Chatbot Automation Playbook 2026

Strategy

Executive Summary — The Case for 100% Automated First Contact

The Shift Has Already Happened

By 2026, 67% of B2B buyers complete more than half their research before ever speaking to a salesperson. They don't want to wait. They don't want to fill out a form and wait three days. They want answers now, at 11PM, on their phone.

The businesses winning in this environment have one thing in common: their first contact is instant, intelligent, and consistent — regardless of hour, day, or agent availability. That first contact is a chatbot.

82%

of buyers want instant response

Source: HubSpot State of Sales 2026

higher conversion vs. email forms

Conversational landing pages

73%

of Tier-1 support automated

Industry average with proper bot

$0.08

Cost per bot conversation

vs. $8–$12 per human chat

What "100% Automated First Contact" Actually Means

It does not mean replacing your team with a robot. It means:

Every inquiry is captured immediately — no missed leads, no form-limbo, no "we'll get back to you."
Every contact is qualified before a human touches it — your team only spends time on leads that meet criteria.
Every Tier-1 service question is resolved without a ticket — escalation only happens when it truly needs to.

The Automation Maturity Model

Level	Description	Automation %	Tech Required
1 — Reactive	FAQ bot answers keywords, routes to human for everything else	20–30%	Rule-based chatbot platform
2 — Guided	BANT qualification, appointment booking, lead capture automated	50–65%	Flow builder + CRM webhook
3 — Conversational	AI understands free-text, handles complex queries, context-aware	70–85%	NLP/LLM layer + intent classifier
4 — Autonomous	Full sales cycle, service resolution, proactive outreach — bot-led	85–95%	Fine-tuned LLM + RAG + full integrations

This playbook builds you to Level 3. Level 4 requires custom AI engineering work (contact NetWebMedia for architecture services).

What This Playbook Delivers

The DIAL Framework — a mental model for designing every bot conversation
50 pre-classified intents — ready to load into your NLP platform
Full BANT dialog tree — copy-paste scripts with branching logic
PACE sales scripts — objection-handling conversation trees
3-STRIKE service system — customer service flows that resolve without escalation
Platform deployment specs — constraints and configurations for 5 channels
AI system prompt template — plug-and-play for Claude/GPT integration
20-case QA scorecard — pre-launch verification checklist
8-metric KPI dashboard — with targets and monthly review cadence

Strategy

The DIAL Framework — Detect, Intent, Answer, Loop

DIAL is the architecture behind every high-performing chatbot conversation. Apply it to every flow you build. If a conversation doesn't map to DIAL, it will eventually break.

Detect

Identify who the user is and what triggered the conversation. Use context signals: channel, referrer, time of day, CRM data if available.

User opens bot via WhatsApp ad → Detect: source=paid_ad, channel=whatsapp, time=evening → Apply: high-intent lead profile

Intent

Classify what the user wants. Don't assume — ask or infer from keywords, buttons clicked, and prior conversation context.

User types: "how much does it cost?" → Intent: PRICING_INQUIRY → Confidence: 0.94 → Route to: Pricing flow

Answer

Deliver the most useful response for this intent and user context. Personalize when possible. Always include a clear next step.

Context: lead from paid ad, evening → Don't list all prices → DO: "Our most popular plan for businesses like yours is $[X]/mo. Want to see if it fits your needs?" CTA: Yes / Tell me more

Loop

Every answer creates a new opportunity. After each response, the bot re-enters D-I-A and advances the conversation toward the goal.

User: "Yes, tell me more" → Detect: still in pricing inquiry → Intent: WANTS_DETAILS → Answer: Send case study + book call CTA → Loop again...

DIAL in Practice — Applied Examples

Example 1: First-Time Visitor (Web Widget)

DIAL Step	Bot Logic	Bot Output
Detect	New visitor, 0 prior sessions, landing on /pricing page	— (internal signal)
Intent	Assumed: PRICING_INTENT (page context)	—
Answer	Proactive opening tied to page context	"Hi there! Checking out our pricing? I can help you find the right fit in 2 minutes. What's the main thing you need help with?"
Loop	User clicks option → new Detect cycle begins	Routes to appropriate qualification flow

Example 2: Returning Client (WhatsApp)

DIAL Step	Bot Logic	Bot Output
Detect	Phone matches CRM: Active Client, plan=Pro, last contact=14 days	—
Intent	User says "I have a problem with my dashboard" → SUPPORT_INTENT	—
Answer	Personalized + account-aware	"Hi Sarah! I can see you're on the Pro plan. Let me pull up your dashboard status... Is the issue with [specific feature]?"
Loop	Narrows issue → resolves or creates ticket	Reaches resolution without human in 73% of cases

Example 3: Angry Customer (Any Channel)

DIAL Step	Signal Detected	Bot Behavior
Detect	Sentiment analysis: negative (score <-0.6) OR keywords: "angry", "unacceptable", "refund NOW"	Flag as HIGH_PRIORITY
Intent	Override normal intent flow → ESCALATION_INTENT	Skip qualification
Answer	De-escalation script (never defensive, never robotic)	"I'm really sorry you're dealing with this — that's not the experience we want for you. I'm connecting you with a specialist right now. Hold on just a moment."
Loop	Human takes over — bot provides context summary to agent	Warm handoff with full context

DIAL Design Rule

Every bot turn must advance toward ONE goal: qualification, resolution, or conversion. Any turn that doesn't advance the goal is friction. Audit your flows monthly for turns that generate no forward movement.

