Roadmap — NCR Batch to Real-Time ALX
NCR Counterpoint is batch-first. MCP requires real-time. The roadmap routes around NCR’s architecture in Phase 1, then progressively replaces the batch dependency.
Phase 1 — Wrap & Ship (2026)
Goal: Get ALX live. Prove the model.
Canary builds the MCP server on top of Counterpoint’s existing REST API. Agent queries hit a Redis cache (refreshed every 60–300 seconds) — near-real-time at approximately 1–5 seconds. Payment via existing Worldpay/NCR integration. “Put it on my account” settles via Counterpoint customer account + stored payment method.
Honest limitation: Inventory is near-real-time, not live. Acceptable at this stage — better than any competitor has today.
Stack: MCP server (Python), Redis cache, Counterpoint REST API, Worldpay.
Phase 2 — Event Triggers (2027)
Goal: Move from polling to push.
Webhook layer on top of Counterpoint — inventory update events trigger on sale, receive, and adjustment. Cache invalidated by events, not just timers. Associate availability connected via store scheduling integration.
Result: Inventory accuracy from 60–300 seconds to near-instant on transaction events.
Phase 3 — Streaming Pipeline (2027–2028)
Goal: True real-time layer independent of NCR’s batch cycle.
Event streaming (Kafka or Kinesis) as a parallel data pipe. NCR transaction events tap directly into the stream. Store sensor data (RFID, computer vision) feeds in. ALX reads from the stream, not from the NCR API.
Result: ALX has live store state. NCR’s batch cycle becomes irrelevant to the customer agent experience.
Phase 4 — Autonomous ALX (2028–2030)
Goal: ALX learns the store and anticipates demand.
ALX develops a store-level demand model from streaming history. Predictive inventory signals push to customer agents before they ask. ALX initiates conversations: “Based on your last visit, this item is running low — want me to hold one?” Cross-store network: customer agent queries any enrolled store via a single MCP endpoint.
Result: Canary’s back office hub becomes a demand prediction network. ALX is not reactive — it is proactive and personalised.
Technology stack by phase
| Phase | Key technology |
|---|---|
| 1 | MCP server, Redis cache, Counterpoint REST API, Worldpay |
| 2 | Webhooks, event bus, scheduling system integration |
| 3 | Kafka/Kinesis, RFID/CV sensor tap, streaming MCP |
| 4 | ML demand model, proactive agent messaging, multi-store ALX network |