I run a landscaping company. Four people, one truck, a seasonal crew that doubles in May. We started using AI in February. Ninety days later it has saved us roughly 38 hours a month and changed how we quote.
This is the actual rollout — not a case study with the rough edges sanded off.
The starting state
Our office manager spent two days a week on the phone, writing emails, drafting estimates, and chasing invoices. Quotes went out three to five days after a site visit. We were losing roughly one in five jobs to a competitor who quoted faster.
The thesis: if we could get a draft quote in front of a customer the same day, we'd close more of them.
Week 1 — One prompt, one task
I set up Claude on the office manager's laptop. One prompt: rewrite an angry customer email in a firm but polite tone.
That's it. Two emails the first week. She rewrote the AI's output most of the time. But she stopped dreading those emails.
Lesson: don't introduce three workflows at once. Find the most painful 15-minute task and automate that. Adoption isn't a software problem — it's a small-wins problem.
Week 3 — Voice memo to estimate draft
We added a second prompt:
Here's a transcript of a site visit voice memo. Draft an estimate for a 220 sqft Holland-paver patio at $22/sqft, plus 6/sqft for cuts, plus 3/sqft for border. Include 4" base prep, polymeric joint sand, and 3-year warranty language.
She'd record the site visit on her phone, paste the transcript, and Claude drafted the estimate. She corrected pricing where Claude got the regional cost wrong, fixed the warranty wording, sent it the same day.
Quotes went out same-day for jobs under $5K starting week three. Close rate on those bumped from roughly 40% to 58% by week six.
Week 6 — Bigger jobs
She started using the same prompt for $10K–$30K jobs. We tightened the prompt with more pricing inputs, added our boilerplate, told Claude to never invent line items. By week eight every estimate that left our office had been drafted by AI and reviewed by her.
The big shift: she went from writing estimates to reviewing them. Different skill, less draining, way faster.
Week 9 — Photo to invoice
Last move: crew takes photos of completed work, she pastes them and a one-line description, AI generates the invoice narrative for the customer receipt.
Today's photos: 4 photos of a 280sqft patio, 1 photo of the polymeric sand application, 1 photo of the as-built drainage trench. Write a 3-sentence customer-facing summary explaining what was completed and what the customer is responsible for in the next 30 days (curing, watering joints, etc).
Customers started replying to invoices with thank-you notes. Hadn't happened before. The narrative was warmer than what we used to send.
What we measured
| Metric | Before | After 90 days |
|---|---|---|
| Hours/week on admin | ~16 | ~6.5 |
| Estimate turnaround | 3–5 days | Same day |
| Close rate (jobs under $10K) | ~40% | ~58% |
| Customer response to invoices | Rare | Common |
That's roughly $3,200/month in retained labor cost plus a 45% lift in close rate on small jobs. The whole stack costs us $20/month in API.
What didn't work
- First prompt I wrote was too generic. "Write a landscaping estimate" produced beige nonsense. The estimate prompt only got useful when I shoved in specific pricing logic.
- I tried to introduce four workflows in one week. She quit using AI for two days. I rolled back to one workflow at a time after that.
- Voice memos with crew chatter in the background confused the transcript. We started recording in the truck cab, doors closed.
The framework
If you're a contractor wondering where to start, this is the order I'd run again:
- One painful 15-min task. Email rewriting, follow-up generation, anything you dread.
- Same task for 2 weeks. Build the muscle of "ask AI first."
- Add one workflow. Probably estimates if you do project work.
- Tighten that prompt for 3 weeks. Don't add more until the first one is boringly reliable.
- Add the next one. Photo-to-invoice, follow-up sequences, quote-to-job conversion.
This is what we run other contractors through when they hire us at the Tool Shop. The order matters. Don't skip the boring 2-week single-task phase.
What's next
We're testing a dispatch automation that books the next-day site visit when a customer says yes. If it works, that closes the loop: lead in, qualified, quoted, scheduled, completed, invoiced — every step gets at least one AI-assisted shortcut.
I'll write that one up around day 180.