I've written about automation costs for small businesses, and the number that usually stops people isn't the software price.

An office manager (analyst, designer, salesperson - substitute your use case) spending 10 hours a week on data entry, follow-up emails, and Monday morning reports. At $35/hour, that's $18,200 a year. It never shows up on a line item. It hides inside salaries, inside "that's just how we do things," inside the assumption that someone needs to do this by hand.

Maybe you're not ready to hand AI the follow-up emails and that's fine. The data entry and Monday reports are where most people start. And your office manager probably has opinions about which tasks they'd happily never do again.

Why It Matters

While the direct cost is real, the opportunity cost is worse. Every hour a business owner spends on $35/hour tasks is an hour not spent selling, building client relationships, or making decisions that actually move revenue. The stuff that's worth $200/hour. Most small businesses are funding their operations budget with their growth budget and don't realize it.

The Part Most People Miss

The ceiling on what counts as "automatable" moved and most businesses didn't notice. Nat Eliason built an AI agent called Felix on OpenClaw and manages it entirely over Telegram. It handles product builds, Stripe payments, and sales outreach — no human employees in the loop. That's the extreme case. But the direction matters.

Intercom reported that the five heaviest users of their internal Claude Code system aren't engineers — they're design managers, customer support leads, and PMs. The work getting automated first isn't the technical stuff. It's the repetitive coordination work that eats everyone's week. I've always thought the best ROI for automation is the monotonous death-by-a-thousand-cuts tasks — regardless of the math. Tasks that aren't necessarily difficult but keep you from doing other things or require constant context switching. Automation doesn't need to be singularly focused on moonshots.

Most businesses still draw the "automatable" line where it was two years ago. The line moved.

What Automation Actually Costs (And Saves) a Small Business →

Lessons Learned

• "Start with one workflow, not the whole business." Pick the one that causes the most pain, prove the ROI, then do the next one. You risk $50, not $5,000.
• "Automate chaos and you get faster chaos." Can't describe the steps on paper? Can't automate them. Fix the process first, then let a tool run it.
• "The ROI math is simpler than you think." Hours saved per week, times the hourly cost of the person doing it, minus the tool cost. Most automations pay for themselves in 1-3 months. After that, every month is margin.

Read the rest →

Worth Reading

How We Use Claude Code Today at Intercom — Fin Ideas (Brian Scanlan) 90% of code changes authored by Claude Code. 13 plugins, 100+ skills. The stat that matters: the five heaviest users are design managers and PMs, not engineers.

Everything Is Computer — Perplexity AI Hardware-tethered agents with persistent memory that run on your machine. AI stops living in a browser tab and starts living on your desktop.

Codex for Non-Developers — JJ Englert on X Practical guide for using OpenAI Codex without an engineering background. Setup, skills, workspace config, and real use cases across business functions.

Not sure where the hidden costs are in your business? I do free 15-minute automation audits — no pitch, just an honest look at what's costing you time and what's worth automating first. Hit reply.

— Collin