# Operations Control Tower Architecture

Companion to **Building an AI Operations Control Tower for eCommerce**.

Primary question: **What requires human attention now?**

This is **not** the 8 a.m. Slack brief ([`operations-daily-brief-schema.md`](operations-daily-brief-schema.md)). It is the **always-on** prioritization layer over signals.

```text
Business Systems → Signals → AI Operations Layer → Detect / Investigate / Prioritize → Operations Control Tower
```

## Surfaces

| Domain | Signal feed | Default |
| ------ | ----------- | ------- |
| Sales | `salesByPeriod` | Watch |
| Orders | Exception queue | Investigate |
| Inventory | ATP vs ads | Page if advertised stockout |
| Customers | Duplicate / credit | Quality / B2B AM |
| Support | Tag spike | Brief, not invented UX |
| Returns | Rate vs baseline | Refund / quality |
| Marketing | Spend vs ATP | Recommend pause |
| Fulfillment | SLA | Exception agent |

## Prioritize

Rank by **blast radius**, not model confidence. Cap the tower’s “now” list (example: 7). Overflow is watch.

Writes stay on domain agents with Cedar. The tower **recommends**; it does not purchase.

## Related

- Signals: [`store-monitoring-signal-catalog.md`](store-monitoring-signal-catalog.md)
- Daily brief: [`operations-daily-brief-schema.md`](operations-daily-brief-schema.md)
- Multi-agent split: [`multi-agent-when-to-split.md`](multi-agent-when-to-split.md)