1 The product in one line
A single piece of software that owns your stock, and pushes it out to — and pulls sales back from — Trade Me and Shopify automatically. One unit of stock, listed everywhere, sells once, disappears everywhere.
It's the thing eBay sellers buy off the shelf (Linnworks, 3Dsellers, Finale) — except no one sells a good version for Trade Me + Shopify in New Zealand. So we build it. For TYDI it's the operating system. Beyond TYDI, it might be a product (§9).
2 What we're modelling it on — and why we build, not buy
| Reference tool | Best at | What we borrow |
|---|
| Linnworks (inventory-hub gold standard) | Central catalogue → many channels, real-time sync, unified orders | The engine architecture — hub-and-spoke, single source of truth |
| Finale Inventory (refurb specialist) | Per-unit IMEI/serial tracking, condition grading, one-of-a-kind stock | The data model — every physical unit is its own record |
| 3Dsellers (top eBay all-in-one, ~200k sellers) | AI listing creation, templates, bulk editor, repricer, feedback | The listing intelligence — templated, AI-drafted, bulk-managed |
| Sellbrite (single-operator simplicity) | The 80/20 subset for a solo seller, not a 100-channel enterprise | The scope discipline — keep it lean |
Why build instead of buy
- None of the eBay tools touch Trade Me. 3Dsellers, Linnworks, Sellbrite, ChannelAdvisor — zero Trade Me support.
- Tradevine stopped supporting new Shopify connections in 2023 — a live, unresolved forum thread of stranded sellers. That dead bridge is the whole gap.
- Shopify's own Codisto / Marketplace Connect pointedly omits Trade Me — so the obvious default never arrives for NZ.
- The one healthy incumbent (Omnivore/TradeRunner, AU-built) is a commodity lister — not refurb-specialised, weak on unique qty-1 stock, priced at 0.8% of sales.
The gap that matters most: all the cheap tools do quantity sync — they don't do unique-unit sync. That distinction is the heart of the product (§4).
3 The core concept — the Hub as single source of truth
The Stock Hub · master catalogue (Postgres)
1 physical unit = 1 record
qty 1 · grade · serial/IMEI · cost · status · photos
Trade Me
- Selling API
- poll sales every 60–120s
Shopify
- Admin API
- webhook on sale (instant)
↓ SOLD ON ONE → WITHDRAWN ON THE OTHER ↓
The rule that makes it work: the channels never talk to each other — they only talk to the Hub. A sale anywhere updates the Hub, and the Hub updates everywhere else. Add eBay AU or Reebelo later and it's just another spoke — the Hub doesn't change.
4 The one thing off-the-shelf tools get wrong (our make-or-break)
Generic tools assume…
"5 identical widgets." When one sells, they decrement 5 → 4. Fine for commodity stock.
Refurb electronics are…
"5 different physical phones, each unique, each qty 1." When phone #X sells, there's no "4 left" — there's a different phone. You must withdraw that exact listing entirely.
This is listing-withdrawal sync, not quantity sync — precisely where Sellbrite/Codisto-class tools break and oversell. Getting it right is the core of the build. Every unique unit = one Hub record (qty 1) → one Trade Me listing + one Shopify listing, all linked, all withdrawn together the instant it sells anywhere.
(The Hub also handles normal multi-quantity stock — accessories, parts, new-old-stock — as ordinary qty pools. But the unique-unit case is the differentiator.)
5 What it does — the modules
Ordered as a product, front-door to back-office.
Module 1 — the front door
Intake & GradingScan the device → capture IMEI/serial, grade (A/B/C), battery health, cosmetic notes, photos, cost, supplier. Hub auto-mints a SKU (IPH13-128-BLK-A-{serial}) + master record. Where refurb businesses live, and where generic tools have nothing.
Module 2 — single source of truth
Master CatalogueOne record per physical unit, qty 1, with a parent "model" grouping. Status flows: in-stock → listed → sold → dispatched → (returned → re-graded → re-listed). Everything reads from here.
Module 3 — the photo→listing pipeline, slotted in
AI Listing PublisherClean photos (PhotoRoom), vision-LLM drafts the title (leads "Refurbished"), fills attributes, writes the description + subtitle, suggests a price. One template per product line. Batch-approve → publish to Trade Me and Shopify in one action.