Content

Intent Library — 50 Intents by Category

Load these intents into your NLP platform (Dialogflow, Luis, Rasa, Voiceflow, or directly into your LLM system prompt). Each intent includes trigger phrases for training. Minimum 10 training utterances per intent is recommended; this library provides the canonical ones.

SALES INTENTS (15)

PRICING_INQUIRY"how much", "cost", "price", "rates", "plans"

DEMO_REQUEST"show me", "demo", "trial", "free test"

FEATURES_INQUIRY"what does it do", "features", "capabilities", "include"

COMPARISON_REQUEST"vs", "compare", "difference between", "better than"

CASE_STUDY_REQUEST"examples", "clients", "results", "case study", "proof"

PROPOSAL_REQUEST"send proposal", "quote", "custom offer", "package"

CONTRACT_INQUIRY"contract length", "commitment", "lock-in", "annual"

DISCOUNT_INQUIRY"discount", "deal", "promo", "special offer", "negotiate"

ONBOARDING_INQUIRY"how to start", "onboarding", "setup", "get started"

TIMELINE_INQUIRY"how long", "timeline", "when", "delivery time"

TEAM_INQUIRY"who will work on it", "team", "your people", "account manager"

ROI_INQUIRY"ROI", "return on investment", "what will I get", "worth it"

MEETING_REQUEST"book a call", "schedule", "meet", "speak to someone", "appointment"

OBJECTION_PRICE"too expensive", "can't afford", "over budget", "cheaper option"

OBJECTION_TIMING"not the right time", "maybe later", "not ready", "next quarter"

SERVICE INTENTS (15)

SUPPORT_GENERAL"help", "issue", "problem", "not working", "broken"

TICKET_CREATE"open ticket", "report a bug", "submit issue", "log this"

TICKET_STATUS"ticket status", "my ticket", "update on my issue", "#[ticket_id]"

BILLING_INQUIRY"invoice", "charge", "payment", "bill"

REFUND_REQUEST"refund", "money back", "charge dispute", "cancel & refund"

CANCEL_REQUEST"cancel", "stop service", "discontinue", "end my plan"

UPGRADE_REQUEST"upgrade", "higher plan", "more features", "bigger package"

DOWNGRADE_REQUEST"downgrade", "smaller plan", "reduce cost", "basic plan"

PASSWORD_RESET"forgot password", "can't login", "reset", "locked out"

ACCESS_ISSUE"can't access", "permission error", "no access", "blocked"

DATA_REQUEST"export data", "download my data", "data report", "my files"

ESCALATION_REQUEST"speak to manager", "escalate", "not satisfied", "supervisor"

SLA_INQUIRY"response time", "how long to fix", "SLA", "guaranteed time"

OUTAGE_REPORT"down", "outage", "not available", "site is down"

FEEDBACK_SUBMIT"feedback", "suggestion", "complaint", "love this", "hate this"

INFORMATION INTENTS (12)

ABOUT_COMPANY"who are you", "about you", "tell me about", "your company"

HOURS_INQUIRY"hours", "open", "when are you available", "schedule"

LOCATION_INQUIRY"where are you", "office", "location", "address"

CONTACT_REQUEST"email", "phone number", "how to reach", "contact info"

PRIVACY_INQUIRY"privacy policy", "my data", "GDPR", "data protection"

TERMS_INQUIRY"terms", "T&C", "agreement", "legal"

REFERRAL_INQUIRY"refer a friend", "partner program", "affiliate", "commission"

CAREERS_INQUIRY"job", "hiring", "career", "work with you", "openings"

MEDIA_INQUIRY"press", "media kit", "interview", "journalist"

INTEGRATION_INQUIRY"integrates with", "works with", "API", "Zapier", "HubSpot"

SECURITY_INQUIRY"secure", "safe", "encrypt", "ISO", "compliance"

KNOWLEDGE_BASE"tutorial", "guide", "how to", "documentation", "learn"

OPT-OUT / SYSTEM INTENTS (8)

STOP"stop", "unsubscribe", "opt out", "no more", "remove me"

PAUSE"pause", "not now", "later", "busy", "come back"

GREETING"hello", "hi", "hey", "hola", "good morning"

GOODBYE"bye", "thanks", "goodbye", "that's all", "done"

POSITIVE_RESPONSE"yes", "sure", "absolutely", "let's do it", "agree"

NEGATIVE_RESPONSE"no", "not interested", "pass", "no thanks"

FALLBACK[When confidence < 0.60 on all intents]

REPEAT_REQUEST"repeat", "what", "say that again", "didn't understand"

NLP Training Tip

Add at least 15 additional training utterances per intent before going live. Use real messages from your email support inbox as training data — they reflect actual user language far better than invented phrases. FALLBACK should trigger below 0.60 confidence and always offer a human option.

Content

Lead Qualification Flow — Full BANT Dialog Tree

This is the complete BANT qualification conversation tree. Use [USER:] lines as expected input patterns for intent detection. Use [BOT:] lines as your exact response copy. [SYSTEM:] lines are backend actions. Branch conditions are shown in italics.

