AI Commerce OS — Project Planning

Small businesses that take a high volume of phone orders lose revenue every day to missed and unanswered calls. AI Commerce OS is a multi-tenant SaaS platform that answers inbound calls with an AI voice agent, takes and confirms orders over a conversational WhatsApp agent, automatically recovers missed calls into WhatsApp conversations, and routes every order into a single management dashboard.

The platform is delivered in clearly defined phases. Phase 1 is a focused, fixed-price MVP / pilot that proves the core value — answer calls, recover missed orders, manage them in one place. Phases 2 and 3 extend the MVP into a full commercial SaaS with loyalty, marketing automation, deeper AI memory, and enterprise capabilities, commissioned separately once the pilot succeeds.

No voice model is built from scratch. The system integrates proven voice AI platforms (Vapi / Retell) with the WhatsApp Business API and a custom backend and dashboard. The client owns all source code and all platform accounts.

During peak hours, staff cannot answer every call. Each missed call is a potential lost order and a customer who may not call back. Static auto-replies do not solve this — a customer who reaches voicemail or an unanswered line rarely completes the order on their own.

The goal is therefore not only to answer more calls, but to actively recover the ones that are missed and to make ordering effortless across both voice and chat — while giving the business clear visibility and control over every order in one place.

AI Commerce OS is a multi-tenant platform. A single deployment serves many businesses (tenants), each fully isolated. The platform is sold as a subscription, with three operating roles:

Tenant isolation: each shop's data — orders, customers, and AI memory — is strictly separated. One shop can never access another shop's data.

Core capabilities

The architecture favours a single shared platform with strict logical isolation per tenant. This keeps operating costs and complexity low at MVP stage while remaining a true multi-tenant SaaS that can scale.

Multi-tenancy approach (MVP): a shared database and schema with a tenant_id column on every tenant-scoped table. Every backend query is automatically scoped to the authenticated user's shop and branch. At scale, the model can evolve toward separate schemas or databases per tenant — deliberately deferred to avoid operational overhead at MVP scale.

Every choice is selected for reliability, maintainability, and full client ownership — no proprietary lock-in, and a single TypeScript language across front and back end.

The platform is built in three phases. Each phase is a standalone, fixed-price contract delivering working value on its own. This protects the client's investment — the pilot proves the concept before any commitment to the larger build.

AI voice and messaging carry a real per-use cost. Before subscription prices are set, the platform meters and models cost-to-serve per shop, so plans are always priced with margin. Figures are approximate mid-2026 rates; confirmed against live provider pricing at contract time.

7.1 Per-use cost to serve (per shop)

7.2 Suggested subscription plans

Usage caps matter: voice has real per-minute cost. Each plan includes a fair-use allowance, with overage billed per unit.

8.1 One-Time Build (Phase 1 MVP)

8.2 Third-party accounts & minimum setup

8.3 Approx. monthly platform overhead

Minimum to launch the pilot: verified Twilio + one number, Vapi/Retell account, verified Meta WhatsApp Business account, LLM API key with credit, one VPS, a domain. Upfront out-of-pocket to go live (ex. development) is roughly $50–$120 plus per-use charges.

Work proceeds through a clear, repeatable sequence with regular check-ins.

1. 1

   Discovery

   Confirm languages, order flows, branches, and pilot business details.

2. 2

   Audit & setup

   Provision accounts (Twilio, Vapi/Retell, Meta, LLM, hosting) under client ownership.

3. 3

   Build

   Implement data model, roles, dashboards, and AI channels in the build order below.

4. 4

   Review

   Demo each milestone; gather and apply feedback.

5. 5

   Pilot

   Run the live system with the pilot business; capture usage and cost data.

6. 6

   Handover

   Documentation, walkthrough, and full transfer of code and accounts.

Phase 1 build order

1. 01Multi-tenant database schema, authentication, and role system.
2. 02Order dashboard for all three roles, validated with test orders.
3. 03Twilio + voice AI integration — voice orders flowing into the dashboard.
4. 04WhatsApp Cloud API + conversation agent — chat orders into the same dashboard.
5. 05Missed-call → WhatsApp recovery, tightening the two channels together.
6. 06Customer memory wired into both voice and WhatsApp.
7. 07Lean marketing site, plan assignment, and pilot launch.

Engagement model: fixed-price, paid contract per phase. Phase 1 (MVP / Pilot) is contracted now. Phases 2 and 3 are optional and commissioned separately once the pilot demonstrates value.