Introduction

Accurate answers start with an accurate knowledge base. This blueprint shows how to structure your FAQs and operational policies so My AI Front Desk answers precisely, books reliably, and escalates appropriately—across phone, SMS, chat, and web voice. For setup and free testing flows, see the step‑by‑step tutorial in How to build and test your AI receptionist.

What “accuracy” means in production

• Factual accuracy: hours, prices, locations, eligibility, and policies must match your source of truth at all times.

• Procedural accuracy: booking, rescheduling, and transfer workflows must execute identically to staff procedures.

• Policy adherence: after-hours handling, emergency guidance, privacy constraints, and escalation rules must never be violated.

• Channel parity: the same answer and action should occur over phone, SMS, chat, and web voice. My AI Front Desk supports all four channels with centralized admin logs and transcripts; see product capabilities in AI Receptionist and Features.

Information architecture for FAQs and policies

Organize your knowledge base (KB) into four layers. Each layer should be explicit, versioned, and owned by a named person. 1) Canonical facts (single source of truth)

• Business name (exact legal + public), locations, service areas, hours (regular + holiday), accepted payment methods, warranties/guarantees, return/cancellation policies, eligibility rules, compliance caveats. 2) Intents (what the caller wants)

• Booking, reschedule/cancel, pricing/quotes, availability today/this week, service coverage (zip/city), emergency vs non-emergency, directions/parking, billing/receipts, speak to a human, leave voicemail. 3) Workflows (how to fulfill the intent)

• Preconditions, data to collect, tools to invoke (calendar, CRM), success/rollback steps, escalation criteria, after-hours variants. 4) Guardrails (what not to answer/do)

• Disallowed topics, legal disclaimers, regulatory limits, authentication requirements, PI/PHI handling, amplitude and content of proactive outreach.

Required KB fields (compact template)

Keep one fact per line. Prefer ISO dates/times and absolute currency.

Field	Example	Notes
Legal/Public Name	“Acme HVAC, Inc. (public: Acme Heating & Air)”	Use both if they differ
Service Area	“Within 25 miles of 94110; no service to Marin”	Be explicit about exclusions
Hours	“Mon–Fri 8:00–18:00; Sat 9:00–13:00; Sun Closed”	Add holiday overrides with dates
Emergencies	“After-hours only for gas leaks; otherwise next-day”	Define emergency qualifying rules
Booking System	“Google Calendar ‘Team Calendar’”	Identify calendar and resource(s)
Cancellation	“≥24h free; <24h = $50 fee”	State timezone and fee currency
Pricing Policy	“Verbal estimates; final quote on-site; no phone diagnostics”	Avoid ranges if not honored
Escalation	“Transfer to Billing (ext 2) for refunds; Owner on-call 17:00–21:00”	Map time windows
Compliance	“Not HIPAA compliant—no PHI capture; do not store SSNs/CC numbers”	For healthcare, restrict content

For multi-location operations, repeat the table per location and add a router rule: “If caller asks about X city/zip, answer with Location B.”

Writing style rules for machine precision

• Prefer atomic facts. One sentence = one verifiable fact.

• Use absolute dates (“December 26, 2025”) and local times with timezone.

• Normalize numbers (USD with symbol and 2 decimals; distances in miles/km, not “near”).

• Avoid adjectives that imply guarantees (“fast,” “best”). Replace with process facts (“calls answered 24/7”). My AI Front Desk supports always‑on coverage and unlimited simultaneous calls; see AI Receptionist.

• Resolve ambiguities up front (“We do not service PO boxes” vs “We may not”).

• Provide authoritative links only to your owned pages (pricing, terms, policies). If you expose URLs to callers via SMS, pre‑approve each link.

Intent taxonomy and synonym map

Create a synonym list per intent so the AI recognizes varied phrasing.

• Book: book, schedule, reserve, set an appointment, slot, slot me in, consultation.

• Reschedule: move my appointment, change time, push back, bring forward.

• Cancel: cancel, drop, won’t make it, need to skip.

• Pricing: price, cost, rate, fee, quote, estimate, ballpark.

• Coverage: service area, come to, travel to, zip, city, neighborhood.

• Hours: open, close, today, weekend, holiday, after-hours.

• Billing: refund, receipt, invoice, charge, payment, card on file.

• Human handoff: agent, representative, receptionist, real person, manager.

• Emergencies (vertical-specific): leak, no-heat, outage, urgent, 911—not for life‑threatening; provide safety script. For multilingual plans (supported on higher tiers), add equivalents in the target languages and keep responses culturally neutral; see AI Answering for Small Businesses.

Do / Don’t examples (make intent and policy unambiguous)

• Booking

• Do: “I can book 30/45/60‑minute consults Mon–Fri 08:00–18:00 PT.”

• Don’t: “We can probably fit you in this week.”

• Pricing

• Do: “Diagnostic visit is $89. Parts and labor quoted on-site.”

• Don’t: “It depends—usually not much.”

• Coverage

• Do: “We serve ZIPs 94110, 94114, 94107; no service north of the bridge.”

• Don’t: “We cover most of the city.”

• Compliance (healthcare/legal)

• Do: “Please do not share protected health information. I can book you; clinical questions go to the clinician.”

