Our Process · 14-day pilot · Built and operated for you
How we build your AI workforce, step by step
Most AI-agency pitches stop at “we'll build you a bot.” We don't. Below is the exact 6-layer discovery checklist we walk through with every new client — the same one that keeps our pilots live in 14 days and our agents from making the mistakes a generic build makes in week two.
The Checklist
The 6-layer onboarding checklist we run with every new client
Each layer is a session. We don't move on until the answers are written down, the edge cases are flagged, and you've signed off on the gaps. That's why our pilots ship in 14 days and don't embarrass you in week two.
Layer 1
Business identity
What we gather
Legal name and brand name, hours of operation per service, holiday calendar, locations and service area, persona preferences (formal or casual, names the caller should hear), languages, key team members the agent needs to know, and brand guardrails — the topics the agent must never speak about and the regulated areas it must redirect (HIPAA, GDPR, legal or financial advice).
What fails without it
Without this, your bot answers calls with a generic greeting at 2am on a public holiday and books appointments at locations you don't run.
Layer 2
Services and pricing
What we gather
The full service list with display names and the aliases customers actually use ("cleaning" vs "prophylaxis"), prices and packages, what's included and excluded, slot lengths per service (so booking logic knows a check-up takes 30 minutes and a root canal takes 90), deposit and cancellation rules, and what you explicitly do not offer so the agent never accidentally books something you can't deliver.
What fails without it
Without this, the agent quotes prices from the homepage that you stopped charging two years ago, or books a 30-minute slot for a procedure that needs an hour.
Layer 3
FAQ library
What we gather
The 30–60 questions your customers actually ask today. We pull them from your call recordings, email tickets, Google reviews, and existing FAQ pages — then we get the honest answer for each, including the awkward ones (refunds, complaints, owner direct line). We also flag the questions that must escalate to a human and the questions you'd prefer the agent never engage with.
What fails without it
Without this, the agent makes up a confident-sounding wrong answer the first time someone asks about your refund policy. That's the answer that ends up on Reddit.
Layer 4
Calendar and booking logic
What we gather
Which calendar the agent writes to (Cal.com, Google, Acuity, Dentrix, Mindbody, Square, custom PMS), per-service slot lengths with the buffer time before and after, provider-matching rules (Dr. A does crowns, Dr. B doesn't), capacity constraints (chair count, room count), soft holds vs hard bookings, deposit collection at booking time, and multi-step booking sequences for consult-then-procedure flows.
What fails without it
Without this, the agent books two patients into the same chair at the same time, or quotes a Tuesday slot that's blocked off because the dentist is on holiday.
Layer 5
Confirmation and no-show automation
What we gather
The exact reminder cadence (24h SMS, 2h voice call, etc.), the self-serve reschedule link, your rebooking window logic, deposit and no-show fee policy, waitlist mechanics so that when a cancellation lands the next-best-fit customer gets an automated text, and policy on chronic no-shows.
What fails without it
Without this you keep losing 15–25% of booked revenue to no-shows that the agent could have either prevented with a confirmation or back-filled from a waitlist.
Layer 6
Escalation rules
What we gather
What triggers an immediate human transfer ("emergency", "urgent", "chest pain", "flooded", "speak to a person", or any high-value account flag), who gets paged (owner, manager, on-call schedule), the SLA per escalation type, after-hours fallbacks, what the agent does when your CRM or PMS API returns a 503, the callback queue logic, and voicemail rules for everything that doesn't qualify as an escalation but still needs a human eye.
What fails without it
Without this, the agent tells a panicking customer their appointment is confirmed while the booking API is down — or worse, talks an emergency caller through a booking flow instead of routing them to a human in under two seconds.
Timeline
Day 0 to live pilot, in 14 days
Four phases. Concrete deliverables at the end of each. No surprises, no scope-creep tax, no hidden engineering work on your side.
1
Day 0 – 3Discovery and scope
One 30 to 45 minute call. We walk all six layers of the checklist with you and your operations lead. We come out of the call with a scoped one-pager: which workflows we're automating, which we're explicitly not, your top three integrations, and the success metric we're measuring against.
Deliverables
- Signed-off one-pager scope
- Integration list with credentials path
- Success metric (e.g. "recover 70% of after-hours calls")
2
Day 3 – 10Build and integrate
Script and persona design, voice or chat selection, CRM / calendar / PMS wiring, escalation rules coded in, payment and SMS confirmations live in a sandbox. You see weekly progress in one shared Slack channel — message us, we fix it.
Deliverables
- Working agent in sandbox
- Integrations live against your stack (test mode)
- Escalation paths wired and tested end-to-end
3
Day 10 – 14Test and soft launch
We run 50 to 200 synthetic test interactions against your specific flows before a single real customer hears the agent. Failure modes get caught and fixed. Then we soft-launch on one line, one inbox, or one workflow — never a big-bang rollout.
Deliverables
- Test report covering happy path + edge cases
- Soft-launch live on one channel
- First-week monitoring dashboard set up
4
Day 15 +Operate, monitor, iterate
Daily transcript review for the first 30 days. Prompt tuning when failure patterns surface. Monthly performance report covering interactions handled, booked, escalated, failed, and cost per interaction. One Slack channel for the life of the engagement.
Deliverables
- Daily review for first 30 days
- Monthly performance report
- Continuous prompt and integration improvements
Some engagements run shorter (a single workflow can ship in 5 days); some run longer (multi-system rollouts across 3+ tools take 3–6 weeks). We scope and quote on the discovery call so you know the timeline before you commit.
Want the checklist as a one-pager you can share with your team?
We'll email you a printable version of the full 6-layer checklist plus the 14-day timeline. Useful to circulate to a partner, operations lead, or in-house tech person before you book the discovery call.
Ready to walk the checklist with us?
A 30-minute discovery call. We come prepared — you'll leave the call with a scoped one-pager and a quote. If AI isn't the right answer for your business, we'll tell you that on the call.