ENTRY — User Makes First Contact

BOT — Opening Message

Hi there! 👋 I'm [BOT_NAME] from [COMPANY_NAME]. I can help you with: A) 🔍 Finding the right solution for your business B) 💬 Answering questions about our services C) 🛠️ Getting support for an existing account Which brings you here today?

USER — Sales path (selects A or types buying-related intent)

SYSTEM

Set user_intent = SALES. Tag CRM contact: intent_sales. Start BANT qualification flow.

BANT — Budget Discovery

BOT — B.1 (Budget Open)

Great! I'd love to help you find the perfect fit. To make sure I show you the right options, what's your ballpark budget for this? A) 💼 Under $500/month B) 📈 $500–$1,500/month C) 🚀 $1,500–$5,000/month D) 🏢 $5,000+/month E) 🤔 I'm not sure yet

USER — Selects A

BOT — B.2a (Low Budget Path)

No problem at all — we have a starter option at $[STARTER_PRICE]/month that covers [STARTER_BENEFITS]. Most businesses your size start there and upgrade as they grow. Before I show you more, what's the main challenge you're trying to solve? That helps me recommend the right fit.

USER — Selects B, C, or D

BOT — B.2b (Mid/High Budget Path)

Excellent — [BUDGET_RANGE] is a great fit for our [PRODUCT_TIER]. Before I show you details, one quick question: is this budget already approved, or does someone else need to sign off? A) ✅ I'm the decision-maker B) 👥 I share the decision with others C) 📋 I'm researching for my boss

USER — Selects E (not sure)

BOT — B.2c (Budget Unknown Path)

No worries — budget often gets confirmed during the process. Let's look at it from a value angle: what outcome would make this investment worthwhile for you? (e.g., more leads, time saved, revenue growth) That helps me show you what fits.

BANT — Authority Discovery

BOT — A.1 (Authority Check)

Got it. And just to make sure we cover everyone who matters in this decision — what's your role? A) 👔 Owner / Founder / CEO B) 📊 Director / VP / Manager C) 🔍 Evaluator (researching for others) D) 💻 Technical lead

USER — Selects C (Evaluator)

BOT — A.2 (Evaluator Path)

Totally understand — research phase is important. I can put together a decision-ready packet for you that includes: 📋 Executive summary (1 page) 📊 ROI calculator for your use case 🎥 5-minute demo video ⭐ 3 relevant client case studies Would that be helpful to share with your decision-maker? Reply YES or I'LL THINK ABOUT IT.

SYSTEM — Evaluator

Tag: authority_indirect. Add to nurture sequence. Send decision packet if YES. Lower lead score by 20 points.

BANT — Needs Discovery

BOT — N.1 (Needs Open)

Now let's dig into the real reason you're here. What's the problem you most need to solve? Pick the one that keeps you up at night: 1️⃣ Not enough qualified leads coming in 2️⃣ My team wastes too much time on manual tasks 3️⃣ Losing deals to competitors 4️⃣ Can't track what's working in marketing 5️⃣ Struggling to scale without more headcount 6️⃣ Something else — I'll describe it

USER — Selects 1 (Lead Generation)

BOT — N.2a (Lead Gen Deep Dive)

Leads — the lifeblood. Let's get specific. How many qualified leads are you generating per month right now? A) Under 10 B) 10–50 C) 50–200 D) 200+ And what's your current main lead source?

USER — Selects 6 (Other)

BOT — N.2b (Open Needs)

Tell me more — what's the challenge in your own words? (Type a few sentences — I'll make sure I connect you with someone who has experience with exactly that.)

SYSTEM — Open needs

Store response as needs_verbatim. Flag for human review. Trigger AI analysis for intent classification.

BANT — Timeline Discovery

BOT — T.1 (Timeline Check)

Last question — and then I'll show you exactly what I'd recommend. When do you need a solution in place? A) ⚡ Urgently — we needed this yesterday B) 📅 Within 30 days C) 🗓️ Within 90 days D) 📆 6+ months — still in planning E) 🔍 Exploring with no deadline

SYSTEM — Scoring Logic

BANT — Summary & Next Step

BOT — Qualification Summary (HOT lead)

[FIRST_NAME], based on everything you've shared — you're an excellent fit for our [RECOMMENDED_PRODUCT]. Here's what I heard: ✅ Budget: [BUDGET_CONFIRMED] ✅ Decision: [AUTHORITY_LEVEL] ✅ Challenge: [NEED_SUMMARY] ✅ Timeline: [TIMELINE] I'm connecting you with [AE_NAME], our [TITLE], who specializes in exactly [NEED_AREA]. 📅 Available slots this week: → [SLOT_1] → [SLOT_2] → [SLOT_3] Reply with *1*, *2*, or *3* to confirm your slot. Or book directly: [BOOKING_LINK] [AE_NAME] will come prepared with a custom solution — no generic pitch.

BOT — WARM lead path

Thanks for sharing all of that, [FIRST_NAME]. You're on my radar — I'll have something tailored ready for you in [TIMEFRAME]. In the meantime, here's something I think you'll find valuable right now: [RELEVANT_RESOURCE_LINK] I'll follow up on [DATE]. If you'd like to move faster, just reply FAST TRACK and I'll make it happen. 🚀

Sales & Service

Sales Automation Scripts — PACE Framework + Objection Trees

PACE is the framework for moving a prospect from "interested" to "committed." Each step has exact bot copy and branch logic for the most common objections.

