Chatbot Training Data Specification

Vellocity Partner Support Knowledge Base

Created: 2026-01-09 Version: 1.0 Integration: ChatbotPro + Bedrock Knowledge Base

---

Overview

This specification defines the structure and format for training data used by Vellocity's partner-facing chatbot. The system leverages:

1. Explicit Q&A Pairs - High-confidence, direct answers
2. Document Chunks - RAG-retrieved context for complex queries
3. Capability Metadata - Structured product information

---

Data Types

Type 1: Q&A Pairs (Highest Priority)

Direct question-answer pairs for common queries.

Database Table: chatbot_data Type Value: qa

Format:

{
  "type": "qa",
  "type_value": "What is LinkedIn Insights?",
  "content": "LinkedIn Insights is a Vellocity capability that analyzes your LinkedIn activity to provide GTM intelligence. It shows which content drives AWS Marketplace pipeline, measures audience-ICP alignment, and maps warm introduction paths to decision-makers. It's available on the Command+ plan and costs 5-15 credits per analysis."
}

Guidelines: - Questions should be natural language (how users actually ask) - Include variations: "What is...", "How does...", "Can I...", "How much..." - Answers should be 1-3 sentences, complete but concise - Include specific details (pricing, credits, plan tier)

---

Type 2: Text Content (Document Chunks)

Chunked documentation for RAG retrieval.

Database Table: chatbot_data Type Value: text

Format:

{
  "type": "text",
  "type_value": "LinkedIn Insights - Content-to-Marketplace Correlation",
  "content": "The Content-to-Marketplace Correlation analysis (15 credits) shows which LinkedIn activity drives AWS Marketplace results. It correlates LinkedIn posts with PDP traffic spikes, identifies content themes that precede demo requests, and tracks topics that correlate with Private Offer inquiries. This provides multi-touch attribution across LinkedIn, Website, and AWS Marketplace, with content ROI scoring."
}

Guidelines: - Title should be descriptive and unique - Content should be self-contained (doesn't require context) - Optimal chunk size: 200-500 words - Include the feature/capability name in content

---

Type 3: Capability Metadata (Structured)

Programmatically generated from capability registry.

Format:

{
  "type": "text",
  "type_value": "Capability: linkedin_graph_analysis",
  "content": "**LinkedIn Graph Analysis** (linkedin_graph_analysis)\n\n- **Description:** Analyze LinkedIn social graph for GTM insights\n- **Plan Tier:** Command+ ($799/month)\n- **Credit Cost:** 5-15 credits per analysis\n- **Analysis Types:** Post Performance (5), Audience Insights (10), Relationship Graph (15), ICP Alignment (10), Content Correlation (15)\n- **Category:** GTM Intelligence\n- **Status:** Generally Available"
}

---

Q&A Extraction Templates

Pattern 1: Feature Overview

Q: What is [Feature Name]?
A: [Feature Name] is a Vellocity capability that [primary function]. It helps AWS Marketplace partners [key benefit]. Available on [Plan Tier] at [Price], it costs [Credits] credits per use.

Q: What does [Feature Name] do?
A: [Same as above, rephrased]

Pattern 2: Pricing/Access

Q: How much does [Feature Name] cost?
A: [Feature Name] is included in the [Plan Tier] plan at $[Price]/month. Each analysis costs [X-Y] credits depending on the type. Your plan includes [N] credits per month.

Q: Which plan includes [Feature Name]?
A: [Feature Name] requires the [Plan Tier] plan ($[Price]/month). It's not available on [Lower Tiers].

Pattern 3: How-To

Q: How do I use [Feature Name]?
A: To use [Feature Name]: 1) Navigate to [Menu Path], 2) [First Action], 3) [Second Action]. Results typically appear within [Time].

Q: How do I set up [Feature Name]?
A: Setting up [Feature Name] requires: [Prerequisites]. Go to [Path] and [Steps].

Pattern 4: Comparison

Q: How is [Feature Name] different from doing it myself?
A: Building [Feature Name] yourself would take [Time] and cost $[Cost]. Vellocity provides this capability out-of-the-box, saving you [Value Prop]. We handle [Technical Details] so you can focus on [Partner Focus].

Q: Why should I use Vellocity instead of [Alternative]?
A: [Specific comparison points]

Pattern 5: Capability-Specific

Q: Can [Feature Name] [specific capability]?
A: Yes, [Feature Name] can [specific capability] through the [Analysis Type] feature. This costs [Credits] credits and [Additional Details].

