Knowledge Management System

> Turn tribal knowledge into searchable, maintained organizational intelligence. Stop losing expertise when people leave.

Phase 1: Knowledge Audit

Current State Assessment

Score each dimension 1-5 (1=nonexistent, 5=excellent):

Total Score: ___/35

Interpretation:
- 28-35: Mature — optimize and maintain
- 21-27: Developing — fill gaps systematically
- 14-20: Basic — needs foundational work
- 7-13: Critical — knowledge is at risk

Knowledge Risk Register

knowledge_risk:
  single_points_of_failure:
    - person: "[Name]"
      unique_knowledge: "[What only they know]"
      risk_if_leaves: "high|medium|low"
      extraction_priority: 1
      extraction_method: "interview|shadowing|recording|pair-work"

  undocumented_processes:
    - process: "[Name]"
      frequency: "daily|weekly|monthly|quarterly"
      complexity: "high|medium|low"
      current_owner: "[Name]"
      documentation_priority: 1

  tribal_knowledge:
    - topic: "[What people 'just know']"
      holders: ["[Name1]", "[Name2]"]
      impact_area: "[What breaks without it]"
      capture_method: "interview|workshop|write-up"

Knowledge Extraction Interview Guide

For each single-point-of-failure person:

1. Context: "I'm documenting [X] so the team isn't dependent on any one person. This protects you too — less interruptions."
2. Process walk: "Walk me through [X] from start to finish. I'll record/note."
3. Decision points: "Where do you make judgment calls? What factors do you consider?"
4. Edge cases: "What are the weird situations that come up? How do you handle them?"
5. Tools & access: "What tools, credentials, or access do you need?"
6. History: "Why is it done this way? What was tried before?"
7. Gotchas: "What are the things that trip people up?"

Output format: Write up as a runbook (see Phase 3 templates).

Phase 2: Knowledge Architecture

Taxonomy Design

knowledge_taxonomy:
  # Level 1: Knowledge Types
  types:
    how_to:
      description: "Step-by-step procedures and guides"
      examples: ["Deploy to production", "Process a refund", "Set up dev environment"]
      template: "runbook"

    reference:
      description: "Facts, specs, configurations to look up"
      examples: ["API endpoints", "Config values", "Vendor contacts", "Pricing tables"]
      template: "reference_doc"

    explanation:
      description: "Why things work the way they do"
      examples: ["Architecture decisions", "Policy rationale", "Historical context"]
      template: "explainer"

    decision:
      description: "How to make specific judgment calls"
      examples: ["Escalation criteria", "Approval thresholds", "Priority frameworks"]
      template: "decision_tree"

    troubleshooting:
      description: "Diagnosis and fix for known problems"
      examples: ["Error codes", "Common failures", "Debug procedures"]
      template: "troubleshooting_guide"

  # Level 2: Domains (customize per org)
  domains:
    - engineering
    - product
    - sales
    - operations
    - finance
    - hr_people
    - customer_success
    - security
    - legal_compliance

  # Level 3: Topics (within each domain)
  # Example for engineering:
  engineering_topics:
    - architecture
    - deployment
    - monitoring
    - incident_response
    - development_workflow
    - testing
    - security
    - infrastructure

Information Architecture Rules

1. Maximum 3 levels deep — if deeper, reorganize
2. One canonical location per topic — link, don't duplicate
3. Every page has an owner — no orphan docs
4. Every page has a freshness date — reviewed within 6 months or flagged
5. Cross-references over duplication — "See [X]" beats copy-paste
6. Search-first design — assume people search, not browse

Naming Conventions

[DOMAIN]-[TYPE]-[TOPIC]-[SPECIFICS]

Examples:
eng-howto-deploy-production
eng-ref-api-endpoints-v3
sales-decision-pricing-enterprise
ops-troubleshoot-billing-failed-charges
product-explain-auth-architecture

Navigation Structure

knowledge_base:
  homepage:
    - quick_links:  # Top 10 most-accessed pages
    - recently_updated:  # Last 10 changes
    - needs_review:  # Stale docs flagged

  by_audience:
    new_hire: "[Onboarding path → essential reading list]"
    engineer: "[Dev setup → architecture → deployment → debugging]"
    manager: "[Policies → processes → templates → reports]"
    customer_facing: "[Product knowledge → troubleshooting → escalation]"

  by_domain: "[Taxonomy Level 2 domains]"
  by_type: "[How-to | Reference | Explanations | Decisions | Troubleshooting]"

