Gerrit Huizinga (@gerrith777) header
Gerrit Huizinga (@gerrith777) profile picture

AI that pays for itself in real small businesses. I run AI inside a home-improvement and outdoor-living company in Indianapolis. Hours, dollars, payback.

Indianapolis, IN
Joined January 2016
Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)Photo by Gerrit Huizinga (@gerrith777)
I run AI inside a home-improvement and outdoor-living company in Indianapolis. The question owners ask me is simple: it saves time, but what's it worth? Each week I'll break down one real system: • how it's built • what it's worth, in hours and dollars • where to start Monday If you run a real business and want AI that pays for itself, follow along. Free 2-minute Automation Score in the reply below.
1
2
42
Three calls in my lead system no AI gets to make: No permission to text: email or a person instead. An opt-out word: everything stops. A 10 PM miss: held until 8 AM. In testing, opt-outs: 5 for 5. Dollars: not measured. Start: ask who in your office tracks text permission.
1
27
The most important button in my missed-call system does nothing by itself. AI writes the callback. Approve just sits there until a person reads it. What's it worth? Dollars not measured yet. A person still has to hit Approve. Start Monday: nothing goes out without that tap.
21
I built a morning Meta ad report that never connects to the live account. The stack: a Meta CSV export, booked appointments from the CRM, about 60 lines of Python, and an approval queue.
2
37
Step 3: Five verdicts: KILL, SCALE, REFRESH, KEEP, and LEARNING (too new to judge). In testing, both ads I built to fail got killed, the tired one flagged, the winner scaled, two new ads left alone. One bad day did not get an ad cut.
1
5
Step 4: Every change sits in an approval queue. Nothing moves until you say yes. Score it by booked appointments per ad dollar, before and after. Monday: export 30 days of ads and booked appointments. Set your target from your own margins.
4
Fast AI on leads is easy. Safe is the work. Four rules: 1. No permission, no text 2. "Stop" ends it 3. No texts 9 PM-8 AM 4. Complaints or doubt go to a person With AI sorting, in testing: 0 dangerous misroutes on 32 new messages. Dollars: not measured. Monday: write them down.
23
Which job would you never let AI do alone? Mine: texting a customer without a permission check, and changing ad spend. Both wait for a person's yes. Dollars: not measured. Start Monday: write down your two, and name the person who says yes to each.
17
Sort what Google or an AI tool could cite about you into four buckets: Own: your website Claim: your Google Business Profile, directories Earn: local news, forum threads Contest: wrong info Dollars: not measured. Start Monday: claim your profile, fix wrong info.
3
2
70
Customers call when a promised update never came. 7 touchpoints: welcome, measure, order + ETA, install confirmed, day-before, demo walkthrough, 7-day check-in. In testing, 40 jobs: 10 overdue. Worth: count "where's my order?" calls. Start Monday: send the order and ETA (#3).
17
Square-foot pricing hides jobs that lose money. In testing, a flat $/sq ft left a 2nd-story awning thin. Start Monday: cost the next $5K+ quote (materials, crew hours, drive, truck), then add a reserve. BLS median, construction laborers: $22.66/hr (May 2025, public estimate).
26
Most "dead" quotes aren't dead. They're unpaid follow-up. In testing, 150 CRM rows, nothing sent:
2
40
Sorting replies by keyword failed on fresh wording (6/15). So for now, any unclear reply pauses the follow-up and goes to a person.
1
7
Discounts come last. Close-out is a real goodbye, then stop. Score it: old quotes that reopen × close rate × gross profit, minus the cost of the hours and the tool. Start with last quarter's open quotes.
1
6
One landing page for every ad is a leak I check on a phone: 1. Headline matches the ad 2. Stars in the first screen 3. Booking up top 4. Promised counts match the screens 5. Tap-to-call 6. One page per service Dollars: not measured. Start with your biggest ad's page.
20