Q: Does [Feature Name] integrate with [System]?
A: [Integration answer with specifics]

---

Extracted Q&A Examples

LinkedIn Insights Q&A Set

[
  {
    "q": "What is LinkedIn Insights?",
    "a": "LinkedIn Insights is a Vellocity capability that analyzes your LinkedIn activity to provide GTM intelligence. It shows which content drives AWS Marketplace pipeline, measures audience-ICP alignment, and maps warm introduction paths to decision-makers. It's available on the Command+ plan."
  },
  {
    "q": "How much does LinkedIn Insights cost?",
    "a": "LinkedIn Insights is included in the Command+ plan at $799/month. Analysis costs range from 5-15 credits: Post Performance (5), Audience Insights (10), Relationship Graph (15), ICP Alignment (10), and Content Correlation (15)."
  },
  {
    "q": "Can LinkedIn Insights show which posts drive AWS Marketplace trials?",
    "a": "Yes, the Content-to-Marketplace Correlation analysis (15 credits) shows which LinkedIn posts correlate with PDP traffic spikes, demo requests, and Private Offer inquiries. It provides multi-touch attribution across LinkedIn and AWS Marketplace."
  },
  {
    "q": "How do I connect LinkedIn to Vellocity?",
    "a": "Navigate to Integrations → Third-Party Services and click 'Connect LinkedIn'. You'll authenticate via OAuth, granting Vellocity read access to your profile and post analytics. We never access private messages."
  },
  {
    "q": "What's the difference between Audience Insights and ICP Alignment?",
    "a": "Audience Insights (10 credits) shows who engages with your content - demographics, engagement patterns, and high-value followers. ICP Alignment (10 credits) specifically measures how well your audience matches your Ideal Customer Profile, with match percentages by segment."
  },
  {
    "q": "Can I see warm introduction paths to prospects?",
    "a": "Yes, the Relationship Graph analysis (15 credits) maps mutual connections to target accounts and suggests warm introduction paths. It shows relationship strength based on engagement patterns and identifies the best path to decision-makers."
  },
  {
    "q": "How long does LinkedIn analysis take?",
    "a": "Most LinkedIn Insights analyses complete in under 60 seconds. Results include key metrics, AI-interpreted patterns, specific recommendations, and an executive summary."
  },
  {
    "q": "Is my LinkedIn data secure with Vellocity?",
    "a": "Yes. We use OAuth 2.0 authentication and never store your LinkedIn credentials. All data is encrypted at rest and in transit. We're SOC 2 compliant and all data stays within AWS (FTR compliant)."
  },
  {
    "q": "How often should I run LinkedIn analysis?",
    "a": "We recommend monthly analysis for ongoing optimization, with additional runs before major campaigns or product launches. The Content Correlation analysis is especially valuable quarterly to track attribution trends."
  },
  {
    "q": "Can I analyze competitor LinkedIn accounts?",
    "a": "Limited analysis is possible using public data only. Full analysis including audience insights and engagement patterns requires connecting your own LinkedIn account."
  },
  {
    "q": "Why can't I just use LinkedIn's native analytics?",
    "a": "LinkedIn's native analytics show engagement metrics but don't connect to AWS Marketplace results. Vellocity's LinkedIn Insights provides the crucial attribution layer - showing which posts actually drove marketplace trials and pipeline. This correlation would take 6-8 weeks to build yourself."
  },
  {
    "q": "What LinkedIn permissions does Vellocity need?",
    "a": "We request read access to: your profile info, post analytics, follower demographics, and engagement metrics. We never post on your behalf or access private messages. You can revoke access anytime from LinkedIn settings."
  }
]