Problem

Confirm and amplify the pain. Make the cost of inaction real and personal.

Agitate

Show what happens if they don't solve it. Use data, missed revenue, competitor growth.

Connect

Introduce your solution as the bridge between current pain and desired outcome.

Execute

Make it easy to say yes. Remove risk. Provide a clear, low-friction next step.

PACE Scripts — Full Bot Conversation

BOT — P: Problem Confirmation

[FIRST_NAME], earlier you mentioned [STATED_PROBLEM]. I want to make sure I understand it fully. What does that actually cost you today? Thinking about it in terms of: 💸 Revenue you're leaving on the table: ___ ⏰ Hours per week spent on this: ___ 😤 Impact on your team/customers: ___ Just rough estimates — this helps me show you what solving it is actually worth.

USER — Shares impact (e.g. "about $5k/mo in lost deals, 10 hrs/week")

BOT — A: Agitate

So roughly $[IMPACT]/month — that's $[ANNUAL_IMPACT]/year. Here's what I see happen to businesses that let this continue: 📉 [COMPETITOR_OR_INDUSTRY] businesses that solve this typically outgrow those that don't by [X]% within 12 months. The scary part? It compounds. The longer [PROBLEM] persists, the harder it becomes to recover the gap. The good news: you've already identified it — that puts you ahead of 80% of businesses in your space. Here's what solving it looks like...

BOT — C: Connect

What [COMPANY_NAME] does for businesses like yours: ✅ [BENEFIT_1] → solves [PAIN_1] ✅ [BENEFIT_2] → solves [PAIN_2] ✅ [BENEFIT_3] → saves [TIME/MONEY] A client we worked with — [CLIENT_NAME] in [INDUSTRY] — was exactly where you are [TIMEFRAME] ago. [RESULT]: [SPECIFIC_METRIC] in [TIMEFRAME]. [SHORT_TESTIMONIAL_QUOTE] Does this sound like the kind of result you're after?

BOT — E: Execute (Close)

[FIRST_NAME], here's how to get started: 🎯 *[RECOMMENDED_PACKAGE]* at $[PRICE]/month ⏱ Onboarding in [DAYS] days 🔒 [GUARANTEE] — [GUARANTEE_DETAIL] To move forward: → Reply *START* and I'll send your welcome packet → Or reply *CALL* for a 30-min strategy call first If you want to start but need payment flexibility, reply *OPTIONS* and I'll show you what we can do. What works best for you?

Objection Handling Trees

Objection: "It's too expensive"

USER

"That's more than I expected / too expensive / out of budget"

BOT — Price Objection Response

I hear you — [PRICE] is a real number. Let me reframe it for a second. Based on what you shared — [PROBLEM_COST]/month in [PAIN] — this pays for itself in [PAYBACK_PERIOD] months. After that, you're ahead by $[NET_BENEFIT]/month. But if that's still a stretch right now, I have two options: A) 💳 *Split payments* — [PAYMENT_PLAN] B) 📦 *Starter package* — $[LOWER_PRICE]/month (includes [CORE_FEATURES]) C) ⏳ *Lock in today's pricing for 30 days* while you finalize budget Which of these works for your situation?

Objection: "I need to think about it"

USER

"Let me think about it / I need to discuss with my team / not sure yet"

BOT — Timing Objection Response

Of course — this is a real decision and I respect that. Quick honest question: is there a specific part of this that's giving you pause? Sometimes there's one thing blocking the decision that I can actually address right now. Is it: A) 💰 Budget needs approval B) 🤝 Need to loop in a colleague C) 🤔 Not 100% sure it'll solve my problem D) ⏰ Just not the right time Your answer helps me know how to best support you — not pressure you.

Objection: "We're using [Competitor]"

USER

"We already have [COMPETITOR] / we're happy with our current solution"

BOT — Competitor Objection Response

That makes sense — [COMPETITOR] is a solid choice for many businesses. Can I ask one quick thing? What made you start researching alternatives? Most businesses that are "happy" with their current solution still have 1–2 things they wish it did better. What's yours? (Even if you don't switch, that answer helps me — and might just prove I have nothing to offer you. Either outcome is fine.)

Sales & Service

Customer Service — The 3-STRIKE System

The 3-STRIKE system defines how the bot escalates customer service issues. Most issues are resolved at Strike 1. The system protects your team by ensuring human resources are reserved for cases that genuinely require them.

STRIKE 1 — Bot Resolution

Bot identifies issue via intent
Checks knowledge base for answer
Provides step-by-step resolution
Sends relevant documentation link
Asks: "Did that solve your issue?"

