SGA Gen4 Social MCP — Product Roadmap

From integration to per-practice operating surface across the organic channel stack.

Owner: Dakota Milner Stack: sga-intranet (BE Fastify + Drizzle + Postgres; FE React + Vite) Status: Phase 1 + 2 shipped Date: 2026-05-18

TL;DR. We just unlocked 30 Gen4 (PromoRepublic) tools from the original 6 by re-OAuthing with the right scopes. The warehouse already lands 83 practice locations and the network-wide rollups every night. This roadmap turns that into a per-practice and per-region operating surface across five phases over about seven weeks.

Per-practice data is the rate-limiting step. Every downstream phase depends on Phase 3 closing the practice ↔ Gen4 location bridge.

Current state — what's already live

Gen4 tools available

Locations in warehouse

260

SGA practices total

32%

Gen4 coverage

Phase 1 + 2 · Foundation

ELT warehouse, persistent OAuth, network rollup

Shipped

marketing.gen4_snapshots · raw JSON-API payloads, one row per tool call, JSONB data column
marketing.gen4_sync_runs · audit log, status, errors
marketing.gen4_locations · 83 locations cataloged with raw attributes
marketing.gen4_oauth_state · rotated refresh tokens persist across service restarts
/admin/gen4 · sync trigger, runs table, snapshot drawer, live tool catalog
/marketing/gen4 · network-wide social audience, content, top posts
scripts/gen4-oauth.py · self-contained DCR + PKCE + RFC 8707 helper

Phase 3 — Practice ↔ Gen4 bridge Now

1 week · the unblock

Every Gen4 datapoint maps to an SGA practice

Critical path

Goal. Close the location bridge so every single-location tool call lands keyed to a known SGA practice. This is the foundation for every downstream phase.

Work

Auto-matcher · fuzzy match Gen4 name + city + state against shared.practices with a confidence score. Wire as a one-shot endpoint POST /api/v1/gen4/locations/auto-match.
Admin map page · /admin/gen4/locations. Three tabs: matched (≥80 confidence), needs review (40–79), unmatched (<40). Manual confirm or reassign per row.
Per-practice sync runner · for each mapped location, call six single-location tools (social performance, social content, GBP, local visibility, reviews, reputation). Stamp location_id on each snapshot row.
Cron · nightly at 4am UTC. 83 × 6 = ~500 calls per night. Confirm rate limits first.

Exit criteria

Every mapped practice has snapshots refreshed nightly
Admin can map a new location in under 30 seconds
GET /api/v1/gen4/snapshots?practice_id=X returns rows

Phase 4 — Practice-level surface Next

1.5 weeks · the headline outcome

Per-practice Organic Channels view

Champions + HQ ops

Goal. Click any practice → see its organic performance at a glance. Audience, posts, GBP, reviews, reputation. Plain answers to "how are we doing?"

Page

/marketing/practices/:practiceId — add an "Organic Channels" section between existing KPI tiles and the campaign breakdown.

Cards

Card	Tool source	What you see
Social Performance	`get_social_performance_single_location`	Audience by network with sparkline + delta. Engagement total + breakdown.
Content Performance	`get_social_content_performance_single_location`	Posts published + top performing post with thumbnail + publishing rhythm.
Google Business Profile	`get_gbp_performance_single_location`	Views, calls, direction requests, photo views.
Local Visibility	`get_local_visibility_single_location`	Average ranking, ranked keywords count, search engines visible on.
Review Performance	`get_review_performance_single_location`	Average rating, review velocity, recent reviews list.
Reputation	`get_reputation_insights_single_location`	Sentiment trend, top topics, alerts.

Empty state

Practice without Gen4 mapping shows a "This practice doesn't have Gen4 coverage. Request setup →" CTA that links to the admin map page.

Exit criteria

Clicking any of the 83 mapped practices shows real per-channel data
The 177 uncovered practices show the empty state cleanly
A champion can self-serve the answer to "how was our IG last week" in under 10 seconds

Phase 5 — Network intelligence Next

1 week · pattern spotting

Cross-network rollups + outlier detection

HQ marketing leadership

Goal. Marketing leadership can answer "which practices need a call this week" in under 2 minutes.

