Cloudflare Workers Roadmap

Roadmap for expanding Cloudflare Workers across the Arlem stack. Workers run at the edge (~50ms latency vs 500ms+ for Vercel/Azure), cost effectively nothing at our scale, and bypass ad blockers when deployed on first-party subdomains.

Current State

Tier 1: Quick Wins (Done / Ready)

Edge Analytics (Done)

Worker: edge-analytics at edge.arlem.com.au

• Replaces Vercel /api/ping proxy for analytics event ingestion
• Geo enrichment from request.cf (country, region, city, lat/lon, timezone)
• Bot filtering before events reach Azure
• KV segment lookup returns visitor segment code in ~5ms
• ctx.waitUntil() forwards enriched events to Azure asynchronously
• Client-side fallback to /api/ping if Worker is unreachable

Impact: Eliminates Vercel function invocation per page view, reduces latency from ~500ms to ~50ms, bypasses most ad blockers.

KV Segment Cache (Done)

• Hourly scheduledBigQuerySync writes segment codes to VISITOR_SEGMENTS KV namespace
• 2-hour TTL ensures stale data self-cleans if sync breaks
• Bulk writes via Cloudflare API (up to 10,000 per batch)

D1 Feature Flags + A/B Test Config

Effort: Low | Impact: Medium

Store feature flags and A/B test configurations in D1 (Cloudflare’s SQLite-at-edge database). The edge analytics Worker reads them on every request and returns active experiments alongside the segment code.

• No external service dependency (LaunchDarkly, Statsig, etc.)
• Configuration changes propagate globally in seconds
• A/B test assignment happens at the edge, not in client JS
• Reference: Philip Walton’s A/B testing at the edge pattern

Schema:

CREATE TABLE feature_flags (
  key TEXT PRIMARY KEY,
  enabled BOOLEAN,
  segments TEXT, -- JSON array of segment codes this applies to
  percentage REAL -- rollout percentage (0-100)
);

CREATE TABLE experiments (
  id TEXT PRIMARY KEY,
  variants TEXT, -- JSON array of variant configs
  allocation TEXT, -- JSON allocation weights
  active BOOLEAN
);

Tier 2: Medium Effort, High Value

Edge Personalisation Manifest

Effort: Medium | Impact: High

Return a personalisation manifest from the edge worker alongside the segment code. The manifest tells the client what to show without additional API calls.

{
  "segment": 3,
  "personalisation": {
    "hero_cta": "See what others are saying",
    "show_social_proof": true,
    "discount_code": null,
    "exit_intent": true
  }
}

• Manifest keyed by segment + experiment variants
• Stored in KV or D1 alongside segment data
• Client reads manifest once on page load, applies all personalisation client-side
• Eliminates multiple round-trips for personalisation decisions

Smart Image Serving (CDN Worker Enhancement)

Effort: Medium | Impact: Medium

Enhance the existing cdn-api Worker to handle smart image transformations:

• WebP/AVIF format negotiation based on Accept header
• Responsive image serving based on CF-Device-Type header
• Art direction (crop/resize) at the edge using Cloudflare Images
• Cache transformed variants in R2 or Workers Cache API
• Reference: Cloudflare’s R2 image reference architecture

Cloudflare Queues for Event Buffering (Done)

Edge analytics Worker produces to a Cloudflare Queue. A consumer Worker batches events and forwards to Azure in bulk. Service Bus remains for Telegram, financial, and other non-analytics queues.

Tier 3: Ambitious / Future

Durable Objects for Real-Time Visitor State

Effort: High | Impact: High

Use Durable Objects to maintain real-time visitor state at the edge:

• Each active visitor gets a Durable Object (created on first event, garbage collected after inactivity)
• Tracks session state: pages viewed, time on site, cart contents
• Can compute journey stage in real-time instead of hourly BigQuery batch
• Enables real-time triggers: “visitor has been on product page for 60s, show social proof”
• Reference: Similar to how Shopify Oxygen uses Workers for storefront personalisation

Workers AI + Vectorize for Product Search

Effort: High | Impact: Medium

• Embed product descriptions and customer reviews using Workers AI
• Store vectors in Vectorize
• Semantic search at the edge: “something soft for a modern bedroom”
• Could power a conversational product finder

Cloudflare Workflows for Checkout Orchestration

Effort: High | Impact: Medium

Use Workflows (durable execution on Workers) for multi-step checkout flows:

• Cart abandonment sequences triggered at the edge
• Inventory reservation with automatic release
• Payment processing orchestration with retry logic

Workers Analytics Engine for Real-Time Dashboards

Effort: Medium | Impact: Low-Medium

Write analytics events to Workers Analytics Engine (columnar storage at edge) alongside Azure forwarding:

• Real-time query API for dashboards
• No BigQuery cost for operational queries
• 90-day retention, SQL-like query interface
• Good for “what’s happening right now” views

What Stays on Azure

Not everything should move to Workers. These stay on Azure Functions:

Principle: Workers handle the hot path (every page view). Azure handles the warm path (hourly syncs, AI processing, financial ops).

Deployment Notes

KV Namespace Setup

# Create production namespace
wrangler kv namespace create VISITOR_SEGMENTS

# Create preview namespace for local dev
wrangler kv namespace create VISITOR_SEGMENTS --preview

# Update wrangler.toml with returned IDs

Secrets

# Edge analytics Worker
cd workers/edge-analytics
wrangler secret put AZURE_FUNCTION_CODE

# Azure Functions (for KV sync)
# Set in Azure Portal > Function App > Configuration:
#   CLOUDFLARE_ACCOUNT_ID
#   CLOUDFLARE_KV_NAMESPACE_ID
#   CLOUDFLARE_API_TOKEN (needs Workers KV write permission)

DNS

Add a CNAME record for edge.arlem.com.au pointing to the Worker. Cloudflare handles this automatically when using custom domains in wrangler.toml, but the domain must be on a Cloudflare-managed zone.

Monitoring

• Workers analytics dashboard shows request count, latency percentiles, error rates
• Azure Function logs show forwarded events arriving
• KV metrics show read/write patterns and cache hit rates