Module 4 — the link, the heart
Multichannel Sync EngineShopify → Hub: orders/paid webhook (seconds) → withdraw on Trade Me. Trade Me → Hub: poll sold-items every 60–120s → set Shopify to 0 / unpublish. Oversell safety: atomic claim, compare-and-set, reconciliation poll (§7).
Module 5 — back office
Unified Orders & DispatchEvery TM + Shopify order in one queue, resolved to the exact serial. Scan-to-confirm on pick (scanned IMEI must match). NZ carrier labels (Starshipit / GoSweetSpot / NZ Post). Tracking pushed back; listing withdrawn on the other channel.
Module 6 — the lifecycle tools fail at
Returns / Re-grade / Re-list LoopReturn received → inspect → re-grade (may downgrade) → re-price → re-list as a fresh qty-1 record. An explicit state machine so units never get stuck as ghost stock.
Module 7 — the decision layer
Reporting & Per-Unit MarginBecause every unit is serialised with its cost, true per-unit profit (sale − cost − fees − shipping) is trivial — plus days-to-sell by model/grade, ageing-stock alerts, sell-through by channel.
Phase-2 modules (defer): Q&A auto-responder, feedback automation, age-based markdown repricing, Sponsored Listings management. All supported by the Trade Me API; none are day-one critical.
6 How it all comes together — one phone's journey
A refurbished iPhone 13 arrivesOperator scans it: IMEI captured, graded B, battery 89%, small scuff, 4 photos, cost $380. → Hub creates IPH13-128-BLK-B-{serial}, qty 1, status in-stock.
AI Publisher drafts the listingTitle "Refurbished Apple iPhone 13 128GB Black — Grade B, 89% Battery, 12-Month Warranty", attributes filled, price $620. Operator approves. → Live on Trade Me and Shopify, linked.
It sells on Trade Me at 2:14pmNext poll (≤2 min) the Hub sees the sale → marks sold → sets the Shopify listing to 0 / unpublishes it. The website can no longer sell the same phone.
The order drops into the unified queueOperator picks it, scans the IMEI — matches ✓ — prints a CourierPost label, packs. → Hub marks dispatched, pushes tracking to Trade Me, marks shipped via the API.
Per-unit margin booked automaticallySold $620 − cost $380 − TM fee − shipping = real profit on that exact device, in the dashboard.
(If returned)Inspect → maybe re-grade to C → re-price → the Hub spins up a new qty-1 listing. Loop closed.
Steps 2–4 are ~90% software. The human scans the box at intake and tapes it at dispatch. That's the whole "AI-first" promise made concrete.
7 The Trade Me ↔ Shopify link — how the sync really behaves
The honest engineering picture, because this is the part that's genuinely hard.
| Direction | Mechanism | Speed | Reliability move |
|---|
| Shopify → Trade Me | orders/paid webhook → ack <1s, queue the job, withdraw on TM | Seconds | Verify HMAC; process async; reconciliation poll as backstop |
| Trade Me → Shopify | Poll SoldItems (LastHour) every 60–120s | Up to ~2 min | inventorySetQuantities with compare-and-set so the Hub never overwrites a unit Shopify just sold |
The unavoidable truth: Trade Me has no webhooks, so TM sales are only knowable by polling. That creates a small oversell window for a unique unit listed on both channels — a website buyer could grab the same phone in the ~2-min gap. Low probability per item, but high cost per incident (refund + reputation hit on the 99.16%-feedback account that is the asset).
Mitigation, in order of strength
- Route unique high-value units to ONE primary channel. Premium/rare devices go TM-only or Shopify-only; only commodity stock dual-lists. Eliminates the race where it hurts — the default policy.
- Poll Trade Me every 60–120s — cheap, small payloads, well within the rate limit.
- Compare-and-set on every Shopify write — turns a silent double-sale into a flagged conflict.
- Reconciliation job that alerts on any unit found live on two channels.
It can't be closed to zero — but for a refurb business it can be made rare and cheap, and policy (1) makes it a non-issue for the stock that matters.
Acknowledge the trade-off, though. Policy (1) works by not dual-listing the unique high-value units — which means it reduces the multi-channel listing this Hub exists to enable. So the Hub's clearest, unambiguous value is the AI listing pipeline plus commodity / multi-quantity stock (accessories, parts, new-old-stock as ordinary qty pools), where dual-listing carries low oversell risk. Reserve dual-listing for that low-risk stock; route premium qty-1 units to a single channel. The Hub still earns its keep — the listing engine and the commodity multichannel reach are the bulk of the value — but the "list once → sell on every channel" line is, by design, narrowed for the unique units where the oversell race actually bites.
