[object Object]

A regional distributor receives 180 sales orders a week through email, EDI, and a customer portal. Three order entry clerks spend most of their day rekeying. The Wave 1 agentic sales scenario in Business Central handles roughly 70 percent of those orders end to end; the discipline is in defining which 30 percent always need a human.

What Automates

Inbound orders parsed from email or EDI. Validation against stock, pricing, customer-specific terms. Standard orders processed with fulfillment kickoff. Human review on exceptions (pricing overrides, stock shortages, new customers). The parser handles structured EDI cleanly; it handles unstructured emails with a confidence score, and the score drives the autonomous-versus-review decision.

Inbound order processing:
1. Parse: extract customer, lines, quantities, ship-to, terms
2. Match: link to customer master, item master, price agreement
3. Validate: stock, credit limit, ship-date feasibility
4. Decision: confidence above threshold = process; below = queue

The match step trips up most automation in older tools. Item aliases and unit conversions are where the agent earns its keep.

Decision Points

Confidence thresholds determine autonomous versus escalated handling. High-confidence standard orders process autonomously. Anything below threshold hits human queue with context. Tuning thresholds matters; start conservative.

threshold 0.95: 55% auto-process, 0.4% rework
threshold 0.90: 68% auto-process, 0.9% rework
threshold 0.85: 76% auto-process, 1.8% rework

The right threshold depends on the cost of a wrong order. For commodity items the cost is low; for configured products the cost is high. Set per item category if your catalog mixes both.

Purchase Scenarios

Requisitions to approvals to POs to receipts. Routine items flow autonomously within policy. Purchasing team focuses on strategic sourcing and exception handling. Day-to-day order-taking becomes largely invisible. The policy is the contract between the agent and the finance team.

purchase_policy:
  auto_approve_below: 500
  preferred_vendors_only: true
  budget_check: required
  delivery_window_days: 30

Quarterly review of the policy is non-optional. Budget thresholds drift faster than people remember.

SMB Impact

BC’s agentic features give small and mid-market orgs ERP-level automation that previously required custom integration work or enterprise ERP budgets. The accessibility shift is real. A 50-person company gets the order automation that used to require a six-figure consulting engagement plus a custom integration platform.

The trade-off is that the SMB needs an in-house owner for the agent configuration. Small businesses without an IT person sometimes fall back to manual processing because nobody owns the threshold tuning.

Exception Queue Design

The exception queue is where the human work concentrates. Design it for triage, not investigation. Each exception arrives with the original document, the agent’s confidence score, the field that failed validation, and the suggested fix. The clerk approves, edits, or rejects in three clicks.

async function loadException(orderId) {
  const ex = await fetch(`/api/v2.0/companies(${cid})/salesOrderExceptions(${orderId})`).then(r => r.json());
  return {
    document: ex.sourceDocument,
    confidence: ex.confidence,
    failedField: ex.validationFailure,
    suggestedFix: ex.suggestion
  };
}

A well-designed queue handles 30 to 50 exceptions per hour per clerk. A poorly designed one buries the clerk in raw data.

Integration with CRM

The sales scenario integrates with Dynamics 365 Sales for opportunity-to-order conversion. The opportunity in CRM becomes the sales order in BC through a triggered action; inventory reservation happens in BC and reflects back to CRM as quote line availability. This avoids the dual-write configuration that was previously required.

Measurement

Track auto-process rate, rework rate, and exception-handling time weekly. The leading indicator of trouble is exception time creeping up; that means the queue is not triaged fast enough and orders are aging.

What to do this week

Define the threshold per item category, design the exception queue with the clerk who will operate it, and run a one-week shadow pilot where the agent processes orders but the clerk reviews every output before fulfillment. Tune thresholds based on the shadow week before going live.

[object Object]
Share