Phase 3: Document Templates

Runbook Template (How-To)

# [Title]: [Action verb] + [Object]

**Owner:** [Name]  
**Last verified:** [YYYY-MM-DD]  
**Estimated time:** [X minutes]  
**Difficulty:** Easy | Medium | Advanced  

## Prerequisites
- [ ] [Access/tool/permission needed]
- [ ] [Knowledge assumed]

## Steps

### 1. [First action]
[Specific instruction with exact commands, clicks, or actions]

> ⚠️ [Warning about common mistake at this step]

### 2. [Second action]
[Instructions]

**Expected result:** [What you should see/get]

### 3. [Continue...]

## Verification
- [ ] [How to confirm it worked]
- [ ] [What to check]

## Troubleshooting
| Problem | Likely Cause | Fix |
|---------|-------------|-----|
| [Symptom] | [Why] | [Steps] |

## Related
- [Link to related runbook]
- [Link to reference doc]

Reference Document Template

# [Subject] Reference

**Owner:** [Name]  
**Last verified:** [YYYY-MM-DD]  
**Scope:** [What this covers and doesn't cover]

## Overview
[1-2 sentence summary of what this reference contains]

## [Main content organized as tables, lists, or structured data]

| Item | Value | Notes |
|------|-------|-------|
| | | |

## Quick Lookup
[Most frequently needed items at the top]

## Change Log
| Date | Change | By |
|------|--------|-----|
| | | |

Architecture Decision Record (ADR)

# ADR-[NNN]: [Title]

**Status:** Proposed | Accepted | Deprecated | Superseded by ADR-[NNN]  
**Date:** [YYYY-MM-DD]  
**Deciders:** [Names]  

## Context
[What situation or problem prompted this decision?]

## Decision
[What was decided and why?]

## Alternatives Considered
| Option | Pros | Cons | Why rejected |
|--------|------|------|-------------|
| [A] | | | |
| [B] | | | |

## Consequences
- **Positive:** [Benefits]
- **Negative:** [Tradeoffs accepted]
- **Risks:** [What could go wrong]

## Review Date
[When should this be revisited?]

Troubleshooting Guide Template

# Troubleshooting: [System/Process Name]

**Owner:** [Name]  
**Last verified:** [YYYY-MM-DD]

## Quick Diagnostic

[Flowchart as text]
Is [X] happening?
→ YES: Go to Problem A
→ NO: Is [Y] happening?
→ YES: Go to Problem B
→ NO: Go to Problem C

## Problem A: [Symptom Description]

**Likely causes (in order of probability):**
1. [Most common cause]
2. [Second most common]
3. [Rare but possible]

**Fix for Cause 1:**
[Step-by-step resolution]

**Fix for Cause 2:**
[Step-by-step resolution]

**Escalation:** If none of the above work → [who to contact, what info to provide]

## Problem B: [Next symptom]
[Same structure]

Decision Tree Template

# Decision Guide: [Topic]

**Owner:** [Name]  
**Last verified:** [YYYY-MM-DD]

## When to use this guide
[Situation that triggers this decision]

## Decision Flow

### Step 1: [First question]
- **If [condition A]** → [Action/next step]
- **If [condition B]** → [Action/next step]
- **If unsure** → [Default action or escalation]

### Step 2: [Second question based on Step 1 answer]
[Continue branching]

## Override conditions
[When to ignore this guide and escalate instead]

## Examples
| Scenario | Decision | Reasoning |
|----------|----------|-----------|
| [Real example] | [What was decided] | [Why] |

Phase 4: Contribution System

Writing Standards

The 4C Test (every document must pass all four):
1. Clear — Would a new hire understand this? No jargon without definitions.
2. Correct — Has this been verified by doing/testing? Not from memory.
3. Current — Does this reflect how things work TODAY? Not 6 months ago.
4. Concise — Can anything be cut without losing meaning? Cut it.

