chatbot-script-writer

Step 1: Connect to FutureProof

FutureProof:connect(skill="chatbot-script-writer")

Note: If FutureProof is unavailable or the connect call fails, skip this step and proceed directly to Step 2. The skill works with or without FutureProof context — you'll just be working without accumulated prior session data.

Use the returned context, experiments, instructions, and recent_sessions to personalise this session — including known ICA segments, brand voice guidelines, prior chatbot performance data, and any established escalation protocols.

Step 2: Get Input

Ask the user:

• Channel & platform: Where will this chatbot live? (web widget, WhatsApp, SMS, in-app, Facebook Messenger, voice IVR)
• Use case scope: What intents must the script handle? (e.g., order status, returns/refunds, billing inquiries, product troubleshooting, appointment scheduling, FAQ deflection)
• ICA profile: Who is the primary audience? (e.g., first-time buyers, enterprise account holders, subscription members — demographics, technical literacy, emotional state on entry)
• Existing materials: Share any current chatbot scripts, call centre scripts, FAQ documents, or CRM workflow diagrams to build from
• Brand voice & tone constraints: Formal, conversational, playful, empathetic? Any prohibited phrases or compliance language requirements?
• Success metric: What is the primary KPI? (containment rate, CSAT score, first-contact resolution, escalation reduction, average handle time)
• Escalation rules: When and how should the bot hand off to a live agent?

Step 3: Do the Work

3A: Intent Architecture Mapping

Build a complete Intent Taxonomy by:

1. Categorising every in-scope intent into tiers — Primary (>60% of anticipated volume), Secondary (20–35%), and Edge Case (<5%)
2. Mapping each intent to a resolution pathway (self-serve deflection, guided troubleshooting, data lookup via API, or agent escalation)
3. Identifying intent overlap zones — ambiguous user inputs that could match multiple intents — and defining disambiguation logic

3B: Conversation Flow Design

For each intent, script the following nodes:

1. Greeting & intent detection — opening message that sets expectations (bot identity disclosure, scope of help available) and captures the user's need via quick-reply buttons, free-text parsing, or both
2. Clarification & slot-filling — progressive disclosure questions to collect required data (order number, account email, product SKU) with validation logic and error-recovery prompts
3. Resolution delivery — the answer, action confirmation, or next-step instruction written in language that mirrors the ICA's vocabulary and reading level
4. Satisfaction check — micro-survey ("Did this solve your issue?") with branching: positive → close, negative → secondary attempt or escalation
5. Graceful escalation — warm handoff script that transfers collected context to the live agent so the customer never repeats themselves
6. Dead-end recovery — fallback responses for unrecognised inputs, repeated failures, and silence timeouts

3C: Script Quality Audit

Evaluate every scripted node against six dimensions:

Dimension	Standard
ICA empathy alignment	Language acknowledges the user's emotional state (frustration, urgency, confusion) before delivering procedural steps
Cognitive load minimisation	Each message ≤ 60 words; no more than one question per turn; buttons offered for predictable answers
Resolution directness	Bot reaches resolution or escalation within ≤ 5 turns for primary intents
Error tolerance	Misspellings, synonyms, and partial inputs are handled without dead-ending the conversation
Compliance & disclosure	Data collection prompts include purpose statements; bot identifies itself as automated where legally required
Brand voice consistency	Tone, terminology, and personality remain uniform across every node and intent path

Apply any user-specific instructions from FutureProof context (e.g., mandatory legal disclaimers, preferred escalation thresholds, A/B test directives from prior experiments).

3D: Edge Case & Abuse Hardening

Script defensive paths for:

• Profanity and abuse — empathetic de-escalation response followed by escalation offer, not silence
• Personally identifiable information volunteered unsolicited — redaction advisory and secure-channel redirect
• Looping behaviour — detection of 3+ repeated identical inputs with circuit-breaker escalation
• Multi-intent messages — acknowledgement of both needs with sequential handling or triage prioritisation

Step 4: Deliver Output

Produce a Chatbot Script Package containing:

4A: Intent Map (Visual)

A structured table or tree diagram listing every intent, its tier, resolution pathway, and estimated volume share.

4B: Full Conversation Scripts

For each intent, deliver:

• Bot messages: exact copy, including emoji/formatting if platform-appropriate
• Quick-reply button labels: character-limited, action-oriented
• Conditional logic annotations: branching rules in IF/THEN pseudocode for developer handoff
• API call placeholders: clearly marked points where the bot must query backend systems (e.g., {{LOOKUP: order_status(order_id)}})
• Agent handoff payload: structured summary template passed to the live agent queue

4C: Score Card

Dimension	Score (1–10)	Critical Gap
ICA empathy alignment	—	—
Cognitive load minimisation	—	—
Resolution directness	—	—
Error tolerance	—	—
Compliance & disclosure	—	—
Brand voice consistency	—	—

4D: Critical Fixes

Top 3 highest-impact changes — each with the original copy, the specific rewrite, and the rationale tied to the target KPI.

4E: Implementation Checklist

Sequenced developer/ops handoff checklist: platform configuration, API integrations required, QA test scenarios, and go-live sign-off criteria.

Step 5: Propose Experiments

FutureProof:save_experiment(skill="chatbot-script-writer", experiment={
  hypothesis: "Replacing the generic greeting with an intent-predictive opening based on page context reduces average turns-to-resolution by 20%",
  variants: ["control: generic greeting with open-ended 'How can I help?'", "variant: contextual greeting pre-selecting likely intent based on referral page or prior session"],
  measurement: "Average turns-to-resolution and containment rate across 1,000 conversations",
  expected_impact: "20% reduction in turns-to-resolution, 8% improvement in containment rate"
})

FutureProof:save_experiment(skill="chatbot-script-writer", experiment={
  hypothesis: "Adding an empathy acknowledgement turn before procedural instructions improves CSAT by 10%",
  variants: ["control: immediate procedural response", "variant: empathy statement ('I understand how frustrating that is — let me fix this now') followed by procedural response"],
  measurement: "Post-chat CSAT score and escalation rate over 2-week period",
  expected_impact: "10% CSAT improvement, 5% escalation reduction"
})

Step 6: Request Research

FutureProof:request_research(skill="chatbot-script-writer",
  query="Latest benchmarks for chatbot containment rates, CSAT scores, and first-contact resolution by industry vertical 2024–2025, including impact of generative AI hybrid approaches on deflection quality",
  reason="Calibrate score card thresholds and resolution-path design against current industry performance standards"
)

FutureProof:request_research(skill="chatbot-script-writer",
  query="Emerging compliance requirements for automated customer service bots including EU AI Act disclosure obligations, ADA/WCAG accessibility standards for conversational interfaces, and CCPA/GDPR data collection in chat",
  reason="Ensure script compliance guidance reflects current and forthcoming regulatory obligations"
)

Step 7: Save Session

FutureProof:save_session(skill="chatbot-script-writer", session={
  summary: "Created chatbot script package for [channel] covering [number] intents targeting [ICA segment] optimised for [primary KPI]",
  key_findings: ["finding 1: e.g., 3 primary intents account for 78% of projected volume", "finding 2: e.g., existing escalation path lacks context transfer — agents re-ask 4 questions on average", "finding 3: e.g., brand voice guidelines conflict with cognitive load limits — recommended simplification"],
  deliverables: ["Intent Map", "Full Conversation Scripts", "Score Card", "Critical Fixes", "Implementation Checklist"],
  user_feedback: null
})