General Platform Q&A Set

[
  {
    "q": "What is Vellocity?",
    "a": "Vellocity is an AI-powered GTM platform for AWS Marketplace partners. It helps ISVs create optimized marketplace listings, manage co-sell partnerships, generate marketing content, and track pipeline attribution - all integrated with AWS Partner Central and Marketplace."
  },
  {
    "q": "How much does Vellocity cost?",
    "a": "Vellocity offers four plans: Launch ($99/month) for getting started, Scale ($299/month) for growing partners, Command ($499/month) for established partners, and Command+ ($799/month) for enterprise features including advanced analytics and co-sell capabilities."
  },
  {
    "q": "Does Vellocity integrate with AWS Partner Central?",
    "a": "Yes, Vellocity integrates with AWS Partner Central for ACE opportunity sync, co-sell pipeline tracking, and CPPO proposal generation. We also integrate with AWS Marketplace for listing management and performance analytics."
  },
  {
    "q": "What AI capabilities does Vellocity use?",
    "a": "Vellocity is powered by AWS Bedrock, using Claude and Amazon Nova models for content generation, analysis, and insights. All AI processing stays within AWS for security and compliance (FTR compliant)."
  },
  {
    "q": "Can I use my own AWS credentials?",
    "a": "Yes, Vellocity supports Bring Your Own Cloud (BYOC). You can connect your AWS account via IAM role delegation, and AI workloads run in your AWS environment using your Bedrock quotas."
  },
  {
    "q": "How do I get started with Vellocity?",
    "a": "Sign up at vellocity.io, complete the onboarding wizard to set up your company profile and brand voice, then connect your AWS Marketplace listing. Most partners are operational within 1 hour."
  },
  {
    "q": "Does Vellocity support Azure and GCP Marketplaces?",
    "a": "Yes, Vellocity supports AWS Marketplace, Azure Marketplace, and Google Cloud Marketplace. You can manage listings across all three from a single dashboard."
  },
  {
    "q": "What is the Co-Sell Hub?",
    "a": "The Co-Sell Hub is Vellocity's partnership management center. It helps you discover co-sell partners, plan joint GTM campaigns, sync ACE opportunities, generate CPPO proposals, and track partnership performance."
  },
  {
    "q": "Can Vellocity help me create AWS Marketplace listings?",
    "a": "Yes, Vellocity's Listing Content Studio generates optimized marketplace listings including product descriptions, feature highlights, and pricing sections. It uses your brand voice and follows AWS Marketplace best practices."
  },
  {
    "q": "How does Vellocity help with SEO for AWS Marketplace?",
    "a": "The SEO + AI Visibility feature analyzes your listing's searchability on AWS Marketplace and tracks mentions in AI assistants like ChatGPT, Claude, and Perplexity. It provides optimization recommendations to improve discovery."
  }
]

---

Import Process

Step 1: Extract Q&A from Documentation

# Run extraction script
php artisan docs:extract-qa --output=storage/chatbot/qa-pairs.json

Step 2: Review and Curate

• Review extracted Q&A for accuracy
• Remove duplicates
• Add missing variations
• Verify pricing/plan information is current

Step 3: Import to Chatbot

# Import Q&A pairs to chatbot
php artisan docs:import-qa \
  --source=storage/chatbot/qa-pairs.json \
  --chatbot=1 \
  --type=qa

Step 4: Import Document Chunks

# Import document chunks for RAG
php artisan docs:import-chunks \
  --source=docs/features \
  --chatbot=1 \
  --audience=partner

Step 5: Train Knowledge Base

# Build vector embeddings
php artisan chatbot:train --chatbot=1

---

Quality Checklist

Q&A Quality

• Question is natural language (how users actually ask)
• Answer is complete but concise (1-3 sentences)
• Specific details included (pricing, credits, steps)
• No jargon without explanation
• Accurate and up-to-date

Document Chunk Quality

• Self-contained (doesn't require external context)
• Includes feature/capability name
• 200-500 words optimal length
• Clear section heading as title
• No broken internal references

Coverage Checklist

• All features have basic Q&A (What is, How much, How to)
• All plan tiers documented
• All credit costs documented
• Integration questions covered
• Troubleshooting questions included

---

Maintenance Schedule

Task	Frequency	Owner
Review Q&A accuracy	Monthly	Support
Update pricing/plans	On change	Product
Add new feature Q&A	On release	Engineering
Review retrieval quality	Weekly	Support
Retrain knowledge base	After updates	Engineering

---

Appendix: Capability Q&A Template

Use this template to create Q&A for each capability:

## [Capability Name] Q&A

### Basic Questions
- What is [Capability Name]?
- What does [Capability Name] do?
- How does [Capability Name] work?

### Pricing Questions
- How much does [Capability Name] cost?
- Which plan includes [Capability Name]?
- How many credits does [Capability Name] use?

### How-To Questions
- How do I use [Capability Name]?
- How do I set up [Capability Name]?
- How do I get started with [Capability Name]?

### Comparison Questions
- Why should I use [Capability Name] instead of [alternative]?
- How is [Capability Name] different from doing it myself?
- What's the ROI of [Capability Name]?

### Technical Questions
- Does [Capability Name] integrate with [system]?
- What data does [Capability Name] access?
- Is [Capability Name] data secure?

### Troubleshooting Questions
- Why isn't [Capability Name] working?
- What if [Capability Name] shows incorrect results?
- How do I contact support about [Capability Name]?

---

Document Version: 1.0 Last Updated: 2026-01-09