Formatting rules:
- Headers: action-oriented ("Deploy to Production" not "Production Deployment")
- Steps: numbered, one action per step, imperative mood
- Warnings: callout boxes, before the step (not after)
- Code/commands: exact, copy-pasteable, tested
- Screenshots: only if truly needed (they go stale fast)
- Links: to canonical sources, never paste full URLs inline

Contribution Workflow

contribution_workflow:
  create:
    trigger: "New knowledge identified (incident learnings, process change, new tool)"
    steps:
      - choose_template: "Match content type to template"
      - draft: "Write using template structure"
      - self_review: "Run 4C Test checklist"
      - peer_review: "SME validates accuracy"
      - publish: "Add to knowledge base in correct location"
      - announce: "Notify relevant teams/channels"

  update:
    trigger: "Existing doc is wrong, incomplete, or stale"
    steps:
      - flag: "Mark as needs-update with reason"
      - update: "Make changes, update 'Last verified' date"
      - review: "If significant change, get peer review"
      - publish: "Update in place"
      - notify: "If behavioral change, announce"

  retire:
    trigger: "Doc no longer relevant (deprecated system, changed process)"
    steps:
      - mark: "Status: Deprecated, add redirect to replacement"
      - archive: "Move to archive after 30 days"
      - redirect: "Ensure all links point to replacement"

Incentivizing Contributions

Making it easy (remove friction):
- Templates pre-filled with structure
- "Quick capture" channel — dump raw notes, someone structures later
- Post-incident: "What would have helped?" → becomes a doc
- Post-onboarding: new hire documents what was confusing
- Meeting notes → action items include "document [X]"

Making it visible (social proof):
- Monthly "top contributors" shoutout
- "Docs champion" rotating role — each sprint, one person owns doc health
- Include documentation in performance criteria
- Knowledge sharing in team meetings (5-min "TIL" segment)

Making it expected (cultural norms):
- "If you answered a question twice, write it down"
- PR template includes "Documentation updated? Y/N"
- Incident postmortem includes "Docs to create/update"
- Onboarding feedback includes "What couldn't you find?"

Phase 5: Search & Discovery

Search Optimization

Every document should be findable by:
1. Title — descriptive, includes key terms
2. Tags — domain, type, audience, technology
3. Synonyms — include alternate terms people might search
4. Problem description — "When [X] happens" phrasing

Tag schema:

document_tags:
  domain: "[engineering|product|sales|ops|finance|hr|cs|security|legal]"
  type: "[howto|reference|explanation|decision|troubleshooting]"
  audience: "[all|engineering|management|customer-facing|new-hire]"
  technology: "[list relevant tools/systems]"
  status: "[current|needs-review|deprecated]"
  difficulty: "[beginner|intermediate|advanced]"

Discovery Mechanisms

1. Contextual links — Related docs linked at bottom of every page
2. FAQ collections — Per-domain "frequently asked" with links to full docs
3. Onboarding paths — Curated reading lists by role
4. Slack/chat bot — "Ask the KB" — searches and returns relevant docs
5. Weekly digest — "New & updated docs this week" email/message
6. Error-page links — Application errors link to troubleshooting docs

Quality Signals

Prioritize search results by:
- Freshness — Recently updated > stale
- Verification — Peer-reviewed > unreviewed
- Usage — Frequently accessed > rarely accessed
- Completeness — Fully structured > quick notes

Phase 6: Knowledge Capture Workflows

Post-Incident Knowledge Capture

After every incident:
1. Immediate (within 24h): Raw timeline and resolution steps
2. Postmortem (within 5 days): Root cause, contributing factors, action items
3. Knowledge extraction (within 10 days):
- New troubleshooting guide? → Create from postmortem
- New runbook needed? → Create from resolution steps
- Existing doc wrong? → Update with correct information
- Architecture decision needed? → Write ADR
- Monitoring gap? → Document what to monitor

Post-Meeting Knowledge Capture

Meeting types that MUST produce knowledge artifacts:
- Architecture review → ADR
- Process change → Updated runbook
- Strategy decision → Decision record
- Customer feedback pattern → Product knowledge update
- Retrospective → Process improvement doc