Page

Upgrade existing /marketing/gen4 from "stats wall" to "decision surface."

Sections

Top performers leaderboard · top 10 by audience growth, posts shipped, engagement, review velocity
Practices needing attention · bottom 10 + flag thresholds (rating < 4.0, posts/week < 2, audience flat or declining)
Network avg vs practice deviation · table with z-score columns
Regional rollups · by state, by acquisition cohort (Gen4 vs legacy SGA vs pilot)
Reputation heatmap · sentiment by region, by service line

Phase 6 — Operations tools Later

1.5 weeks · Korrine's command center

Central inbox + triage queues

Korrine + social ops team

Goal. Korrine stops opening individual Gen4 dashboards per practice.

Page

New /tasks-app/social-ops — lives next to standup + sprints in the tasks shell.

Sections

Section	Tool source	Action
Centralized inbox	`get_inbox_conversations`	All DMs across all locations, sort by unread
Failed posts triage	`get_failed_posts_all_locations`	Queue per practice with error + retry/dismiss
Low publishing alerts	`get_low_publishing_locations`	List of practices below threshold posts/week
Scheduled posts calendar	`get_scheduled_posts_all_locations`	Cross-practice month view, filter by network

Phase 7 — AI insights + automation Later

2 weeks · data becomes decisions

Alerts, recommendations, weekly digests

Network-wide quality of life

Auto-alerts · configurable thresholds. Slack/email when metrics dip (negative sentiment rising, review velocity drops, audience flat for N days)
Cross-practice content recommendations · "Wyoming Dental's post about veneer financing got 500 engagements. Try at Woodyard Periodontics."
Sentiment trend detection · flag practices where negative topics are emerging
Weekly digest per champion · auto-generated email with practice scorecard + recommended actions

Sequencing + timing

Phase	Length	Cumulative	Depends on
3 · Practice bridge	1 week	Week 1	Phase 1+2 shipped
4 · Practice surface	1.5 weeks	Week 2.5	Phase 3
5 · Network intelligence	1 week	Week 3.5	Phase 4 (uses per-practice data)
6 · Ops tools	1.5 weeks	Week 5	Phase 3 (location mapping)
7 · AI insights	2 weeks	Week 7	Phases 3 + 4 + 5

Phases 5 and 6 can run in parallel once Phase 4 is shipped, shaving ~1 week if needed.

Open questions + risks

Coverage gap. 177 of 260 practices have no Gen4 footprint. Either backfill mappings via Gen4 onboarding ops (Korrine), or accept "we only know about the 83." Need a call before Phase 4 ships, since the empty-state UX depends on the answer.

Rate limits. Nightly per-practice sync = ~500 calls. Gen4 docs don't publish a limit. Phase 3 should include a small canary run (~10 practices) before unleashing all 83.

Token rotation guard. Phase 2 fixed the rotation-loss bug, but if Gen4 ever revokes a client_id, the system is dead until a human re-OAuths. Long-term we should add a healthcheck endpoint that alerts on invalid_grant.

Reputation/review tools. Returned Access Denied at Phase 1 scopes. The new scope set should fix this; Phase 3 will confirm. If still blocked, escalate to Gen4 support or fall back to direct Google Business Profile API for reviews.

Warehouse migration. If we eventually move Coupler.io off the bench and use it as the analytics warehouse, this whole snapshot table strategy changes. Decide before Phase 7 (where insights logic gets heavy).

Out of scope

Posting or scheduling from inside SGA Intranet. Stay in Gen4 for now.
AI-generated content suggestions. Lives in the Content Engine, not the analytics surface.
Patient sentiment NLP beyond what Gen4 returns natively.
Migration of all 177 uncovered practices to Gen4. That's a Korrine business decision, not engineering.

Recommended starting point

Begin Phase 3 immediately. It's the rate-limiting step for every downstream phase, and the work is mechanical (matcher + admin page + cron). Ship a canary sync before committing the full nightly cron.