How we collect data

High level

Mercidia connects to configured commerce, analytics, and advertising sources through their official APIs. Launch connections are read-oriented and request only the permissions needed to load source records. Pilot is separate: it is available only for explicitly approved rules and supported provider actions; unavailable or unsupported actions fail closed.

What we read

• Orders, refunds, products, and related revenue fields from Shopify.
• Events, sessions, and conversion reporting from GA4 when the service account has property access.
• Campaign and performance records from Meta Ads using the read-only ads_read permission.
• Google Ads remains an evaluation connector and is not part of the launch-ready source claim.

What Pilot can do

• Evaluate approved rules against available source data and record a proposed, pending, skipped, or completed result.
• Run a supported provider action only after approval, role, budget, and rollback checks pass.
• Keep unsupported or unavailable provider actions unapplied and visible for review.

Meta Ads currently uses ads_read and is read-only; it cannot pause or enable campaigns or change budgets or bids through this connection. Pilot behavior also depends on provider support and account authorization.

Technical details — how data is retrieved

Key points below map directly to the codebase so you can follow the exact technical process.

1. API fetching layer: The frontend uses fetchJson in frontend/src/lib/api.ts to call backend endpoints. On the server it calls the real API base URL (via getApiBaseUrl()) and attaches an API key when available. In the browser it proxies calls to /api/proxy.
2. Endpoints and unavailable states: Dashboard data is loaded by loadDashboardData() from workspace-scoped /recommendations and /connectors endpoints. Network or connector failures produce empty or unavailable states; they are not replaced with fabricated records.
3. Connector syncs: The launch paths include Shopify, Meta Ads, and GA4, with diagnostics and sync endpoints invoked through server-backed fetchJson wrappers. Google Ads route coverage remains evaluation-only until hosted lifecycle validation is complete.
4. Automation actions: The automation system reads persisted rules, campaign metrics, and action records from the backend. A run records the provider result when one exists, approval state, daily-cap decision, and audit event; unavailable or unsupported platform data does not become an applied action.

Technical details — how metrics are derived

Metric calculations are implemented server-side helper functions that the frontend consumes. The key functions are in frontend/src/lib/dashboard.ts and frontend/src/lib/businessConfigMetrics.ts. Highlights:

• Recommendation-based dashboard metrics: calculateDashboardMetrics(recommendations) reduces recommendation records to totals (cost, revenue, impact). It builds channel-level aggregates and derives:
  • totalSpend — sum of recommendation.metrics.cost
  • totalRevenue — sum of recommendation.metrics.revenue
  • overallRoi — (revenue - spend) / spend
  • marketingEfficiency — revenue / spend
  • potentialSavings — totals.impact (sum of recommended savings)
• Paid media blend: Available campaign sources are aggregated by spend, revenue, impressions, clicks, and conversions, then used to compute derived metrics such as roi, ctr, and cpa. Email remains preview-only rather than a live sync source, and Google Ads values remain evaluation data until hosted validation is complete.
• Business-config derived metrics: Independent of connectors, deriveBusinessMetrics(config) (in businessConfigMetrics.ts) uses the business setup inputs like currentMonthlyRevenue and totalMarketingBudgetMonthly to compute:
  • netProfit — revenue - total monthly costs (including marketing and fixed/operational costs)
  • profitPerDollar — netProfit / marketingSpend
  • marketingEfficiency — revenue / marketingSpend * 100
  • Break-even orders, gross/net margin per order, and spend share
• Attribution: Frontend fetches an attribution summary via fetchAttributionSummary() in api.ts. The attribution breakdown compares first-touch, last-touch, and linear allocations over recorded journeys, with the lookback window and evidence limits shown beside the result. On failure, the frontend uses an empty summary state.

Privacy & security

We store the data needed to provide reconciliation, source health, decision records, and approved workflows. Connector credentials are encrypted at rest and accessed server-side through scoped connector paths. Read our full Privacy Policy for details.