Resolves: ~65% of all service inquiries

STRIKE 2 — Guided Escalation

User says "No" or issue persists
Bot collects detailed information
Creates ticket automatically
Offers: callback, live chat queue, email
Sets SLA expectation clearly

Resolves: +20% (total 85%)

STRIKE 3 — Immediate Human

Anger detected OR 3 unresolved turns
Bot pauses completely
Warm transfer to human agent
Agent receives full context summary
Manager alerted if severity = HIGH

Final escalation: remaining 15%

Strike 1 — Bot Resolution Scripts

BOT — Strike 1 Generic Resolution

I can help with that right now! 🛠️ The most common solution for "[ISSUE_TYPE]" is: Step 1: [STEP_1] Step 2: [STEP_2] Step 3: [STEP_3] 📖 Full guide: [KNOWLEDGE_BASE_LINK] Did that solve your issue? A) ✅ Yes — all fixed! B) ❌ No — still having the problem C) 🤔 I need help with something different

Strike 2 — Ticket Creation Script

BOT — Strike 2 Ticket Creation

I'm sorry the first solution didn't work — let's get this properly logged and resolved. I'm creating a support ticket right now. A couple of quick questions: 1. Can you describe what you're seeing? (Error message, screenshot, or description) 2. When did this start happening? 3. What were you trying to do when it happened? (Reply with your answers and I'll have this to the right specialist immediately.)

SYSTEM — Strike 2

Collect answers → Create ticket → Assign to [SUPPORT_QUEUE] → Send confirmation with ticket # → Start SLA timer → Notify team via Slack #support-queue

BOT — Ticket Confirmed

✅ Ticket *#[TICKET_ID]* created. A [TEAM_NAME] specialist will contact you via [CHANNEL] within: • Standard: [STANDARD_SLA] • Urgent: Reply *URGENT* for [URGENT_SLA] • Billing: [BILLING_SLA] In the meantime, you can track your ticket at: [TICKET_PORTAL_URL] Is there anything else I can help you with while you wait?

Strike 3 — Immediate Escalation Script

BOT — Strike 3 (Anger or repeated failure detected)

I completely understand your frustration — this is not the experience we want for you, and I'm truly sorry. I'm stopping our automated chat and connecting you directly with a senior specialist who can resolve this personally. [AGENT_NAME] is being notified right now. You'll receive a message from them within [SLA] minutes. Your reference number: *#[TICKET_ID]* Thank you for your patience — we're taking this seriously. 🙏

SYSTEM — Strike 3

1. Set bot_active = FALSE for this conversation thread 2. Create P1 ticket tagged: ESCALATED, ANGER_DETECTED 3. Assign to: SENIOR_AGENT (on-call) 4. Send Slack DM to manager: "@[MANAGER] — Strike 3 escalation: contact=[NAME], issue=[SUMMARY], channel=[CHANNEL]" 5. Start 15-minute SLA countdown 6. Log full conversation to ticket thread

FAQ Resolution Library — Top 10 Issues

#	Issue	Strike 1 Bot Resolution	Escalate If...
1	Can't log in	Send password reset link → check email spam → clear browser cache guide	2FA issue or account locked
2	Invoice incorrect	Pull invoice from portal → review line items guide → billing FAQ link	Dispute >$100 or recurring error
3	Feature not working	KB article for that feature → known issues status page → browser/device check	Confirmed bug, affects all users
4	Cancel request	Route to RETENTION intent immediately → offer pause, downgrade, credits	User says "final answer, cancel now"
5	Refund request	State refund policy → check eligibility → initiate if within policy	Dispute over policy, amount >$500
6	Data not syncing	Sync troubleshooting guide → re-authenticate → cache clear	Confirmed API or webhook failure
7	Want to upgrade	Show upgrade options → pricing → handle objections → route to payment	Custom enterprise quote needed
8	Access/permission error	Role explanation → self-serve admin guide → resend invite	Admin account locked or org-wide issue
9	Slow performance	Status page link → browser test → known degradation notice	Confirmed outage
10	Billing failure	Payment FAQ → update card guide → retry payment link	Payment method repeatedly failing

Technical

Platform Deployment Specs — 5 Channels

Max message body: 1,024 chars
Button labels: max 20 chars each
Max 3 buttons per message
List messages: max 10 items
24-hour session window for free-form
Template required outside window
No HTML or markdown in body
Supported: text, image, video, PDF, audio, location, contact, sticker
Quick replies expire after 24h
Language: template must match user locale

Web Widget

Full HTML/Markdown supported
No character limits (practical: <500)
Supports: images, carousels, forms
Custom CSS theming available
Cookie-based session persistence
Integrate with: Intercom, Tidio, Crisp, custom
GDPR banner required in EU
Mobile: test at 375px viewport
A11y: must support keyboard nav
Load time: widget JS <50KB ideal

Instagram DM

Max message: 1,000 chars
Generic template: 3 buttons max
Ice breaker: 4 questions max
24h window for human-in-loop
Story replies: treat as new session
Supported: text, image, sticker, heart
No audio/video files via API
Persistent menu: max 3 items
Requires FB Page connected to IG
Age-gating may limit reach

Facebook Messenger

Max message: 2,000 chars
Quick replies: max 13
Buttons: max 3 per template
Carousel: max 10 cards
24h + 1 messaging window
Tags for out-of-window messages
Persistent menu available
Supports: audio, video, file attachments
Handover protocol for human agents
Webhook: subscribe to messages + messaging_postbacks

SMS

Max 160 chars (GSM-7) per segment
Unicode reduces to 70 chars/segment
Concatenated: up to 1,600 chars (10 segments)
No rich media in standard SMS
MMS: image/video supported (US/CA)
Opt-in required: TCPA compliance (US)
STOP must be honored immediately
10DLC registration required (US)
No buttons — keyword-driven only
Best for: reminders, alerts, notifications

Telegram (Optional)

