that's the only way you keep finding new places to apply AI.
There's a real difference between adding AI to a process and rebuilding the process because of AI. Sprinkling takes out a step or two. Reinvention means a 10 step process that used to take seven people now takes three, or one. The hard part is that people inside a legacy process carry process debt, they can't always see past how it's always been done. That's why bringing in an outside perspective, someone without that debt baked in, tends to unlock the bigger redesign. Sometimes companies get pushed into this anyway, when a headcount shift means one person now owns three steps of a workflow with an agent doing the rest.
It's never too early to build an agent, a custom GPT, or a Claude Skill, even for something small. Some of what you build will work well. Some will hallucinate constantly and get scrapped. Both outcomes teach you where the models actually stand today.
The biggest unlock at SmartBug wasn't a tool, it was culture. People were afraid to share how they'd tried using AI to change a legacy process, worried they'd get in trouble for touching something that wasn't broken. Once we fixed that, our roughly 200-person team went from silence to about five use cases shared every day. Find the channel where that sharing happens in your org. If it doesn't exist, start it.
Full Episode here:
https://www.getmasset.com/resources/podcast/479-paul-schmidt