AI Sales System Build (Revenue Command) — n8n + Retell AI + Twilio + CRM Integrations (Multi-Client SaaS)

We run a sales and AI systems consultancy in Australia. We've already built and proven an AI sales stack for our own business — conversational SMS agents, AI voice agents, automated pipeline management and a live revenue dashboard. It works. It's producing real bookings and real revenue right now.

We're now productising it so we can deploy the same system to other sales-driven businesses: recruitment agencies, real estate agents, business brokers and similar.

We need one developer to build the productised, multi-client version. You are not inventing this from scratch — the logic, the sequences, the prompts and the workflows already exist and are running. Your job is to rebuild them properly as a repeatable, multi-tenant system that we can deploy to a new client in under a week instead of three.

What you'd be building

A four-layer system, deployed per client, managed from one place:

1. Daily briefing engine Automated morning brief and end-of-day close-out delivered to the client (email + SMS + dashboard). Pulls from their CRM and calendar: what's booked today, who went quiet, who's due for follow-up, what moved yesterday, what needs a human. Not a data dump — a prioritised, written brief.

2. Conversational SMS agent Two-way AI SMS that qualifies inbound enquiries, books appointments, chases no-shows, nurtures cold leads and escalates to the human when the conversation needs it. Must handle real replies (not just broadcast), write back to the CRM, respect opt-outs, and know when to stop.

3. AI voice agent Outbound calling of new leads and enquiries using Retell AI over Twilio, with an Australian SIP trunk. Batch campaigns plus real-time speed-to-lead calling. Call outcomes, transcripts and dispositions sync back to the CRM. Books directly into the client's calendar.

4. Client dashboard A web dashboard where the client sees their pipeline, lead flow, agent activity, conversion rates by stage, and ROI (what the system cost vs what it produced). Login per client. This is the thing they open every morning — it needs to be genuinely good, not a wall of numbers.

Plus: a provisioning layer so onboarding a new client is configuration, not development.

Integrations required
CRM: HubSpot and GoHighLevel at minimum. Pipedrive and Salesforce as fast-follows. Must be built behind an abstraction layer — we will add CRMs and we do not want to rewrite the agents each time.
Calendar: Google Calendar, cal.com, Calendly

Email : Gmail / Google Workspace, Microsoft 365
Telephony: Twilio (Australian AU1 region), Retell AI
LLM: your recommendation — we care about cost per conversation, not brand names
Tech stack

Our current production stack is Make.com + Retell AI + Twilio + HubSpot. For this build we expect to move the automation layer to self-hosted n8n for cost and multi-tenancy reasons — but if you can make a strong case for a different approach, we want to hear it in your proposal. Tell us what you'd actually do and why.

For the dashboard, we're open on the stack. Something you can maintain, that we can hire for later.

Milestones

Milestone Deliverable Est. hours

1 Foundation Multi-tenant automation infrastructure, client config layer, credential/secrets handling, logging and error alerting. One test client provisioned end to end. ~140
2 Integrations CRM, calendar and email connectors behind a common abstraction. Two-way sync verified against a live sandbox. ~110
3 SMS agent Conversational agent live: inbound reply handling, qualification, booking, escalation, opt-out compliance, full CRM write-back. ~100
4 Voice agent Retell + Twilio AU trunk live. Batch dialler + speed-to-lead trigger. Outcomes, transcripts and recordings synced. Calendar booking from call. ~100
5 Briefing engine Morning brief + daily close-out generating and delivering reliably for the test client. ~55
6 Dashboard Client-facing dashboard, authenticated, live data, ROI reporting. ~150
7 Provisioning + handover New-client setup tooling, templates, documentation, video walkthrough, 30 days of bug support. ~60

~715 hours core + ~15% QA/revision allowance.

Milestones are paid on acceptance. We'd like to start with Milestone 1 as a standalone engagement — if it goes well, the rest follows without re-negotiation.

What we need from you
Demonstrated production work with AI voice agents — Retell, Vapi, Bland or similar. Not a tutorial project. Something real, with real call volume.
Demonstrated n8n or Make.com work at a level beyond simple linear automations — error handling, retries, queues, sub-workflows, multi-tenancy.
Twilio experience including SIP trunking and non-US number provisioning.
Real CRM API work — HubSpot or GoHighLevel specifically is a big plus.
Front-end capability for the dashboard, or a named partner you work with (tell us who, up front).
Written English good enough to work asynchronously across a 5–12 hour timezone gap.
You test your own work. We are not your QA team.
What we don't want
Anyone who has only ever built demos
Agencies who bid, win, then hand it to a junior — we want to talk to and work with the person doing the build
Anyone who needs a spec for every decision. We'll give you the logic and the outcomes; you own the technical calls.
Terms
All code, workflows and IP assigned to us on payment. NDA before we share our existing prompts, sequences and workflow logic.
Weekly 30-minute call at a time that works for both of us, plus async updates.
We supply all accounts, API keys and test data.
Ongoing retainer available after handover for maintenance and new client onboarding — this is the outcome we're hoping for.
Screening questions (put these in the Upwork job form)
Link us to an AI voice agent you have built that has handled real calls at volume. What was the call volume, and what was the connect-to-booking rate?
We're planning to move from Make.com to self-hosted n8n for multi-tenancy and cost. Do you agree with that call? If not, what would you do instead and why?
How would you architect this so that onboarding client #40 takes a few hours of configuration rather than a few days of development? Be specific.
What's your realistic all-in cost per connected AI call minute for a call to an Australian mobile, and how would you bring it down?
What hours do you work, in UTC?
Paid test task (do this before you award the full project)