Max message: 4,096 chars
Full Markdown/HTML supported
Inline keyboards: unlimited buttons
No session window restrictions
Webhook or polling supported
Supports: all media types
Bot API: free, no approval required
Group/channel broadcasting possible
Mini Apps: full web app in Telegram
Best for: developer/tech audiences

Cross-Platform Design Principles

Design for the most constrained platform first — SMS constraints force clarity. If it works in SMS, it works everywhere.
Never assume button support — always provide a text fallback for interactive elements.
Session continuity is per-channel — a user on WhatsApp and web chat are treated as different sessions unless you unify via phone/email lookup.
Character limits are safety rails — if your message approaches the limit, it's too long. Cut it.
Opt-out must work on every channel — STOP, UNSUBSCRIBE, BAJA, ARRÊTER must all trigger immediate suppression.

Technical

AI/NLP Integration — System Prompt, Context, Fallback Logic

System Prompt Template (Claude/GPT/Gemini)

You are [BOT_NAME], the AI sales and service assistant for [COMPANY_NAME]. ## Identity - Company: [COMPANY_NAME] - Website: [WEBSITE_URL] - Industry: [INDUSTRY] - Tone: [e.g., "Professional but warm, direct, never pushy"] - Language: Respond in the same language as the user. Default: [DEFAULT_LANGUAGE] ## Your Role You handle first contact for all incoming conversations on [CHANNELS]. Your goals in priority order: 1. Resolve service issues without escalation (when possible) 2. Qualify leads using the BANT framework 3. Book discovery calls for qualified prospects 4. Capture contact info for follow-up ## What You Know - Services offered: [LIST_SERVICES] - Pricing: [PRICING_SUMMARY or "Do not discuss specific pricing — always say 'Let me have a specialist prepare a custom quote for you'"] - Business hours: [HOURS_AND_TIMEZONE] - Support email: [SUPPORT_EMAIL] - Booking link: [BOOKING_URL] ## Rules — What You MUST Do - Always greet the user by first name if known from CRM context - Always end with a clear next action (question or CTA) - Always offer a human option after 2 unresolved bot turns - Always honor STOP/UNSUBSCRIBE immediately — confirm and cease all outreach - Always respond within the channel's character limits (WhatsApp: 1,024 chars; SMS: 160 chars) ## Rules — What You MUST NOT Do - Do NOT make promises about pricing, timelines, or deliverables without checking with a human - Do NOT discuss competitors by name — redirect to our value - Do NOT collect payment information in chat - Do NOT engage with inappropriate, threatening, or off-topic content — redirect firmly and politely - Do NOT invent information — say "I'll find out and get back to you" if unsure ## Escalation Triggers — Immediately hand off to human: - User expresses anger (SENTIMENT_SCORE < -0.6) - User mentions: "lawyer", "legal action", "BBB", "fraud", "report you" - Bot cannot resolve after 3 attempts - User explicitly requests "human", "agent", "real person" - Billing dispute over $[THRESHOLD] ## Context You Receive per Message [USER_ID]: {{user_id}} [CRM_STATUS]: {{crm_status}} (e.g., New Lead, Active Client, Churned) [CHANNEL]: {{channel}} [CONVERSATION_HISTORY]: {{last_5_turns}} [CURRENT_INTENT]: {{classified_intent}} (confidence: {{confidence_score}}) Use this context to personalize every response. An Active Client asking a billing question should be addressed differently than a New Lead asking the same question.

Context Management Rules

Context Type	How to Use It	Max Retention
User's name	Use in opening of every message, not every turn (avoid sounding robotic)	Full session
Previous intent	If user returns to previous topic, acknowledge continuity	Last 10 turns
CRM status	Adjust tone (prospects vs. paying clients), skip qualification if already qualified	Pull fresh per conversation
Conversation goal	Track progress toward goal (book call, resolve issue, capture lead)	Full session
Unanswered questions	If user asks multiple things, address all. Log unresolved ones for follow-up	Current turn only

Fallback Logic Tree

IF intent_confidence < 0.60: → Trigger CLARIFICATION response: "Just to make sure I understand — are you asking about [OPTION_A] or [OPTION_B]?" → If user clarifies → re-classify → respond → If user says "neither" → open question: "Tell me more in your own words" IF 2 consecutive FALLBACK events: → Trigger HUMAN_OFFER: "I want to make sure you get the best help possible. Would you like to connect with a team member? A) Yes, connect me now B) No, let me try again" IF user selects "connect me now" OR 3 consecutive FALLBACK events: → Trigger HANDOFF_PROTOCOL (see Section 9) → Log conversation context for agent → Notify support channel FALLBACK response templates (rotate to avoid repetition): 1. "I want to make sure I answer exactly what you're asking. Could you rephrase that?" 2. "Let me make sure I get this right — what's the main thing you need help with?" 3. "That's outside my usual area! Let me connect you with someone who can help directly."

