TYDI — AI-First Operating Model

1 The model: AI does the roles, the operator runs the business

The whole point of "AI-first" here is leverage — the work a team would do is run by brain skills and co-work agents, supervised by the operator. AI is the tool; you're the operator. What a traditional refurb retailer staffs as separate roles, TYDI runs as supervised agents:

Cataloguer / copywriter→ co-work + nhb-copy / atlas

Listing / feed admin→ feed pipeline + product-feed skill

Ad manager→ google-ads suite

SEO / content lead→ atlas · seo-sprint · blog-pipeline

Sourcing analyst→ super-scraper + scoring agent

CRM / email marketer→ GHL + segmentation co-work

Customer support→ AI support agent

Analyst / reporting→ briefing + csv-analyzer

~8→2

Team roles, collapsed to two supervised operators + a picker

New tooling to buy — it's already in the brain

Why it matters for the deal: the lean labour line in the FY26 accounts isn't an anomaly to be "fixed" with hires — it's the target operating model. AI leverage is what makes the bootstrap viable and the margins defensible as revenue scales.

Honest note — this is the target, not day-one reality. The "two operators + a picker" chart above is collateralised by an AI listing system that doesn't exist yet — and which the same person running marketing would build. So treat it as a timeboxed Phase 1 build (a few weeks) with a clear fallback: if the build slips, hire a lister to carry the listing workload manually until the pipeline is live. The lean org is what the business converges to once the system is running — not an assumption that it's already running on day one.

2 The operating map — function → AI-first method → brain capability

Every line below uses a skill, MCP or stack component you already own. Nothing here is hypothetical.

Function	AI-first method	Brain capability
Sourcing & deal scoring	Monitor liquidation / returns / clearance sources; score deals by resale demand × margin	`super-scraper` + SerpAPI · `csv-analyzer` · co-work scoring agent
Product cataloguing	Generate SEO titles, descriptions, specs & categorisation from a model number/photo; clean imagery	`write-post`/`nhb-copy` + `atlas-content-generation` · `generate-image`
Multi-channel listing & feeds	Optimised feeds pushed to site + Trade Me + eBay + Mighty Ape + Google Shopping	`sellable-product-feed-optimization` · feed pipeline (Railway/n8n)
AI-shopping readiness	Catalog structured to be discovered & bought by AI agents (ChatGPT / Perplexity / Gemini / Google AI)	`sellable-ai-ready-catalog-audit` + `sellable-agentic-storefront-optimization` · `atlas-schema-markup`
Pricing	Competitor price monitoring + repricing rules	`super-scraper` + SerpAPI · `csv-analyzer` · pricing agent
SEO (brand sites)	Topical content, programmatic pages, technical recovery	`atlas` · `seo-sprint` · `blog-pipeline` · `seo-research` · Ahrefs MCP
Paid ads	Build & run Shopping/Search/PMax; competitor ad intel; search-term hygiene	`google-ads` (+ creation/audit/pmax/cannibalization) · `adlib-analysis`
Landing pages & funnels	Brand-specialist LPs, quiz funnels, CRO audits	`frontend-design` · `sellable-landing-page-generator` · `settledloop` · `quiz-funnel`
Content & video	Per-platform posts, product video, brand content	`write-post` · `video-topics` · `sellable-video-storyboard` · `remotion` · Canva MCP
CRM reactivation	Email/SMS lifecycle, parts cross-sell, behavioural segmentation	GHL · `gmail` · co-work segmentation
Voice of customer	Mine reviews & returns for product + copy insight	`sellable-voice-of-customer`
Customer support	AI support agent for order/product queries	Call-agent pattern (capex-sdlt / gerrards fork) · GHL
Dead-stock liquidation	Batch $1-auction listings to Trade Me / eBay (the consignment clear-out)	Co-work listing agent + feed pipeline
Reporting & ops	Daily performance briefing, inventory reconciliation, dashboards	`briefing-generator` (gerrards-daily pattern) · `sheets`
Tracking	Server-side conversion tracking across every channel	Stape sGTM (MCP) · the Stape/GTM tenant recipe

3 The real edge: built to be sold by AI shopping agents

Commerce is going agentic. Shoppers increasingly ask ChatGPT, Perplexity and Google's AI for "the best refurbished Dyson V8 in NZ" — and the AI picks from catalogs it can read and trust: structured data, clear condition/warranty, clean specs, strong entity signals. Almost no NZ refurb seller is optimised for this. TYDI's catalog — audited and structured with sellable-ai-ready-catalog-audit + sellable-agentic-storefront-optimization + atlas-schema-markup — becomes a first-mover in agentic commerce, not just another human-SEO store.

