Architecture Decks · AI Strategy

System architectures.

Four governed intelligence systems (retention, competitive, sentiment, and content), each rolled up into a bespoke command center, then unified into one governed platform the executive team works from.

Architecture 01 · Customer Retention

Customer Retention Intelligence

GoalSurface churn risk from real customer conversations.OutcomeCSMs act on at-risk accounts before the renewal, not after it.

SOURCEINGEST + STOREINTELLIGENCEACTIONGongcall transcriptsHubSpotCRM + deal contextScrub Pipeline { }PII strip · normalizeSnowflakeclean data storeClaude Codechurn scoring · play { }SlackCSM risk alertsHubSpotrisk score write-backCONSUME · EXECUTIVE SURFACEF3i · BESPOKE { }Retention Command CenterCustom report layer reading the governed data.Monthly · quarterly · ad hoc churn briefings for the CRO.

System. Gong transcripts and HubSpot context flow through a custom scrubbing pipeline that strips PII and normalizes records into Snowflake. Claude Code scores churn risk per account and drafts the recommended play, then writes the signal back to HubSpot and alerts the owning CSM in Slack. The governed result rolls up into a bespoke command center for executive review.

Service
Claude core
Custom { }
Command center
Architecture 02 · Competitive Intelligence

Competitive Intelligence Radar

GoalTrack every competitor move on features, pricing, and positioning.OutcomeA ranked, always-current view of the field, refreshed on a schedule.

TRIGGERCRAWLANALYZEOUTPUTGitHub Actionsscheduled cronApifysite crawlPlaywrightJS-render scrapeNeonsnapshot storeClaude Codediff · extract · rank { }Notioncompetitive boardGoogle Sheetsranking tableSlackweekly digestCONSUME · EXECUTIVE SURFACEF3i · BESPOKE { }Competitive Command CenterCustom report layer reading the ranked intelligence.Quarterly landscape review · ad hoc threat briefings for the CMO.

System. A scheduled GitHub Actions job runs Apify and Playwright crawlers across competitor sites. Snapshots land in Neon, where Claude Code detects changes, extracts the substance, and ranks moves by threat level. Results post to a Notion board, a Google Sheets ranking table, and a Slack digest, and roll up into a bespoke command center for executive review.

Service
Claude core
Custom { }
Command center
Architecture 03 · Market Sentiment

Market Sentiment Listening

GoalHear what the market actually says about your brand and products.OutcomeEarly trend signals, not lagging vanity metrics.

HARVESTLANDANALYZESURFACEApifyforum + community crawlPlaywrightJS-render scrapePostgresraw postsClaude Codesentiment · themes · trends { }SnowflakeaggregatesVercelsentiment dashboardSlacksignal alertsCONSUME · EXECUTIVE SURFACEF3i · BESPOKE { }Sentiment Command CenterCustom report layer reading the trend aggregates.Monthly brand-health report · ad hoc trend alerts for the CMO.

System. Apify and Playwright harvest forum and community discussion into Postgres. Claude Code classifies sentiment, clusters themes, and isolates emerging trend signals. Aggregates roll up into Snowflake and surface through a Vercel dashboard, with high-priority shifts pushed to Slack and consolidated in a bespoke command center for executive review.

Service
Claude core
Custom { }
Command center
Architecture 04 · Dynamic Content Response

Dynamic Content Response Engine

GoalTurn live market signals into content that earns AI and search visibility.OutcomePublished, on-brand answers tuned for AEO and GEO, on a continuous loop.

SIGNALSDATA LAYERGENERATEPUBLISHMEASURE + LEARNUpstream Signalssentiment · intel · trendsDrive + FS MCPbriefs + brand assetsSnowflake / Neonunified signal layerClaude CodeAEO/GEO content engine { }Notion / Mondayhuman reviewGitHubcontent repo + PRVercelNext.js publishCONSUME · EXECUTIVE SURFACEF3i · BESPOKE { }Content Command CenterCustom report layer tracking AEO/GEO visibility and traffic.Quarterly performance review · ad hoc lift reporting, loops back to learn.

System. Signals from the sentiment, competitive, and trend systems plus brand assets from Google Drive and the Filesystem MCP consolidate into a Snowflake or Neon layer. Claude Code generates AEO and GEO optimized content, routed to Notion or Monday for human review, versioned in GitHub, and published through Vercel. A bespoke command center tracks visibility and traffic, then feeds performance back into the data layer to sharpen the next cycle.AEO: Answer Engine Optimization · GEO: Generative Engine Optimization

Service
Claude core
Custom { }
Command center
Architecture 05 · The Unified Platform

One platform. Four systems. Three chairs.

GoalConsolidate the four intelligence systems into one governed surface.OutcomeEvery executive works from a single source of truth, not four disconnected tools.

FOUR SYSTEMSTHREE CHAIRSMEASURE + LEARN · feeds every systemCustomer Retentionchurn risk signalsCompetitive Intelmarket moves + rankingMarket Sentimentbrand + trend signalsDynamic ContentAEO + GEO outputTHE F3i INTELLIGENCE PLATFORMGoverned Data FabricSnowflake · Neon · PostgresClaude Codereasoning + analysis fabric { }F3i · BESPOKE { }F3i Command CenterUnified executive reporting, one source of truth.Monthly · quarterly · ad hoc · live command center.CMObrand + demandCROpipeline + retentionCPOportfolio + segment

System. The four intelligence systems share one governed data fabric and one Claude Code reasoning layer, surfaced through a single bespoke command center. Retention, competitive, sentiment, and content intelligence stop being point tools and become one platform that the CMO, CRO, and CPO consume from the same place. Performance loops back to sharpen every system.

Service
Claude core
Custom { }
Command center