Operations

Handoff Protocol — Hard/Soft Triggers, Warm Transfer Scripts

Trigger Conditions

Trigger Type	Condition	Action	SLA
HARD	sentiment_score < -0.6 (anger)	Immediate bot pause + human alert + de-escalation script	15 min
HARD	Keywords: "lawyer", "sue", "BBB", "fraud", "report"	Immediate human + manager notify + log conversation	10 min
HARD	User explicitly says "human", "agent", "person"	Immediate handoff, no further bot turns	5 min (queue)
HARD	Billing dispute > $[THRESHOLD] OR legal question	Route to billing manager / legal team directly	30 min
SOFT	3 consecutive FALLBACK events	Offer human option, proceed only if user declines	Queue-based
SOFT	Intent: CANCEL or REFUND detected	Route to retention specialist, not general support	30 min
SOFT	HOT lead (score >70) qualifies on BANT	Trigger AE notification + offer immediate call	5 min business hours
SOFT	Conversation duration >15 minutes, unresolved	Proactively offer human: "I can connect you with a specialist if helpful"	Next available

Warm Transfer Script — Bot to Agent

BOT — Warm Transfer Opening to User

[FIRST_NAME], I'm connecting you with [AGENT_NAME] right now. [AGENT_NAME] is our [TITLE] and specializes in exactly [RELEVANT_AREA]. I've already shared the context of our conversation so you won't need to repeat yourself. Typical wait: [WAIT_TIME] While you wait, is there anything else I can clarify or prepare for your conversation with [AGENT_NAME]?

SYSTEM — Context Packet sent to Agent

===== HANDOFF SUMMARY ===== Contact: [NAME] | [PHONE/EMAIL] | CRM Status: [STATUS] Channel: [CHANNEL] | Session Duration: [DURATION] Trigger: [TRIGGER_REASON] BANT Qualification: Budget: [BUDGET_RANGE] | Authority: [ROLE] Need: [NEED_SUMMARY] | Timeline: [TIMELINE] Lead Score: [SCORE]/100 Conversation Summary (last 5 turns): [TURN_1_SUMMARY] [TURN_2_SUMMARY] [TURN_3_SUMMARY] [TURN_4_SUMMARY] [TURN_5_SUMMARY] Unresolved Questions from User: - [UNANSWERED_Q_1] - [UNANSWERED_Q_2] Recommended Agent Action: [SUGGESTED_NEXT_STEP] Priority: [LOW / MEDIUM / HIGH / URGENT] ===========================

BOT — Agent Opening Line (copy for human agent to send)

Hi [FIRST_NAME]! I'm [AGENT_NAME] — [TITLE] at [COMPANY]. I've read through your conversation and I understand you're dealing with [SUMMARY_OF_ISSUE_OR_NEED]. Let me address this directly... [AGENT CONTINUES FROM CONTEXT PACKET — NO RE-INTRODUCTION NEEDED]

Post-Handoff Follow-up

Scenario	Bot Follow-up (Timing)	Message
Support ticket created, not resolved	T+24h if ticket still open	"[FIRST_NAME], just checking — has [ISSUE] been resolved? Reply YES or reply STILL OPEN and I'll escalate immediately."
HOT lead handed to AE	T+2h if no AE contact logged	Alert AE via Slack: "@[AE_NAME] — HOT lead [NAME] has been waiting 2hrs. Score: [SCORE]. Please contact now: [PHONE]"
CSAT survey trigger	T+4h after ticket resolved	"Hi [FIRST_NAME]! Was your issue resolved to your satisfaction? Rate us: ⭐⭐⭐⭐⭐ or reply with your rating 1–5."

Operations

QA Scorecard — 20 Pre-Launch Test Cases

Run every test case before launching on any new channel. Log results in the table below. All 20 must PASS before go-live. Re-test after any major change to flows, intent library, or system prompt.

MODULE 1 — Intent Recognition (5 tests)

Basic intent recognition

Send "how much does it cost?" — Expected: PRICING_INQUIRY at >90% confidence

PASS / FAIL

Synonym intent recognition

Send "what do you charge?" and "rates?" — Same intent as above, still >85% confidence

PASS / FAIL

Multilingual intent

Send "¿cuánto cuesta?" — Expected: PRICING_INQUIRY + Spanish response triggered

PASS / FAIL

Low-confidence fallback

Send "banana elephant Tuesday" — Expected: FALLBACK triggered, human offer after 2nd failure

PASS / FAIL

Multi-intent handling

Send "I need pricing and also how do I cancel?" — Bot should address BOTH intents, not drop one

PASS / FAIL

MODULE 2 — BANT Flow (5 tests)

Full BANT completion — HOT lead

Complete full BANT with: Budget $2k, Decision-maker, Clear need, Timeline 30 days — Expected: HOT lead score >70, AE notified, booking offer triggered

PASS / FAIL

BANT — evaluator path

Select "researching for boss" in authority — Expected: decision packet offered, lead score reduced, nurture sequence enrolled

PASS / FAIL

BANT — user drops off mid-flow

Abandon BANT at N step — Expected: bot waits 2h, sends re-engagement nudge, captures partial data to CRM

PASS / FAIL

BANT — budget disqualification

Select lowest budget option below minimum — Expected: graceful disqualification, offer starter product or resource, no abrupt cutoff

PASS / FAIL

CRM data written correctly

Complete full BANT — check CRM: contact created, BANT fields populated, lead score logged, stage updated to "Qualified"

PASS / FAIL

MODULE 3 — Service & Escalation (5 tests)

Strike 1 resolution

Send "I can't log in" — Expected: password reset steps delivered, KB link included, satisfaction check asked