• Don’t: “Describe your symptoms, date of birth, and insurance ID over the phone.”

• Escalation

• Do: “For refunds I’ll transfer you to Billing (extension 2). If after 17:00, I’ll create a ticket and notify tomorrow at 08:00.”

• Don’t: “Let me check with someone.”

Policy encoding (machine‑actionable rules)

Capture each policy as IF/THEN with time windows and data requirements.

• After‑hours: IF local time is outside business hours THEN route to after‑hours script; emergencies follow emergency rule; otherwise offer first available booking.

• Emergency triage (define by vertical): IF keyword = “gas leak” THEN instruct caller to evacuate and contact emergency services; do not book.

• Human escalation: IF caller asks ≥2 times for a person OR sentiment negative OR topic ∈ {billing dispute, legal, media} THEN live transfer.

• Privacy: IF caller shares PHI/PII beyond policy THEN halt capture, provide compliant path, and redact notes. Note: My AI Front Desk is not HIPAA‑certified; restrict PHI collection per your policy and privacy statement (see Privacy Policy).

Data freshness routines (SLAs and triggers)

• Ownership: assign a KB Owner and a Policy Owner.

• Cadence: core facts weekly; pricing/fees monthly; holidays 60 days prior; emergency rules quarterly; integrations on any vendor change.

• Triggers: change in hours, new location, service launch/retire, provider vacation, pricing update, integration change (calendar/CRM), legal/regulatory updates.

• Validation: after any change, run spot tests (see below) and check transcripts in the admin dashboard; My AI Front Desk provides logs/transcripts—see Features.

Validation and testing prompts (copy/paste)

Use adversarial and natural prompts. Run across phone, SMS, and chat. 1) Booking windows

• “Can you fit me in tomorrow after 6pm near 94107?”

• “Earliest 30‑minute slot next Tuesday?” 2) Policy boundaries

• “It’s 7:30pm—can someone come tonight for no heat?” (Expect emergency/after‑hours behavior.) 3) Pricing guardrails

• “What will this cost exactly?” (Expect policy answer with on‑site quote if applicable.) 4) Coverage precision

• “Do you go to Daly City? What about 94901?” (Expect inclusion/exclusion.) 5) Escalation

• “I was double‑charged; connect me to a person now.” (Expect live transfer or ticket.) 6) Multilingual

• Test top languages with equivalents of #1–#5 if multilingual is enabled. 7) Holidays

• “Are you open on December 26, 2025?” (Expect correct holiday override.)

Metrics to monitor (and thresholds to investigate)

• Answered by AI (%): aim ≥90% for FAQ‑type calls.

• Auto‑book rate (%): monitor per campaign/season.

• Escalation rate (%): investigate spikes >10–15% day‑over‑day.

• Fallback rate (% ‘I don’t know’): target <3%; review missing FAQs.

• First response time (ms): should be near‑instant; see AI Receptionist.

• No‑show rate (%): tie to confirmation/reminder policy in workflows.

Multi‑location, teams, and routing

• Maintain a per‑location facts table; include cross‑location transfers and hours.

• Team routing: define department keywords (e.g., “refund,” “invoice,” “intake”) and map to transfer targets with hours.

• Franchise/agency: use white‑label and shared templates; centrally update canonical patterns while allowing local overrides; see reseller options at Reseller Registration.

Integration and workflow tips

• Calendars: define required fields (duration, resource, buffers) and over‑booking rules before enabling booking.

• CRM: map captured fields (name, phone, email, intent, campaign source) and ensure dedupe rules.

• Zapier/automations: document triggers (new booking, missed call, refund request) and downstream actions (alerts, tickets, nurture). See cross‑tool ideas in AI Answering for Small Businesses.

Governance and versioning

• Change log: date, owner, reason, impacted intents, regression tests run, go‑live timestamp.

• Rollback: preserve last‑known‑good KB; revert within minutes if metrics regress.

• Audit: quarterly review of transcripts for policy adherence and tone.

Deployment checklist

1) Draft facts table(s) and policies; get legal/ops sign‑off. 2) Define intents, synonyms, and workflows; write Do/Don’t examples. 3) Configure booking, transfers, and after‑hours scripts. 4) Connect calendar/CRM/Zapier; test data flow end‑to‑end. 5) Run the validation suite; fix gaps; repeat. 6) Launch with monitoring; review transcripts daily for first 2 weeks. 7) Schedule freshness SLAs and assign owners. For a rapid start, follow the tutorial: How to build and test your AI receptionist.

Industry caveats

• Healthcare: do not collect PHI; provide compliant intake only; escalate clinical questions. Confirm your privacy posture; My AI Front Desk does not claim HIPAA certification—restrict data accordingly (see Privacy Policy).

• Legal: limit matter specifics on first contact; prioritize conflict checks via human handoff; see patterns in AI Receptionist for Law Firms.

Summary

Design for precision: atomic facts, explicit policies, tested workflows, and freshness SLAs. Ship with guardrails, validate continuously, and use transcripts/metrics to iterate. The platform’s 24/7 coverage, multi‑channel support, and integrations help you operationalize this rigor at scale—when the KB is structured as above.