Workflow

Knowledge Base Maintenance Workflow

Your help center stays accurate as the product changes — without a technical writer on payroll.

Your Intercom help center has 140 articles. 40 are from your 2023 product that got renamed. 20 reference UI that no longer exists. Support still answers 'how do I export data?' 8 times a week because the article about it got buried under /advanced/exports-v2/deprecated. Deflection rate is 12%. You should be at 35%+.

Free to startNo credit card requiredUpdated Apr 2026
Tycoon solution

AI Customer Support + AI Head of Content run a continuous KB maintenance cycle. New articles drafted from patterns in recent tickets, stale articles flagged against product changelog, screenshots refreshed from deploys, SEO optimized, and deflection rate tracked. Your KB becomes a living system that grows with the product.

How it runs

  1. 1
    Gap detection from tickets

    AI Customer Support scans inbound tickets (Intercom, Zendesk) for patterns: 5+ customers asking the same question in 14 days without a good KB article = a gap. Generates a draft article titled with the customer's exact language ('how to export data to CSV' not 'Data Export Documentation').

  2. 2
    Draft from resolved tickets

    When a ticket is resolved, AI Customer Support extracts the resolution into a draft KB article. Steps the agent took + screenshots + caveats. Editor reviews and publishes. One resolved ticket becomes one scalable article.

  3. 3
    Staleness detection

    Cross-references KB articles against product changelog (from /workflows/release-notes-production). Articles referencing removed features, renamed UI, or deprecated APIs get flagged. Auto-updates simple changes (renamed button, moved menu item); harder changes surface for human review.

  4. 4
    Screenshot refresh on deploy

    For articles with UI screenshots, AI Head of Content re-captures screenshots after each deploy that touches the relevant pages. Runs in Playwright headless, matches the old screenshot coordinates, replaces in the article. No more 'the button looks different from the screenshot' tickets.

  5. 5
    SEO and discoverability

    Each article optimized for help-center search: title matches customer language, H2s answer sub-questions, keywords placed naturally, related articles linked, structured data for FAQ schema. Articles rank in Google for 'how do I X in [your product]' — reducing tickets from people who Googled instead of logging in.

  6. 6
    Deflection rate tracking

    AI Customer Support measures deflection weekly: articles viewed per ticket, tickets referencing KB articles (good), tickets about topics WITH KB articles (the article isn't discoverable — fix!). Dashboard shows which articles are pulling weight and which are dead weight.

  7. 7
    Quarterly KB audit

    Every 90 days, AI Head of Content does a full audit: retire articles with <10 views/quarter that aren't tied to edge cases, consolidate duplicates (3 articles about the same thing merge into 1 canonical), update all 'last verified' dates, refresh tone for consistency. Output: a leaner, more accurate KB.

Who runs it

hire/ai-customer-supporthire/ai-head-of-contenthire/ai-cto

What you get

  • KB articles match current product, not 6 months ago
  • New articles written within 48 hours of a support pattern emerging
  • Screenshots refresh automatically after UI changes
  • Deflection rate rises from 10-15% to 30-40% over 6 months
  • Support ticket volume drops 20-30% on topics with strong KB coverage
  • KB articles rank in Google for high-intent help queries
  • Zero 'dead' articles — quarterly audit retires what doesn't earn its keep

Frequently asked questions

Our product changes weekly. Can the KB really keep up?

Yes, because the KB maintenance runs on the same release cadence as the product. Every release-notes entry (from /workflows/release-notes-production) triggers a KB check: what articles reference the changed feature? AI Head of Content updates affected articles the same day the feature ships. For major UI overhauls, a bigger refresh batches 10-30 article updates. The KB stays current because it's wired into the release pipeline, not treated as 'documentation we'll update later'.

What about embedded video tutorials — those are expensive and go stale too.

Video is the one place AI doesn't auto-fix. Videos become stale when UI changes but re-recording isn't cheap. AI Head of Content handles this two ways: (1) flags videos referencing deprecated UI and adds a text overlay note ('UI has changed slightly — current path: Settings > Data > Export'), and (2) prioritizes video refresh for articles with >1000 views/quarter, leaves lower-traffic videos as-is. For new tutorials, AI Head of Content writes the script + storyboard; you or a contractor does the recording. Saves 70% of production time.

How does it handle multi-language support — we have customers in EN, ES, JA, DE?

Full i18n workflow. New article drafted in primary language first (usually English), then translated via DeepL + human review for the supported locales. Updates to the primary article propagate to translations with version tracking (so a JA translation doesn't silently diverge from EN). SEO optimization per locale — different keywords rank in different languages. Locale-specific KB search configured in Algolia or your help-center's built-in search.

Our customers are developers. They prefer docs in GitHub README vs a help center. Does this apply?

Developer-facing docs are a different pipeline — more like /workflows/api-docs-maintenance. Tycoon distinguishes: developer docs live in the repo (README, docs/ folder, OpenAPI spec), end-user help lives in Intercom/Mintlify help center. Workflow principles are the same (gap detection from tickets, staleness detection from product changelog, SEO optimization) but the tools and tone differ. Teams with both usually run both workflows — developer docs for integrations, help center for admin/billing/account questions.

Can it actually improve deflection rate or is it just busy work?

Deflection improvements are measurable but take 90+ days to show. The mechanism: articles become discoverable (SEO + in-app search), accurate (match current product), and comprehensive (cover patterns from real tickets). Typical trajectory: month 1-2, minimal change (new articles indexing, old articles cleaned up). Month 3-6, deflection rises 5-15 points (30% → 40%, for example). Month 6+, compound gains as customers learn to search your KB first. The ones who don't see gains usually have an in-app search or help-widget problem, not a KB content problem — AI Customer Support flags which is the bottleneck.

Related resources

Run your one-person company.

Hire your AI team in 30 seconds. Start for free.

Free to start · No credit card required · Set up in 30 seconds