Part 01
Ad Library Audit
Tracks competitor advertising themes, formats, messaging volume, estimated spend intensity, timing, and positioning gaps across a defined monthly reporting cycle.
Build 02 / Controlled Intelligence
Governed marketing intelligence for decision-ready teams.
A controlled intelligence system for campaign performance, competitor monitoring, web visibility, ingestion governance, and executive-ready AI summaries.
Dashboard Evidence
The intelligence layer needs to be readable before it becomes automated.
Marketing Intelligence AI is being shaped as a decision-support surface first: visible data quality, four intelligence views, executive summaries, and controlled evidence before automation scales.
Click to enlarge
Illustrative dashboard panel for the governed Marketing Intelligence AI build.
Intelligence Model
Four intelligence views. One governed data model.
Part 02
Organic Social Intelligence
Benchmarks public organic activity, content themes, format choices, engagement signals, and competitor posting patterns across the tracked market set.
Part 03
Website and Search Intelligence
Connects website traffic, channel visibility, search behaviour, referral sources, and content themes into a more useful competitive visibility layer.
Part 04
Campaign Performance Intelligence
Brings paid campaign metrics, spend signals, impressions, engagement, clicks, and channel-level performance into one decision-support environment.
Decision System Controls
The dashboard is the surface. The control layer is the product.
Marketing intelligence only becomes useful when the source, freshness, quality, review state, and publication status are visible. The system is designed to protect decisions from untrusted inputs before AI summaries or executive readouts are generated.
Source control
Known inputs
Every source is registered, labelled, and reviewed before it becomes part of the decision layer.
Quality gates
Validated data
Schema checks, row counts, required fields, and exception states are visible before publishing.
Review state
Human approval
Automation can collect and prepare data, but trusted publishing remains review gated.
Fallback path
Last-known-good
If ingestion fails, the dashboard can fall back to the latest approved dataset.
Architecture Options
Three viable routes. One selected delivery path.
This page shows the architecture decision behind the build: three viable delivery routes, the trade-offs between speed and enterprise control, and the selected path for a fast but governed portfolio implementation.
Option A
SelectedVercel-led build route
Best fit for developer velocity, preview deployments, serverless API routes, strong local development, and staged review before production release.
Vercel hosting + API routes
Cloudflare DNS / CDN
Supabase data layer
Apify ingestion pilot
Option B
Cloudflare-native route
Best fit when procurement simplicity, existing Cloudflare infrastructure, low V1 cost, and IT-owned access control are the highest priorities.
Cloudflare Pages
Cloudflare Workers
Supabase storage
Access policy gate
Option C
Enterprise cloud route
Best fit when centralised IAM, managed secrets, enterprise logging, monitoring, BigQuery analytics, and long-term data-platform scale are the priority.
Cloud Run
BigQuery
Secret Manager
Cloud Logging
Selected Build Route
Option A: developer velocity with a governed data spine.
Developer velocity
Preview before production
Branch and preview deployments support stakeholder review, ingestion testing, and safer iteration before anything touches the live dashboard.
Serverless control
API routes own the logic
CSV upload, validation, Apify webhooks, data cleaning, dashboard queries, and review gate transitions all run server-side.
Portable data model
No lock-in at the schema level
Raw tables, cleaned tables, source registry, ingestion logs, metrics config, and dimension tables remain portable across the options.
Ingestion Governance
Manual first. Automated second. AI after trust.
V1
Manual CSV upload
Controlled monthly upload through an admin panel. Validates column names, data types, required fields, row counts, and source records before dashboard refresh.
V1.5
Apify pilot with human review gate
One actor, one source, one dashboard. Every run enters review before anything is published. CSV fallback remains available if the actor fails or data is rejected.
V2
Scale across approved sources
Additional actors and data sources activate only after source registry approval, rate-limit configuration, schema validation, and review gate confidence.
V3
AI insight summaries
AI-generated summaries are layered on top of approved data only. The AI explains patterns, drafts briefings, and surfaces gaps, but humans own decisions.
V4
Execution layer
Future CRM, campaign, or marketing automation integrations are deferred until the intelligence layer is stable, reviewed, and trusted.
Data Model
Raw data preserved. Cleaned data published. Governance logged.
Raw layer
Immutable source archive
CSV files and Apify JSON are stored as received. Raw records are not overwritten, which protects auditability and enables reprocessing.
Clean layer
Dashboard-ready tables
Cleaned tables standardise companies, themes, formats, platforms, dates, derived metrics, and display-ready fields.
Control layer
Source registry and logs
Approved sources, review status, ingestion method, operator action, run status, row count, and publication state are logged.
Data Quality Framework
Confidence is visible instead of assumed.
Raw
Collected or uploaded data before review. Hidden from the main dashboard.
Validated
Schema passed, but data still needs review or comparison before being trusted.
Approved
Human-reviewed and published to the dashboard as the current trusted dataset.
Trusted
Repeatedly successful sources can move toward lighter review with exception alerts.
AI Layer
AI summaries sit on approved data, not raw noise.
The AI layer is deliberately deferred until the ingestion, review, and cleaned-table model is stable. Once the data spine is trustworthy, the system can produce summaries, gap analysis, briefing drafts, and campaign recommendations.
Future AI outputs
> Executive competitive summary
> Monthly gap analysis
> Campaign response brief
> Data quality and stale-data warnings
Scope Control
Scope control protects the architecture.
Deferred deliberately
> No multi-actor automation before the pilot proves quality
> No AI agent before clean and approved data exists
> No CRM or marketing automation integration in early phases
> No real-time refresh without a separate approval decision
Always available
> CSV fallback at every stage
> Raw archive before transformation
> Human review gate for pilot automation
> Last-known-good dashboard data if ingestion fails
Signal
Discuss Marketing Intelligence AI
Use the contact form for demos, architecture conversations, speaking opportunities, or collaboration around governed marketing intelligence workflows.