The 7-Day Custom AI Team Build (And What Actually Gets Built)

Every AI vendor pitches "up and running fast." Almost none of them define what fast means, what "running" means, or what "up" leaves on your plate. When we say a first agent is live in seven days, here is what happens inside those seven days — the meetings, the decisions, the code, the tests, and what actually greets you on Day 8.

This is written for SMB owners who have already been through a failed automation buy or a stalled Zapier project. The pattern below is the same across an HVAC shop in Tampa, a dental practice in Cleveland, and an e-commerce brand in Melbourne. Different tools, same skeleton.

Day 1 — the intake call (45 minutes)

Not a demo. Not a discovery-plus-pitch. The Day 1 call has one goal: name the workflow bleeding the most time this week and get concrete about what "done" looks like. You describe the pain in your own words — missed calls, cold quotes, aging AR, no-show recovery, ticket triage — and we ask 15 to 20 questions until we can write the agent's job description in two paragraphs.

By end of call you have: the one workflow we are shipping first, the tools it will touch, the humans it will hand off to, and a written definition of "success in 7 days." No slides.

Days 2–3 — role scoping and workflow map

We audit the actual stack. Not what the sales page says you own — what is live, logged in, and paid for. CRM version, phone system, calendar, quote tool, invoicing, whichever inbox actually gets read. We also read what your team already does manually so the agent inherits the tribal knowledge, not a stripped-down version of it.

Output of Days 2–3: a one-page workflow map with the triggers (inbound call, inbound form, new invoice past 30 days, incoming ticket), the decisions the agent will make on its own, the handoffs to humans, the success signals, and the safety rules (never quote below floor, never book without phone readback, always escalate when confidence is low). This gets signed off before we build.

Days 4–5 — build in your existing stack

The build lives inside the tools you already pay for. We do not migrate you off ServiceTitan, Dentrix, HubSpot, HighLevel, or Shopify. We add the agent layer on top and wire it into the primitives you use every day. If a piece of substrate is missing — usually a phone number, a webhook endpoint, or an outbound-SMS number — we provision it and hand you the credentials.

During Days 4 and 5, we ship the agent, load your knowledge base into it (price book, service areas, hours, escalation rules, common objections, past-decision patterns), and stand up the dashboard. You get a link to the dashboard on Day 5 evening.

Day 6 — testing and role-play

Day 6 is when we run the agent through every scenario we can invent, plus every scenario you have seen in the last 30 days. For a receptionist that means new-lead intake, after-hours emergency, out-of-service-area, angry customer, returning customer whose caller ID we recognize, and price-haggle attempts. For a sales navigator it means aged-quote follow-up, quote-with-discount request, wrong-decision-maker, and reactivation of a lead who went cold 90 days ago.

We record the role-play, share it with you, and mark the ones that need a config change vs. the ones that need an escalation-to-human path. Everything gets a decision. Nothing ships silent.

Day 7 — launch and monitor

The webhook goes live. If it is a phone-based agent, Twilio or your existing carrier gets pointed at the new endpoint. If it is a form-based agent, the form action gets swapped. If it is a triage agent, the inbound inbox rules get updated. From Day 7 forward, every interaction is logged, every escalation is timestamped, and every dollar of pipeline created is attributed to the agent.

Day 7 also installs a soft rate cap for the first 24 hours so you can watch the first live traffic in real time without a runaway. Once the first day is clean, the cap lifts.

What "the team" looks like by tier

The 7-day build ships the first role. The tier decides how many roles run alongside once you scale up.

AI Operator — $1,500/mo. One agent, deep. Three real examples we ship most often at this tier: (1) AI receptionist for a solo dental practice or single-truck HVAC shop — owns inbound calls, books, escalates emergencies. (2) AI sales navigator for a residential realtor — owns speed-to-lead intake, qualifies, books discovery. (3) AI ticket triage for a Shopify brand doing under $2M — classifies inbound support, drafts recommended actions, only escalates true edge cases.

Founding CTO — $2,500/mo. Two to three agents working together. Three real examples: (1) HVAC shop — receptionist + sales navigator on cold quotes + dispatch-brief morning agent. (2) Boutique law firm — intake triage + conflicts checker + client-comms follow-up agent. (3) Multi-location dental group — recall + no-show recovery + insurance follow-up trio on a single dashboard.

Full Custom CTO — $3,500/mo. The full four-role AI department, plus custom code where the off-the-shelf substrate does not cover the workflow. Three real examples: (1) Roofing contractor with 12 crews — receptionist + sales navigator + office manager + AI bookkeeper, all wired to a custom crew-dispatch dashboard. (2) Real estate brokerage — intake + ISA + transaction coordinator + commissions bookkeeper. (3) Multi-location dental group with a custom PMS integration and its own outbound-marketing agent. This is the premium tier where the "custom" part earns its name — every SMB in this bracket needs at least one workflow the market has not productized yet.

What happens in Week 2 and beyond

Once the first agent is stable, we run a weekly optimization sprint — adjust prompts based on real conversations, add rules for new edge cases, tighten handoff points. Every 30 days you get a written measurement of what the agent shipped: calls answered, quotes sent, leads booked, AR recovered, tickets closed. If a tier upgrade is warranted (usually because a second workflow is now the bleeding one), we scope the next role on the same intake pattern — 45-minute call, 7-day build, live on Day 8 of that role's sprint.

What Day 7 does not look like

It does not look like a demo of what could exist if you configured it right. It does not look like a Zapier board you inherit. It does not look like a login you have to babysit. It looks like a role. You know the receptionist is on because the phone is being answered and the calendar is filling. You know the bookkeeper is on because the AR aging report gets shorter every week. You do not have to open the dashboard for it to work; the dashboard is there when you want to verify.