Summit Chase Multi-Client Analytics Portal Architecture
Designed the technical and functional architecture for Summit Chase’s future analytics platform: a multi-client portal with admin and client login, BigQuery-backed reporting, standardized SQL models, campaign classification, conversion mapping, and reusable dashboards for paid media and CRM data. The architecture balanced near-term dashboard delivery with a longer-term Next.js/Supabase portal, while accounting for platform constraints such as Google Cloud service account policies, inconsistent campaign-status fields, and the need to validate every filter against actual source inventories. The design also introduced AI-native execution patterns, with ChatGPT acting as product and MarTech director and Claude Code planned as the implementation environment.
Designed a scalable reporting architecture that converted one-off agency dashboards into a governed multi-client data platform.
Role: Product and technical architect
The Problem
Summit Chase had no dedicated client reporting portal and relied on Google Sheets, Looker Studio, and Supermetrics-based workflows. The agency needed a repeatable reporting foundation for many clients instead of one-off dashboards.
The platform needed to ingest multi-platform paid media and CRM data into BigQuery, normalize client-specific metrics and filters, and expose dashboards through both internal admin and client-facing views. The design also had to work around Google Cloud constraints such as disabled service account key creation.
The architecture had to support multiple ad platforms, client-specific branding, standard dashboard templates, cross-platform campaign classification, role-based access, and later AI-driven setup/automation without overbuilding the first milestone.
Approach & Architecture
Designed a full analytics platform with a client portal, admin portal, BigQuery warehouse, standardized SQL layer, and reusable dashboard templates. The proposed system used Supermetrics or native Google connectors for ingestion, BigQuery for normalization, Supabase for authentication, and Next.js for the portal. The admin workflow included client creation, platform account association, branding configuration, metric toggles, and conversion mapping. The system was framed as a long-term replacement or augmentation for Looker Studio dashboards.
Ad platform and CRM data would be ingested into BigQuery through Supermetrics, Google-native transfers, or direct APIs. A normalized SQL/modeling layer would create common cross-platform metrics and platform-specific tables. A Next.js client portal with Supabase authentication would read from BigQuery-backed APIs and render standardized dashboard pages, while an admin interface would manage clients, accounts, branding, conversion mapping, and dashboard permissions.
Key Decisions & Trade-offs
Hardest Part
Designing a platform flexible enough for many clients and platforms while keeping the initial implementation realistic and grounded in the available source fields.
Technical Detail
Universal filters included Date Range, Comparison, Platform, Account, Campaign, Campaign Status, and Summit Chase-built Campaign Classification. Platform-specific pages were planned for Google Ads, Meta, LinkedIn, Microsoft, TikTok, Reddit, OpenAI Ads, GA4, GTM, HubSpot, and Salesforce.
Proposed stack included Next.js, Supabase, BigQuery, Supermetrics/native connectors, and possibly MCP connectors for Supermetrics and BigQuery.
Planned validation against paid-media-bigquery-fields.xlsx, Google Ads API v23, BigQuery Data Transfer schemas, and platform field inventories.
ChatGPT acted as product, UX, MarTech, and design director while Claude Code was planned as the build environment for implementation. The architecture also considered MCP connectors and AI-assisted scraping of client brand identity for portal theming.
Measured Results
The conversation produced a detailed target architecture and implementation direction for a multi-client agency reporting platform. The system was not confirmed as shipped, but it became a structured product and technical plan for future development.
| Metric | Value | Before | Source |
|---|---|---|---|
| Windsor AI plan cost considered This was a vendor plan considered, not a measured result. | $249/month | n/a | User-stated comparison in chat |
| Windsor AI account capacity considered This was a product-plan constraint, not a measured result. | up to 200 accounts and 10 input sources | n/a | User-stated vendor plan comparison |
Every figure above was recorded during the work itself. Where no number was measured, none is claimed.