This is the difference between "an e-commerce business that uses some AI" and an AI-first one: AI runs the back office and the catalog is engineered for the channel everyone else hasn't noticed yet.

4 The AI photo & video studio — differentiation + CRO

Existing brain workflows that turn one plain product shot into a full campaign — at near-zero marginal cost per SKU.

From one product shot →

Lifestyle imagery — drop the refurbished Dyson or laptop into real-room scenes (your product→lifestyle workflow)
UGC video — AI model + user "review" / demo clips generated from the same shot (your UGC workflow)
A full AI photoshoot per SKU — hero, lifestyle, detail & video, batched with no studio and no model fees

Skills: generate-image · remotion · sellable-video-storyboard · video-topics · Canva MCP

One shoot, three jobs

Product pages→ premium look → higher CR

Social→ always-on organic content

Paid ads→ fresh creative on tap

The differentiation: every Trade Me / eBay listing looks the same — one flat photo. TYDI's brand-specialist pages look like a premium retailer → trust → conversion, and the same assets feed social + ads. A marketplace seller structurally can't match this.

5 Email & lifecycle — the owned channel marketplaces can't touch

The acquired customer database + email list is the asset a Trade Me / eBay seller never gets — and where the slow/dead stock turns into profit.

List size isn't disclosed — the IM only says "active email subscriber list", with no number. DD item: get the size, growth rate, engagement (open/click), platform, and — critically — the consent basis. NZ's Unsolicited Electronic Messages Act 2007 requires opt-in + unsubscribe + sender ID; you can't legally blast an old or unconsented list, so how it was collected determines its real value.

Segment by

Brand bought — Dyson owners → parts/upgrades; Apple buyers → accessories (feeds the brand-specialist sites)
RFM — recency / frequency / value → VIPs, lapsing, one-timers
Category & browse interest
Price sensitivity — deal-seekers → the clearance offers below

Automated flows

Cart & browse abandonment — recover near-misses (the fastest win)
Related products / cross-sell — parts & accessory attach (high margin)
Post-purchase — review (feeds VoC), warranty, replenishment (Dyson filters/batteries on a cycle)
Win-back & welcome — lapsed and new subscribers

The dead-stock money-maker (your idea): the slow/low/dead stock you'd otherwise $1-auction is monetised far better through owned email — offered cheap as a free-shipping-threshold add-on ("add this $29 item to hit free shipping"). It recovers real margin on inventory the books treat as dead, and lifts average order value — on a channel you alone control. You can't do this on Trade Me or eBay — it's the clearest reason owning the customer relationship beats being a marketplace seller. Driven by a live inventory feed + co-work that auto-builds the offers, delivered through GHL. Full email deep-dive, segmentation & the quantified potential →

6 How it's built — co-work + your stack

The thinking layer — Claude co-work

Recurring cognitive work run as brain skills / agents / workflows, on demand or scheduled: cataloguing & copy, deal scoring, listing prep, SEO & content, VoC, reporting. Supervised by the operator — review, approve, ship.

The plumbing layer — your build stack

Railway · Fastify · Postgres · Prisma — the feed / listing / pricing engine
n8n (automation.bridgemedia.nz) — glue & scheduling
GHL — CRM, email/SMS, social planner
Stape sGTM — server-side tracking · super-scraper + SerpAPI — sourcing & pricing data · Astro/Keystatic — the brand-specialist sites

7 Rollout sequence (what to automate first)

Catalogue + list

AI copy + feed pipeline + the $1-auction dead-stock lister. Kills the biggest labour drain on day one.

Feed + SEO

Google Shopping feed + SEO recovery — capture the demand already pointed at TYDI.

Ads + tracking

Shopping/Search/PMax under Stape conversion tracking.

CRM + VoC

Reactivate the database; mine reviews for product/copy.

Agentic-ready

Catalog schema + AI-shopping optimisation — the first-mover edge.

Support agent

AI handles order/product queries; operator handles exceptions.

The through-line: AI is the leverage that lets two operators run a business that produces a team's worth of output — and being AI-shopping-ready first is the edge as commerce goes agentic. That's what "AI-first" buys you: lower cost to run, faster to scale, and a moat the incumbents aren't building.

Every skill, MCP and stack component named here is already present in the brain (verified) or in Greg's build stack. This is an operating-model blueprint, not a build spec — each line becomes its own short project.

← Back to the document hub