Adi Gaber · Chicagoland

Too big to run on memory. Too early to hire a department.So I build the systems in between.

I own a garage-door company in Chicagoland — trucks, technicians, payroll, hiring, training, dispatch, marketing, and everything in between. The last couple of years I've been building the pieces we couldn't buy off the shelf: the analytics, the technician scorecards, the ads attribution, the content system. Working through ChatGPT, Make, Gemini, Claude, Grok, Replit and a few others, at night, with no engineer on the payroll.

I write each one up in full — the problem in the business, what I built, what it changed, and the part I got wrong. If you're running something around this size, tell me what's working for you. I'm still figuring most of it out, and sharing it as I go.

Adi Gaber working through a problem at the whiteboard
The short version

Trucks, technicians, and a payroll to make.

Two years ago all of it came through me — the phone, the dispatching, the ordering, the inventory, payroll, training the CSRs, booking the trucks in for service, running the office meetings. Most of that doesn't any more.

I own and run a garage-door company in Chicagoland. That's the day job, and everything on this site got built around it — at night, on weekends, while the company was growing and needed the attention.

I did years of analytics and product work before this, for one of the biggest retailers in the country, so I'm not coming at it from zero. Every one of these I built myself in the evenings, working through whatever tool fit — ChatGPT, Claude, Gemini, Make, Replit.

We're not where we want to be on any of it. We're just further along on the data than we used to be, and some of what we built we had to tear up and do again. More about me.

01

The numbers move while the job is still open

A lead or a job changes in the CRM and it's in the warehouse about two minutes later, with a full rebuild overnight as the backstop. Nobody runs any of it. It still can't tell me when it hasn't run, though, and that part isn't finished.

02

Coaching stopped being a whole day

Every technician gets one specific thing to work on, picked by an engine off the scorecard rather than by whoever I happened to talk to last. I still have the conversation. I stopped doing the analysis for it.

03

We stopped paying for the SEO platform

About $500 a month once you count the consulting on top of the licence. We rebuilt the parts we actually opened — local rank, site health, position tracking, internal linking — on the data APIs underneath, for a few dollars a month. The city-page content we now write ourselves. Still building.

04

About half of it I built twice

The first scorecard judged people on numbers with somebody else's assumptions baked in. The backup I'd have reached for at two in the morning turned out to be 25 days stale. Both of those are written up.

One number at a time

The number we changed, and what it showed.

About once a week: new videos, new use cases, and what we changed inside the company. Not a newsletter about AI news.

About once a week. Unsubscribe anytime. Privacy

Or follow here