Business plan v2 — one package, AUD, honest infra cost

Young's three rulings, 2026-10-14: Oracle ARM shape confirmed as ours (4vCPU/24GB/200GB) · POS gaps + EVE beta are build work (Sol specs, Min+Tam Docker design) · prices in AUD.

What changed from v1: prices converted to AUD, and the infra cost model is rebuilt around a finding from this morning's probe — our node EXCEEDS Oracle's Always Free allowance, so v1's "$0 infra" case is retired.

1. The product (unchanged from v1)

Hosted online shop (Medusa v2.21.2, MIT) + in-store POS on client tablets (Agilo fork, MIT) + AI staff with per-tenant memory (EVE, Apache-2.0). Licence profile clean end-to-end — the bundle itself is sellable. Backups, updates, TLS, domain, admin onboarding included; we run all of it.

2. Who buys (unchanged)

Single-location AU retailers on Shopify Basic + POS Pro. The switch pitch: one bill lower than their current two bills, plus the AI receptionist they don't have.

3. Pricing PROPOSED, AUD

Rate used: A$1.55 = US$1 (Oct 2026; re-check at first sale — the number moves).

The anchor still holds in AUD: Shopify AU Basic (~A$60) + POS Pro (~A$137) ≈ A$197/mo before processing — Solo undercuts it by A$78 while adding AI staff. (Shopify AU figures are the standard AU list; confirm at sale — regional pricing moves.)

Pass-through, stated upfront: card processing (Stripe AU ~1.75% + A$0.30 domestic) and POS hardware (tenant buys). No GMV cut — still the selling point vs Saleor Cloud's 0.8%.

AI overage PROPOSED: included conversations, then tenant key or metered overage at cost + 20%, billed in AUD.

4. Unit economics — corrected for the allowance finding

The finding (probed 2026-10-14, IMDS + docs)

Node: VM.Standard.A1.Flex, ap-melbourne-1, 4 OCPUs / 24 GB / 200 GB — confirmed ours. Oracle's Always Free covers 1,500 OCPU-hours + 9,000 GB-hours/month = 2 OCPUs + 12 GB always-on 2. We run double that. The excess (2 OCPU + 12 GB) is billable at PAYG unless trial credits cover it.

v1's "$0 infra" case is retired as the default — it only holds while credits last. The margin survives either way because the product is labour-priced, not infra-priced: onboarding is scripted, support is the cost, and tenant #1 stays cash-positive from month one in both cases.

OPEN: one console check — Young or whoever holds the OCI login: confirm whether the account is on trial credits or PAYG, and the actual monthly burn. That single number collapses this table to one row.

AI variable cost (unchanged)

Single-digit AUD per tenant per month at bundled volumes on free/long-context models — inside the tier price, overage rule as backstop.

5. Competitor bench (USD sourced 2026-10-13; AUD approx at 1.55)

6. Build work this plan depends on (other lanes — tracked, not owned)

7. Risks (repriced where the allowance finding bites)

8. Decisions for Young (v1 §8, updated)

1. Tier prices in AUD (A$119 / A$229 / +A$75) — approve or reset.
2. AI overage model — tenant-key-or-metered, as proposed?
3. OCI console check — who looks, and what is the actual burn?
4. First pilot shop — name it.

Sources

• Allowance math: Oracle docs Always Free Resources (A1 Flex: 1,500 OCPU-h + 9,000 GB-h/mo) 2 + community allocation note 1 + live IMDS shape probe 2026-10-14.
• Competitors: medusajs.com/pricing, docs.medusajs.com/cloud/pricing, shopify.com/pricing (all fetched 2026-10-13).
• Stack/sizing/licences: Tam's three studies (see v1 §Sources — unchanged).