The operational backbone for a membership network.
Forum for Naturals is a private membership network for senior marketing and sustainability leaders in the naturals industry. I built and own the systems that run it — and I'm now leading the platform rebuild.
The problem
A specialized network runs on a lot of recurring, manual operations: agencies to review, members to match, a monthly spotlight to assemble, intake to chase, and data to keep clean across multiple systems.
Done by hand, that work doesn't just fail to scale — it drifts. A review attaches to the wrong agency, a pairing repeats, a spotlight ships late. For a network whose whole value is curation, the credibility lives in exactly those details.
The approach
Treat the whole thing as one system, not a pile of tasks. Airtable holds the data; idempotent automations move and match it; an AI layer drafts and summarizes; a member app presents the result.
Every flow is built to be safe to re-run — normalized match keys, status flags, and explicit error capture instead of blind writes. The matching is deterministic; the AI only ever handles the writing.
Five systems, one backbone.
Three Airtable bases
The operational core: members, brands, agencies, reviews, resources, courses, events, and the matching system — modeled with normalized keys and rollups so the data stays filterable, not just full.
Verified-review pipeline
Reviews flow from intake into canonical records, matched on normalized email and name keys with captured record IDs. A failed match is flagged with an error note, never written silently.
Self-assembling Spotlight
The monthly Agency Spotlight assembles itself: an agency submits a form, the submission is logged and tracked, and the page record builds its copy and assets — ready to render.
Member-matching system
Deterministic no-repeat pairings each cycle, with the outreach drafted automatically — the same primitive that became Cycle Conductor. Rules for the matching, AI for the voice.
Custom AI skill library
Six reusable Claude skills — narrative reporting, comparative analysis, wireframe-to-deck, tone-controlled outreach, multi-format orchestration, and the catalog that indexes them.
Dashboards & tracking
Looker Studio reporting on engagement and page views, with Claude as the external AI layer for formula help and turning screenshots into stakeholder presentations.
3 Airtable bases
skills built
per cycle
audited
Now leading the platform rebuild.
The next phase sunsets the off-the-shelf member app for a more controlled stack. The standing principle: document every workflow tool-agnostically first, then build against the written specs — so the tools serve the design, not the other way round.
Role-based visibility is modeled per field rather than per page, which is what makes access auditable instead of a tangle of show-hide rules. The directory and member systems are spec'd from teardowns of how the most mature directories and community platforms structure their schema, filtering, and permissions.
What people ask about this work.
What did you build for Forum for Naturals?
The full operational backbone: three Airtable bases, an agency directory with a verified-review pipeline, a self-assembling monthly Agency Spotlight, a deterministic member-matching system, and a custom library of reusable AI skills. I'm now leading the rebuild onto a Clerk, Airtable, and Sanity stack.
How is the review pipeline kept trustworthy?
Reviews flow from intake into Airtable, where normalized match keys link each one to the correct agency, contact, and brand records. Every automation is idempotent — it captures record IDs, tracks a status, and writes an explicit error note when a match fails, so failures are visible instead of silent.
Why rebuild the platform?
To move from an off-the-shelf member app to a more controlled stack — Clerk for auth, Airtable as the system of record, Sanity for content. The principle driving it is to document every workflow tool-agnostically first, then build against the specs, with role-based visibility modeled per field rather than per page.
Have an operation that should run itself?
This is the shape of system I build — database, automation, AI, and the member-facing result, owned as one loop end to end. If you're running an operation that deserves the same, tell me where it strains first.
Start a project ↗