Product · AI Build
AI-Built Business Systems
Designed and shipped a suite of internal systems for a US home-services group by directing AI: a CRM / estimating / production platform replacing ~$2,000 a month of SaaS, a field estimator ported from Excel and verified to the cent, and a task manager with marketing analytics across 28 months × 16 channels.
- Role
- Product Design · AI-assisted Build
- Client
- Home-services group (US · remote)
- Platform
- Next.js 16 · Prisma · Supabase · Netlify
- Year
- 2026
- Directed with
- Claude · Next.js · Prisma · Supabase
- ~$24k/yr
- SaaS cost replaced (~$2,000 a month before)
- To the cent
- Estimator parity vs. the Excel original, checked by a verification suite
- 28 × 16
- Months × channels of marketing analytics
- 0
- Scheduled-job failures (30-minute analytics sync)
- 28
- Standard reports
- 26
- Row-level security policies
Overview
A growing home-services group ran on a patchwork: about $2,000 a month of SaaS that didn't fit the workflow, a fragile Excel cost sheet that reps couldn't safely show a customer, and marketing data scattered across spreadsheets. I wrote the product requirements and architecture, then directed AI to design and build a set of owned systems that replaced all of it.
The challenge
Each tool was a silo, and none matched how the business actually sold and produced work. The estimator was the sharpest problem: a rep needs to quote in the field, on a phone, sometimes offline — and hand that phone to a customer without exposing margin. Anything that replaced the spreadsheet had to be provably identical to it, or nobody would trust the number.
Before / after
Before ~$2,000 a month of SaaS that didn't fit the workflow
After One owned platform, ~$25–100 a month of hosting
Before An Excel cost sheet reps couldn't show a customer
After A field estimator verified to the cent, safe to hand over
Before Marketing data scattered across spreadsheets
After 28 months × 16 channels, synced every 30 minutes
Approach
- Wrote the PRD and architecture, then directed AI to build a multi-company CRM with a 14-status lead pipeline, 24-status project lifecycle, price book, contracts, invoicing, 28 standard reports and an automations engine.
- Ported the Excel cost sheet into a multi-brand web estimator that generates proposals and contracts — with a verification suite that asserts parity against the workbook to the cent, and an offline-first, customer-safe field UI.
- Built a task manager with an escalation engine, proof-of-work gating and calendar sync, plus an analytics module that syncs three workbooks every 30 minutes into 28 months × 16 channels of spend, bookings and cost-per-booking.
- Hardened it: row-level security on every table (26 policies), and a privilege-escalation defect found and closed with a database trigger.
The AI loop
- Spec
- I wrote the product requirements and architecture first — a 14-status lead pipeline, a 24-status project lifecycle, price book, contracts, invoicing, 28 standard reports and an automations engine — before any code existed.
- Directed
- Then I directed AI to build against that spec: the multi-company CRM, the field estimator ported from the Excel cost sheet, and the task manager with its analytics module.
- Verified
- A verification suite asserts the estimator's proposal and contract totals match the original workbook to the cent; the analytics sync pulls three workbooks every 30 minutes with zero scheduled-job failures; every table carries row-level security (26 policies).
Outcome
One owned platform replacing ~$24k a year of SaaS with ~$25–100 a month of hosting; an estimator the team trusts because it is provably identical to the sheet it replaced; and marketing analytics that stay current without anyone touching a spreadsheet — designed and shipped by one person directing AI.