WristBook — Case Study

Overview

The whole operational backbone, in software.

WristBook is a CRM-and-accounting platform for independent luxury watch dealers — the system of record for inventory, buyers, deals, and books. I built the operational backbone end to end: the AI that does the data entry, the automations that run the lifecycle, and the internal tooling that lets the founder operate the whole business from one cockpit.

The throughline was simple. Take the work a dealer used to do across a notebook and eleven browser tabs, and have software do it — intelligently, automatically, and reliably.

What I built

Three systems, one platform.

01

AI features

Claude-powered · generative where judgment is needed

Grounded watch identification. Snap a photo or type a reference and the spec fills itself in — through a retrieval-augmented pipeline (fingerprint → retrieve real catalog candidates → grounded model pick), not freeform guessing. Every call is cost-tracked.
Invoice & batch extraction. Drop in a PDF invoice and the AI reads every line item, allocates shipping per piece, and pre-fills an entire batch — eight watches in under a minute.
Messy-data import. Spreadsheets, scanned ledgers, bank-statement PDFs, even a photo of handwritten notes become clean, structured records the operator reviews before commit. AI proposes, human approves.
Changelog drafting. Raw commit notes turn into polished, customer-facing release notes on demand.

02

Automation infrastructure

Deterministic & auditable · rule-based where predictability matters

A full lifecycle email engine. Six behavior-triggered sequences — onboarding activation, inactivity re-engagement, plan-limit upgrade nudges, migration upsell, milestone review/referral asks, and a multi-touch churn win-back. Built on a single ledger that doubles as an idempotency guard and scheduler, so an email fires exactly once, cancels itself the moment the customer takes the target action, honors unsubscribe, and ships behind a master safety flag.
An automatic buyer matcher. Every new piece is instantly scored against every client wishlist, so the right buyer surfaces itself. Deterministic and explainable by design.
Billing & delivery automation. Payment lifecycle, plan state, and email deliverability all driven by event webhooks.

03

Operator command center

One founder, the whole business

A founder-facing admin that runs the company. Relationship CRM, a free-access invite system, revenue and funnel analytics, completion meters, and one-click customer communications — the whole multi-tenant business operated from a single dashboard.

How it's built

Real tools, drawn on purpose.

Next.js (App Router) Supabase — Postgres · Auth · RLS Anthropic Claude Stripe Resend Vercel — serverless + cron

Patterns first: every new system reuses the platform's existing conventions instead of bolting on a new layer. And the line between AI and deterministic automation is drawn deliberately — generative where judgment is needed (reading a messy invoice), rule-based where predictability matters (anything that sends money or email).

What it enables

A business that runs itself between logins.

→

Minutes of data entry instead of hours.

The AI does the cataloging — the part that used to eat the day.
→

Onboarding, re-engagement, upgrades and win-backs, all hands-free.

The lifecycle happens without the founder lifting a finger — and never double-fires.
→

One founder operating a multi-tenant SaaS.

The entire platform, run from a single dashboard.