Getting Started

Three tools that work together to help you make smarter purchase order decisions faster.

When you generate an AI comment on a PO

1

Fetch SKU Data

System collects full SKU details: price, coverage, supplier info, return rates

2

Apply Rules First

Check hard gates - negative margin? High return rate? Rules can block immediately

3

AI Agents Analyze

5 specialists evaluate coverage, margins, market trends, and supplier risk

4

Get Recommendation

Agents vote: Buy More, Buy Less, PASS, or Do not Buy - with confidence scores

5

Auto-Approve Decision

High confidence and thresholds met? Auto-approve immediately. Otherwise escalate

6

You See the Result

View recommendation, reasoning, and each agent perspective on the PO

How they connect

Rules
hard gates
AI Agents
decisions
Auto-Approve
Rules filter risk levels; Agents make decisions; Auto-Approve executes them

Rules

Decision gates that check PO data

Rules are conditions that evaluate fields on a purchase order. Like a dictionary: if negative margin is detected, mark as high risk. Rules can block orders, flag them for review, or feed data to agents.

Used by

  • AI Agents as hard gates or scoring inputs
  • Auto-Approve as risk-level filters and blocks

Example

EGM less than 0% - Do not buy (critical block)

Return rate greater than 10% - High risk flag

AI Agents

Specialists that evaluate every PO

Each Agent analyzes a purchase order through a specialist lens: coverage, margins, market trends, supplier risk. Agents vote their recommendation and confidence. Multiple agents form your Council.

Agent types

  • Rule-based: pure logic, no AI, just rule evaluation
  • Prompt-based: AI judgment with custom persona
  • Hybrid: rules as hard gates, AI for nuance

Example

Maya: margin rule fires - Do not buy with high confidence

Alex: weighs all 5 agents, issues final verdict

Auto-Approve

Skip manual review for clear-cut orders

Auto-Approve watches Council decisions and approves orders automatically when every configured threshold is met. You set the bar: confidence level, unanimity, order value, SKU risk.

Checks

  • Agent council decisions and average confidence
  • Minimum agents that must agree (e.g. 5 out of 5)
  • Max order value and allowed order types
  • Excluded suppliers (always manual review)

Example

5 agents agree - Confidence at least 90% - Order under 5k dollars - Auto-approved

Real-World Scenarios

Scenario

✓ Seasonal Restock - Auto-Approved

Restocking a popular kitchen gadget for Q2 season with 6 months of inventory. Margin is 12%, supplier is reliable, sales velocity is steady.

Calvin (Coverage):6 months matches seasonal target

Coverage rule check passes

Maya (Margin):Demand is steady

Sales velocity shows consistent trend, no red flags

Grant (Financials):Margin 12% is healthy

Covers fulfillment costs plus 5% buffer

Rita (Risk):Supplier track record clean

Return rate 2%, lead time reliable

Alex (Arbitrator):Approve with 95% confidence

All 4 agents vote yes, high confidence triggers auto-approve

Result: Auto-approved in seconds. All agents agree, confidence above 95%, all thresholds met.

Scenario

⚠ New Vendor Order - Escalated

First international supplier order for 18K electronics. New vendor with no history. Margin is tight at 8%.

Calvin:4 months coverage is acceptable

Within acceptable range for standard orders

Maya:No sales history available

Cannot assess demand for unknown product

Grant:Margin 8% is risky

Minimal safety margin if fulfillment costs spike

Rita (Risk):New vendor flag - High risk

No relationship history, international, 18K order value triggers escalation

Alex:Escalate for human review

Risk rule blocks auto-approve, requires manual sign-off

Result: Escalated to manual review. Rita risk rule triggers, new vendor requires human approval.

Scenario

✗ Clearance with Returns - Rejected

Clearance inventory at 40% discount. Supplier return rate is 18% (your threshold is 5%). Margin is negative at -2%.

Calvin:3 months coverage acceptable for clearance

Clearance orders allow shorter windows

Maya:Demand is declining

Product in clearance means demand is dropping

Grant:Reject - Negative margin

EGM is -2%, rule immediately fires, no profit possible

Rita:High return risk 18%

Return rate far exceeds 5% threshold, supplier unreliable

Alex:Do not buy - High confidence

Multiple rules and agents all reject, confident veto

Result: Rejected. Multiple rules fire, multiple agents agree - order does not proceed.

Example walkthrough: low-margin product arrives

1

Rules evaluate first

EGM is negative at -3% - the EGM rule fires immediately. Outcome: Do not buy (critical block)

2

AI Agents run

Maya (Margin Hawk): rule blocked, issues Do not Buy with high confidence Calvin, Grant, Rita: each issues their own decision via AI judgment Alex (Arbitrator): synthesizes all voices, Maya financial veto overrides others

3

Auto-Approve checks

Decision is "Do not Buy" - does not clear approval threshold Result: Routed to manual review for a buyer to confirm or override

For a healthy product: all 5 agents agree to buy more with high confidence and no rules block - auto-approve skips the queue entirely.

New agents need approval before going live

After creating an agent, a manager reviews and approves it at Approval. Until then, the agent is in draft and will not run on real orders.