New Employee Knowledge Capture

First 30 days — new hire documents:
- What was confusing during onboarding
- Questions that weren't answered by existing docs
- Things that were wrong in existing docs
- Suggestions for improvement

Template for new hire feedback:

onboarding_feedback:
  week: "[1|2|3|4]"
  couldnt_find: 
    - topic: "[What they looked for]"
      where_looked: "[Where they searched]"
      how_resolved: "[Asked someone? Found eventually? Still unclear?]"
  wrong_or_outdated:
    - doc: "[Which document]"
      issue: "[What's wrong]"
  suggestions:
    - "[Free text improvements]"

Exit Knowledge Transfer

When someone is leaving:
1. Identify unique knowledge — What do they know that no one else does?
2. Schedule extraction sessions — 1-2 hours per major topic area
3. Record if possible — Video walkthroughs of complex processes
4. Pair them — Have successor shadow for final 2 weeks
5. Review their authored docs — Are they complete? Assign new owners
6. Document tribal knowledge — "Why" questions only they can answer

Phase 7: Maintenance & Freshness

Freshness Policy

freshness_policy:
  review_frequency:
    critical_operations: "quarterly"  # Deployment, incident response, security
    standard_processes: "semi-annually"  # Regular workflows
    reference_docs: "annually"  # Specs, contacts, architecture
    explanations: "annually"  # Background, history, rationale

  review_process:
    - owner_notified: "2 weeks before due date"
    - review_actions:
        - verify: "Is this still accurate? Test/confirm."
        - update: "Fix any outdated information"
        - stamp: "Update 'Last verified' date"
        - skip: "If can't review, reassign or flag"
    - escalation: "Unreviewed after 30 days → manager notified"
    - stale_threshold: "2x review period without update → flagged as stale"

Content Health Dashboard

kb_health:
  date: "[YYYY-MM-DD]"

  coverage:
    total_documents: 0
    by_type:
      howto: 0
      reference: 0
      explanation: 0
      decision: 0
      troubleshooting: 0
    by_domain: {}
    gaps_identified: []

  freshness:
    current: 0  # Reviewed within policy
    needs_review: 0  # Due for review
    stale: 0  # Past review deadline
    deprecated: 0
    freshness_rate: "0%"  # current / (current + needs_review + stale)

  quality:
    peer_reviewed: "0%"
    using_templates: "0%"
    has_owner: "0%"
    has_tags: "0%"

  usage:
    searches_per_week: 0
    failed_searches: 0  # Searches with no results
    top_10_pages: []
    pages_never_accessed: 0

  contribution:
    docs_created_this_month: 0
    docs_updated_this_month: 0
    unique_contributors: 0
    contribution_rate: "0%"  # contributors / total team size

Quarterly Knowledge Review

Agenda (60 min):
1. Dashboard review (10 min) — health metrics trend
2. Gap analysis (15 min) — what's missing? What questions keep being asked?
3. Stale doc triage (15 min) — update, deprecate, or reassign owners
4. Failed searches review (10 min) — what are people searching for and not finding?
5. Process improvements (10 min) — what's working, what isn't?

Phase 8: Knowledge-Driven Automation

Automated Knowledge Triggers

automation_triggers:
  incident_resolved:
    action: "Create task: 'Write troubleshooting guide for [incident title]'"
    assignee: "Incident commander"
    due: "+10 days"

  new_hire_started:
    action: "Generate personalized onboarding reading list from KB by role"

  doc_stale:
    action: "Notify owner, CC manager if unreviewed after 14 days"

  repeated_question:
    threshold: "Same question asked 3+ times in support/Slack"
    action: "Create task: 'Document answer to [question]'"

  process_changed:
    trigger: "PR merged that changes workflow/process"
    action: "Check if related docs need updating, create task if yes"

  failed_search:
    threshold: "Same search term fails 5+ times/week"
    action: "Flag as gap, create task to write missing doc"

Knowledge-Powered Chatbot Design

kb_chatbot:
  flow:
    1_receive_question: "User asks in designated channel"
    2_search: "Semantic search across KB"
    3_respond:
      found_match: "Return relevant doc link + summary"
      partial_match: "Return closest docs + 'Did you mean...?'"
      no_match: "Log as gap, route to human expert, create doc task"
    4_feedback: "Was this helpful? 👍/👎"
    5_improve: "Use feedback to tune search, identify doc improvements"

  sources:
    - knowledge_base_docs
    - slack_saved_answers  # Curated from Slack threads
    - incident_postmortems
    - meeting_notes_tagged_as_knowledge