8 Build vs buy — the call
Omnivore/TradeRunner exists and does Shopify↔Trade Me today. So why build?
- Refurb-specialised where Omnivore isn't. Per-unit IMEI/grade/battery-health, unique-unit withdrawal sync, scan-to-confirm dispatch, single-channel routing — things a commodity lister doesn't do.
- It owns the AI listing pipeline. Omnivore syncs stock; it doesn't draft titles, fill attributes, or fix the CRO gaps. The Hub does — that's where the margin is.
- Flat cost vs 0.8% of sales. On a $794k → $1.85M store, 0.8% is $6k → $15k/yr to a third party, forever. Building it once removes that tax.
- It's the operating layer for the whole thesis. Every other AI-first lever plugs into the Hub. You're building the spine of the business, not buying a sync tool.
Recommendation: build it for TYDI. Use Omnivore only as a stop-gap if you want stock synced before the Hub is ready.
9 The bigger idea — could this be a product?
The instinct is right but needs calibrating. The honest read: the white space is real but thin — a niche, not a land-grab.
~25k
Active Shopify stores in NZ (~3,100 in Canterbury), growing ~9%/yr
low-000s
Realistic overlap — sellers running both Shopify + a serious Trade Me store
$49–99
/mo — a small, defensible SaaS; low-hundreds-of-$k ARR ceiling, not a venture outcome
- Documented, unresolved pain: the "Tradevine Shopify is gone" thread is real sellers stranded with no clean answer. Omnivore's %-of-sales and TradeRunner's weekly sync are weaknesses to wedge against.
The smart sequencing (the actual "something cool"): build it for TYDI first, where it pays for itself as the operating system on a real high-volume store. TYDI becomes the proof-of-concept and first case study. Then the productisation decision is de-risked — you've already built and battle-tested it. "Build for one, sell to many," but only commit to the "many" after the "one" proves it. Exactly your SettledLoop pattern. Treat productising as optional upside on a build justified for TYDI alone.
10 Phased build (mirrors the data dependency)
| Phase | Ships | Why this order |
|---|
| 1 — Catalogue + Intake | Master record model + the scan/grade front door | Get the data model right first; everything hangs off it |
| 2 — AI Listing Publisher | Photo→listing pipeline → publish to TM + Shopify | The 80/20 of value; fixes every CRO gap |
| 3 — Sync Engine | Unique-unit withdrawal sync, both directions + safety net | Protects the listings once live; the make-or-break module |
| 4 — Orders + Dispatch | Unified queue, scan-to-confirm, NZ labels, tracking pushback | Recovers the daily fulfilment grind |
| 5 — Returns loop | Re-grade → re-list state machine | Closes the lifecycle the tools fail at |
| 6 — Reporting | Per-unit margin, ageing, sell-through | The decision layer |
| Phase 2+ | Q&A bot, feedback, repricing, more channels | Leverage once the spine runs |
Built on your stack: Node/Fastify/Postgres/Prisma/Railway, secrets in ~/.brain-secrets/, Shopify custom app (GraphQL Admin API), Trade Me OAuth 1.0a app (register fresh under the new in-trade owner — see the API doc's §2).
11 Decisions for you
1. Scope — full Hub, or just the AI Listing Publisher bolted onto Omnivore for sync, build the sync engine later? Lean: catalogue + publisher first (the value); add the sync engine before you trust it for unique units.
2. Channel-routing policy — dual-list everything, or route unique high-value units to one channel? Lean: route high-value to one channel — kills the oversell risk where it matters.
3. Productisation — TYDI-only for now, or design multi-tenant from day one? Lean: TYDI-first, but keep tenant-isolation in the data model so productising later isn't a rewrite — the SettledLoop lesson.
4. Pre-settlement — want a Phase 1+2 build brief (data model + enrich prompt + approve-screen wireframe) ready to start in the sandbox before settlement?
Grounded in June-2026 product research and the Shopify/Trade Me API docs. Endpoint behaviour, Tradevine's Shopify status, and competitor pricing should be re-verified at build time. Not legal advice — confirm Trade Me API T&Cs (no caching/redistributing their data) for any competitor-facing feature.
→ See the full operating-system vision (Stock Hub)
← Back to the document hub