Lead qualification and pipeline
crm.lead_qualification
crm.pipeline_progress
crm.customer_retention
AI improves lead scoring, outreach timing, and churn prediction. Same actor model. Same evidence trail.
The Loop Engine
Enterprise AI fails in chaos. It succeeds in structure.
The Loop Engine turns operational processes into bounded state machines - finite, observable, and improvable. Every human action and AI decision runs inside a loop. Every loop closes with evidence.
Finite states
No freeform AI improvisation
Complete trace
Every transition recorded
Apache 2.0
Core primitives - open source
Outcome-aligned
Billing tracks loop completions
Systems of record don't learn. Control systems do.
Traditional enterprise software
- Stores data. Does not learn from it.
- AI guesses what to do next in unbounded environments.
- Outcomes are logged. Decisions are not.
- No actor attribution - who did what is reconstructed later, if at all.
- Compliance is a separate audit step.
The Loop Engine
- ✓Every completed loop generates training data automatically.
- ✓AI reasons over finite states - what is allowed, what comes next.
- ✓Decision evidence is recorded at transition time, not reconstructed.
- ✓Every actor - human or AI - leaves the same evidence trail.
- ✓Compliance is a byproduct of operation, not a separate workflow.
The Loop Engine is not a workflow tool. It is a control layer.
EVENT STREAM→
SIGNAL ENGINE→
LOOP ENGINE→
DECISION (human or AI)→
ACTION→
OUTCOME→
LEARNING
The Loop Engine is the control layer. It does not replace your ERP or CRM. It makes them improvable.
LOOP TYPE 01
Self-Learning Loops
Each completed loop produces a structured dataset: transition history, decision evidence, actor attribution, and outcome metrics. That dataset becomes the training signal for the next iteration - without any data pipeline or ETL step.
Loop: scm.replenishment
LOW_STOCK_SIGNAL
↓ [system:demand-signal]
REPLENISHMENT_RECOMMENDED
↓ [system:replenishment-engine]
PO_CREATED
↓ [drew.kim@lumebonde.co - human approval]
PO_RECEIVED
↓ [system:fulfillment]
STOCK_STABILIZED - CLOSED
Outcome logged:
demand_forecast: 200 units
actual_demand: 260 units ← learning signal
supplier_lead_time: 12 days
actual_lead_time: 16 days ← learning signal
stockout_occurred: true
The next loop adjusts supplier lead time estimates automatically.
Learning signal table
| Attribute | Predicted | Actual | Δ |
|---|---|---|---|
| Demand | 200 | 260 | +60 |
| Lead time | 12d | 16d | +4d |
| Stockout risk | Low | High | Miss |
Loop: scm.procurement
Aggregate: PO-2026-0012
Transitions:
OPEN → PO_CONFIRMED 12s [drew.kim@lumebonde.co]
PO_CONFIRMED → RECEIPT_SCHEDULED 4m [system:replenishment]
RECEIPT_SCHEDULED→ RECEIVED 3d [warehouse:dc-east]
RECEIVED → CLOSED 2m [system:fulfillment]
Correlation: corr_abc123
Every actor - human or AI - leaves the same evidence trail.
LOOP TYPE 02
Fast Feedback Loops
The faster a loop closes, the faster the system improves. Every state transition emits a structured event - actor, timestamp, evidence, outcome - creating a continuous feedback signal that AI can act on immediately.
scm.procurement - Last 30 days
Avg close time: 3.2 days
Exception rate: 4.1%
AI-acted transitions: 67%
Human-approved: 33%
OPEN
IN_PROGRESS
CLOSED
ERROR
AI doesn't wait for a nightly batch job. It sees the result of every action as it happens.
LOOP TYPE 03
Signal Loops
Raw events are noise. Signal loops extract structured, high-confidence alerts from the event stream and use them to trigger downstream loops - without manual monitoring.
Signal pipeline:
Raw events:
sale_recorded SKU: BRS-001 qty: 847 location: DC-East
shipment_delayed carrier: FedEx ETA: +4 days
stock_movement BRS-001 DC-East on_hand: 12 units
↓ feature extraction
Signal produced:
type: DEMAND_SPIKE
sku: BRS-001
location: DC-East
confidence: 0.84
on_hand: 12 units (below reorder point: 50)
↓ loop triggered
scm.replenishment: OPEN
aggregate: REPLEN-2026-0041
Signal loops also learn - tracking their own accuracy and adjusting confidence thresholds over time.
Live Signals
●DEMAND_SPIKEBRS-001conf: 0.842 min ago
●STOCKOUT_RISKBRS-003conf: 0.9114 min ago
●THRESHOLD_CLEARCRM-110conf: 0.771 hr ago
Signals trigger loops. Loops close with evidence.
A control layer for any enterprise system
The Loop Engine primitives - state machine runtime, event trace model, actor model, signal definition framework - are open source under Apache 2.0. The same pattern that makes SCM replenishment AI-safe applies to any bounded enterprise workflow.
Procurement, fulfillment, collections
erp.procurement
erp.order_fulfillment
erp.cash_collection
Supplier selection, payment timing, and logistics routing - each loop closes with measurable outcomes.
Ticket resolution and escalation
support.ticket_resolution
support.escalation
Routing efficiency, escalation triggers, and knowledge base gaps - all visible in transition history.
The primitives are open. The SCM implementation is where Better Data runs deepest.
Why AI fails without loops
| AI Weakness | Without the Loop Engine | With the Loop Engine |
|---|---|---|
| Unbounded decisions | AI guesses what to do next | Finite state machine - only valid transitions allowed |
| No training data | Operational data sits in a database, unused | Every closed loop produces a structured learning signal |
| No accountability | Who decided what is reconstructed later | Actor model records every human and AI action at transition time |
| Unclear outcomes | Success is measured manually, inconsistently | Loop completion is a measurable, billable outcome |
| Compliance burden | Separate audit workflow required | Audit trail is a byproduct of operation |
Every actor leaves the same evidence
The Loop Engine does not distinguish between human operators and AI agents at the infrastructure level. Both are actors. Both produce transition evidence. Both are auditable.
Loop ID: scm.procurement
Aggregate: PO-2026-0012
Current state: RECEIPT_SCHEDULED
Transition history:
OPEN → PO_CONFIRMED
actor: drew.kim@lumebonde.co
actor_type: human
timestamp: 2026-02-14T09:14:22Z
evidence: { approved: true, method: "3-way-match" }
PO_CONFIRMED → RECEIPT_SCHEDULED
actor: system:replenishment
actor_type: automation
timestamp: 2026-02-14T09:18:45Z
evidence: { warehouse: "DC-East", eta: "2026-02-17" }
RECEIPT_SCHEDULED → [pending]
actor: agent:commerce-gateway/claude
actor_type: ai-agent
recommended: RECEIVE
evidence: { shipment_scan: "confirmed", quantity_match: true }
status: awaiting human approval
Correlation: corr_abc123
Human, automation, AI agent - one evidence model.
AI agents connect to the Loop Engine via the Commerce Gateway (MCP). They can query loop state, surface exceptions, and recommend next actions within loop guardrails.
Explore Commerce GatewayRelated products
See it in action
Explore how loop-native operations improve throughput, resilience, and compliance.
Book a demo