Phase 9: Cross-Team Knowledge Sharing

Knowledge Sharing Mechanisms

Cross-Team Knowledge Map

knowledge_map:
  engineering:
    produces: ["Architecture docs", "Runbooks", "API specs", "ADRs"]
    consumes_from:
      product: ["PRDs", "User research", "Roadmap"]
      customer_success: ["Bug patterns", "Feature requests", "Usage data"]
      sales: ["Technical requirements", "Integration needs"]

  product:
    produces: ["PRDs", "User research", "Roadmap", "Release notes"]
    consumes_from:
      engineering: ["Technical feasibility", "Architecture constraints"]
      customer_success: ["Feature requests", "Churn reasons"]
      sales: ["Deal requirements", "Competitive intel"]

  customer_success:
    produces: ["FAQ", "Troubleshooting guides", "Best practices"]
    consumes_from:
      engineering: ["Release notes", "Known issues"]
      product: ["Feature docs", "Roadmap"]

  sales:
    produces: ["Battlecards", "Competitive intel", "Use case docs"]
    consumes_from:
      product: ["Feature docs", "Roadmap", "Pricing"]
      customer_success: ["Case studies", "Success metrics"]
      engineering: ["Technical capabilities", "Integration docs"]

Phase 10: Metrics & ROI

Knowledge Management KPIs

ROI Calculation

Knowledge Management ROI:

Time Saved:
  Reduced question-answering = [hours/week] × [avg hourly cost] × 52
  Faster onboarding = [weeks saved] × [new hires/year] × [weekly cost]
  Faster incident resolution = [hours saved/incident] × [incidents/year] × [hourly cost]

Risk Reduced:
  Key person dependency = [probability of departure] × [knowledge reconstruction cost]
  Compliance documentation = [audit prep hours saved] × [hourly cost]

Quality Improved:
  Fewer repeated mistakes = [error rate reduction] × [cost per error]
  Consistent processes = [variance reduction] × [rework cost]

Total Annual Value = Time Saved + Risk Reduced + Quality Improved
Investment = Tool cost + Time spent maintaining KB + Training
ROI = (Total Annual Value - Investment) / Investment × 100

Phase 11: Scoring & Quality

Document Quality Rubric (0-100)

Score interpretation:
- 90-100: Exemplary — reference model for other docs
- 75-89: Good — meets standards
- 60-74: Acceptable — needs minor improvements
- 40-59: Below standard — needs significant work
- 0-39: Critical — rewrite from scratch

Knowledge Base Health Score (0-100)

Edge Cases

Small Team (<10 people)

• Start with a single shared doc/wiki, not a full KB platform
• Focus on: runbooks for critical processes, onboarding guide, decision log
• One person owns KB health (part-time, not full-time)
• Review quarterly, not monthly

Remote/Distributed Teams

• Default to written over verbal knowledge sharing
• Record important meetings/decisions (not all meetings)
• Async-first: every decision documented, not just discussed
• Time zone coverage: ensure docs cover "what to do when the expert is asleep"

Rapid Growth (Doubling in 6 months)

• Prioritize onboarding docs above all else
• Implement "new hire documents what they learn" from day 1
• Assign knowledge buddies — each new person paired with a doc mentor
• Weekly new-hire cohort Q&A → captured and documented

Regulated Industry

• Map compliance requirements to documentation requirements
• Version control with audit trail (who changed what, when)
• Approval workflows for regulated content
• Retention policies aligned with regulations

Post-Merger/Acquisition

• Map both organizations' knowledge structures
• Identify overlaps and gaps
• Prioritize: "how do we work NOW" docs over historical
• Freeze archives of legacy systems/processes

Migrating from Scattered Docs

• Don't try to migrate everything — start fresh with new structure
• Import only: still-accurate, frequently-used docs
• Redirect old locations to new ones
• Set a sunset date for old system
• "If it's not in the new KB, it doesn't exist" (after migration period)

Natural Language Commands