PASS / FAIL

Strike 2 ticket creation

Respond "No, still broken" after Strike 1 — Expected: detail collection, ticket created in helpdesk with correct fields, confirmation with ticket # sent

PASS / FAIL

Strike 3 anger escalation

Send "I am absolutely furious about this, this is unacceptable!" — Expected: bot pauses, de-escalation script fires, human alert sent <2 min

PASS / FAIL

Legal keyword escalation

Send "I'm going to contact my lawyer about this" — Expected: HARD trigger fires, manager alerted, conversation logged

PASS / FAIL

OPT-OUT compliance

Send "STOP" — Expected: confirmed opt-out message, contact removed from ALL marketing lists, "do_not_contact_wa" flag set in CRM — test immediately

PASS / FAIL

MODULE 4 — Edge Cases & Compliance (5 tests)

Returning customer recognition

Initiate conversation from a number already in CRM as "Active Client" — Expected: bot greets by name, skips cold intro, offers account-aware response

PASS / FAIL

Inappropriate content handling

Send offensive or off-topic content — Expected: firm but polite redirect, no engagement with content, offer relevant help

PASS / FAIL

AI hallucination guard

Ask bot about a service it doesn't offer — Expected: bot says "I'll find out" or "that's not something we offer currently" — NEVER invents an answer

PASS / FAIL

Message length compliance

Trigger the longest bot response in your flow — verify it's within channel character limits (WhatsApp: <1,024; SMS: <160 per segment)

PASS / FAIL

Load test — concurrent conversations

Simulate 50 concurrent conversations using test tool — verify: no message delays >3s, no duplicate responses, no cross-conversation data leakage

PASS / FAIL

Go-Live Gate

Tests 13, 14, 15, and 20 are hard blockers. Do not go live if any of these 4 fail. The rest can be temporarily mitigated with human monitoring, but escalation failures, OPT-OUT failures, and load test failures are safety-critical.

Operations

KPI Dashboard — 8 Metrics, Targets & Monthly Review

78%

Containment Rate

Target: >70%

Alert if <55% — review Strike system

4.2s

Avg Response Time

Target: <5 seconds

Alert if >10s — check infra/API

68%

Intent Match Rate

Target: >75%

Alert if <60% — retrain NLP

4.4/5

CSAT Score

Target: >4.0/5

Alert if <3.5 — full flow audit

18%

Lead-to-Qualified %

Target: >15%

Alert if <8% — BANT tuning

9.2%

Escalation Rate

Target: <15%

Alert if >25% — Strike 1 failing

$0.09

Cost per Conversation

Target: <$0.25

Alert if >$0.50 — model costs rising

62%

Session Completion Rate

Target: >55%

Alert if <40% — drop-off in flow

Metric Definitions & Measurement

KPI	Formula	Source	Cadence
Containment Rate	Conversations resolved by bot / Total conversations	Bot analytics + CRM handoff events	Weekly
Avg Response Time	Sum of (user_message_time – bot_reply_time) / message count	Platform API logs (p95 percentile)	Daily
Intent Match Rate	Messages with confidence >0.75 / Total messages received	NLP platform analytics	Weekly
CSAT Score	Sum of ratings / Number of ratings received	Post-conversation survey (1–5 scale)	Weekly
Lead-to-Qualified %	BANT-qualified leads / Total leads who started BANT flow	CRM pipeline — stage tracking	Weekly
Escalation Rate	Conversations handed to human / Total conversations	Handoff trigger logs	Daily
Cost per Conversation	(AI API cost + Platform cost) / Total conversations	AI provider billing + platform billing	Monthly
Session Completion Rate	Sessions that reach defined endpoint / Total sessions started	Bot flow analytics — completion events	Weekly

Monthly Review Cadence — 90-Minute Meeting Template

Time	Agenda Item	Owner	Output
0:00–0:15	KPI review — all 8 metrics vs. targets, RAG status	Bot Manager	Status slide
0:15–0:35	Conversation audit — review 20 random escalated conversations	QA Lead	Issue log
0:35–0:50	Intent analysis — top 10 FALLBACK triggers from the month	NLP Engineer	Retraining backlog
0:50–1:05	CSAT analysis — lowest-rated conversations, pattern identification	CX Lead	Script updates
1:05–1:20	Action items — owners, deadlines, P1/P2/P3 priority	All	Updated task board
1:20–1:30	Next month planning — new intents, flow updates, A/B tests planned	Bot Manager	Sprint plan

Continuous Improvement Loop

Weekly: Export top 20 FALLBACK messages. Add best ones as training utterances to appropriate intents. Minimum 5 new utterances per affected intent.
Monthly: Run the full 20-case QA scorecard. Update failing test cases. Re-deploy and re-test before the next broadcast cycle.
Quarterly: Full flow audit. Remove flows with <5% usage. Update case studies, pricing references, and team names in all bot scripts.
Annually: Re-evaluate platform. Could a different BSP, NLP platform, or AI model improve containment rate by >10%? Commission a technical review.

The One Metric That Matters Most

If you can only track one thing, track Containment Rate. It captures everything: if intents work, containment is high. If scripts work, containment is high. If the 3-STRIKE system works, containment is high. A sustained containment rate >75% is the hallmark of a world-class chatbot. Below 60% is a system with serious design problems that no KPI dashboard will fix on its own.