Stop Buying AI Tools.
Start Building AI Workflows.

Ask any operations manager what AI tools their company uses. You'll get a list. ChatGPT, Notion AI, a Zapier account someone set up last year, maybe a $300/month tool that three people tried once at an offsite.

Then ask them what those tools have automated. Silence.

That's the problem. Companies are buying AI tools the way they used to buy enterprise software — on the assumption that having access to the technology is the same as using it. It's not.

Tools are passive. Workflows are active.

A tool sits there waiting for someone to open it. A workflow runs when the trigger fires — whether anyone is paying attention or not.

The difference sounds obvious until you look at your own stack. Most AI subscriptions are tools. They require a human to initiate, prompt, and act on the output. That's not automation — that's assisted manual labor. Still valuable, but not what the ROI case is built on.

A workflow is different. An email arrives → AI reads it, categorizes it, logs the key data to your CRM, and routes it to the right person. Nobody opened a tab. Nobody typed a prompt. The work just happened.

That's what “AI in operations” actually means when it's working.

Where to find your first workflow

Don't start with strategy. Start with a person.

Pick someone on your team who does the same task every day or every week. Not a complex judgment call — a repeatable thing. Processing invoices. Summarizing call notes. Routing support tickets. Moving data from one system to another. Filling out a report from numbers that already exist somewhere.

Shadow them for one hour. Write down every step. Pay close attention to the steps that are just moving information — reading something, then copying the relevant part somewhere else. That's your workflow candidate.

The standard we use: if a reasonably smart person could do it in their first week on the job using only what's on their screen, AI can probably do it automatically.

The math is simpler than you think

Take that repeatable task. Estimate the honest hours per week it costs. Multiply by 52. Multiply by the fully-loaded cost per hour for the person doing it (salary ÷ 2,000 is a decent ballpark).

Three hours a week for a $60K employee: that's about $4,500 a year in labor cost on a single task. One workflow, built once, eliminates it. Build five workflows and you're looking at $20–25K in annual labor savings — before you count error reduction, faster turnaround, or the fact that the person now has time for work that actually requires their brain.

Most of the workflows we build cost between $5K and $25K to implement. The payback period on the labor math alone is typically under 12 months. That's before you factor in the compounding — once a workflow exists, it scales without adding headcount.

Why the tool-first approach fails

When companies buy tools first, they're betting that their employees will figure out how to apply them. Sometimes they do. More often, the tool gets used for a few weeks, the novelty wears off, and it goes dormant. The subscription keeps renewing.

The workflow-first approach flips this. You identify the specific process, design the automation around it, and the system runs whether anyone thinks about it or not. Adoption isn't a change management problem — the workflow doesn't care if people remember to use it.

This is why the companies actually getting ROI from AI right now aren't the ones with the most tools. They're the ones who picked three or four specific processes and built reliable automation around each one.

What to do this week

Pick one person. One task. Write down every step. Ask: is each step moving information, or is it actual human judgment?

The information-moving steps are your automation target. The judgment steps stay human — for now.

Build the workflow. Run it for 30 days. Measure the hours saved. Then pick the next one.

That's the whole strategy. It's not glamorous, but it compounds fast.