Case Study · AI for Ecommerce

A content system that runs the marketing for a Lake Tahoe tea brand
Tica and Mike run Tahoe Teas as a family. They have a real voice and real products — what they didn't have was the time to re-write the same idea for email, the blog, Instagram, and a text. So we built the software that does it for them, end to end, with Tica approving every word before it ships.
- Client
- Tahoe Teas, Ecommerce
- Role
- Design + AI Engineering
- Engagement
- 2025 to present

Where it started
From one email flow to the whole calendar
The first job was narrow: replace Klaviyo for the welcome flow and the signup popup. We cut that over in May 2026. It worked, and the real problem became obvious — the welcome email was only one of the four places Tahoe Teas needed to show up every week.
In June 2026 we rebranded the app from the “Email System” to the Tahoe Teas Content System and gave it a single job: take one good idea and turn it into a week of on-brand content, without Tica writing any of it from scratch.

The loop
Idea in, published content out, then it learns
The whole system is a closed loop. Each round of publishing feeds the next round of ideas — so it's built to get sharper the more it runs.
01
Ideas, ranked
An AI strategist proposes a ranked queue of what to publish next, grounded in empty weeks on the calendar, the Lake Tahoe season, SEO keywords nobody has covered, recent titles, and what's actually converting.
02
One click drafts every channel
Pick an idea and the Studio writes email, a Shopify blog post, an Instagram caption, and an SMS in parallel. Each is shaped for its own channel rather than copy-pasted.
03
A quality gate before anyone sees it
Each piece is drafted a few ways; a cheaper model judges which is best. A deterministic brand-voice check scores every draft 0 to 100 and strips the giveaway dash before a customer could ever read it.
04
The owner approves from her phone
Tica reviews and signs off on each channel through a passwordless share link. She never touches code, and nothing leaves the building without her yes.
05
Publish, then learn
Approved pieces auto-publish on schedule. Opens, clicks, and Shopify discount-code revenue flow back in to shape the next round of ideas.
One idea → four channels
Welcome flow + broadcasts
SEO articles
Image posts
Short, timely
SMS is built and TCPA-consent-compliant, pending carrier approval (Twilio A2P 10DLC). Instagram posts images; reels with music need a quick manual finish. Facebook cross-posting was deferred on purpose.
Why it can be trusted
Automated, but never on autopilot
Grounded in its own voice
Every draft is checked against the brand's previously approved copy with pgvector, so new writing sounds like Tahoe Teas, not like a model.
Human in the loop, by design
Nothing publishes without an explicit, per-channel yes from the owner. The AI proposes and drafts; a person decides.
Honest about its limits
SMS is built and consent-compliant but waits on carrier approval. Instagram posts images; music needs a manual finish. Facebook was deferred on purpose.
Wired into the store
Subscribers, unique discount codes, and published articles sync with Shopify, and revenue from those codes is attributed back to the content that drove it.

What changed for them
A two-person shop that publishes like a team
Instead of staring at a blank calendar, Tica opens a ranked list of ideas that already know what the store needs. One click drafts the week. She reviews it on her phone, approves what she likes, and the rest publishes on schedule.
The brand voice holds steady across every channel because it's grounded in their own past writing — not a generic template, and not a tone that drifts from tool to tool.
The math
What building it in-house saves
$4,400/yr
Subscriptions replaced
AI writer, social scheduler, blog retainer, design tool
~12 hrs/mo
Owner time reclaimed
Writing & scheduling, valued at ~$40/hr
~$10,100/yr
Estimated total savings
Avoided tools plus reclaimed labor
The honest version: a custom build like this pays for itself by retiring a stack of subscriptions and handing Tica back the better part of a workday each week. We've kept the numbers deliberately small — they count only the tools the system replaces and the hours it gives back, with no revenue claims attached.
AssumptionsReplaced tools: AI writer (~$468/yr), Buffer (~$180/yr), a freelance blog/SEO retainer at 2 posts/mo × $150 (~$3,600/yr), and Canva (~$120/yr). Reclaimed time: ~12 hrs/mo at a conservative ~$40/hr (~$5,760/yr). The email-sending platform itself is counted at $0 — the brand still needs one. Real benefit is likely higher.
Built with
- Next.js 16
- React 19
- Tailwind v4
- Drizzle + Neon (pgvector)
- Resend
- Shopify Admin GraphQL
- Instagram Graph API
- Twilio
- Notion
- Vercel Blob + Cron
- AI SDK v6 + AI Gateway
- Claude Opus 4.6 + Haiku 4.5
Doing the same work by hand?
Tea, software, or anything in between — if you keep re-typing the same idea in four places, there's usually a system worth building. Tell me what's eating your week.