Customizing Agents

Tailor each agent's behavior to match your business rules and preferences.

Agent Configuration

Go to Settings → Agents to customize:

1. Autonomy Level

Choose for each agent:

• Full Autonomy — agent acts independently; you see results after the fact
• Human-in-Loop — agent recommends; you approve before execution
• Audit-Only — agent monitors and logs; you decide everything

When to use each:

• Full Autonomy: routine tasks (quoting, tracking, standard customer inquiries)
• Human-in-Loop: sensitive decisions (carrier blocking, escalations)
• Audit-Only: until you trust the agent's behavior

2. Decision Thresholds

Set parameters that guide agent behavior:

Quoting Agent:

• Min FSC score: (default 70) — don't quote carriers below this
• Insurance required: (default yes)
• Fraud risk threshold: (default 50) — block carriers above this score
• Territory: (default all) — restrict to specific states if needed

Carrier Vetting Agent:

• FMCSA rating: (default "Acceptable") — allow Conditional or block Unsafe
• Insurance amounts: (default $100K cargo, $1M GL)
• Claims threshold: (default 2 per year) — flag if carrier has more claims
• Ownership screening strictness: (default standard)

Tracking Agent:

• Delay threshold: (default 30 min) — alert after X min late
• Route deviation: (default 5 miles) — flag if truck goes X miles off-route
• Update frequency: (default every 2 min) — how often to check telematics
• Escalation timeout: (default 2 hours) — auto-escalate if delay exceeds this

Customer Service Agent:

• Auto-answer topics: select which questions agent answers independently
• Escalation trigger: angry tone, refund request, dispute, etc.
• Response time target: 5 min, 30 min, 1 hour
• Escalation team: who receives escalations

3. Communication Templates

Customize how agents communicate:

Quoting Agent — (mostly silent; shows quotes)

Tracking Agent:

• "Your shipment is on time, arriving [ETA]" → customize language and branding
• "Your shipment is [X hours] late. We're working with the carrier." → customize urgency

Customer Service Agent:

• "Where's my shipment?" → template response with tracking link
• "I need to reschedule." → template response with rescheduling portal link

Go to Settings → Communication Templates to edit. All templates support variables:

• {shipper_name} — customer name
• {load_id} — load reference
• {eta} — estimated arrival
• {carrier_name} — carrier company
• {delay_minutes} — how many minutes late

4. Escalation Rules

Define when agents escalate decisions to you:

Examples:

"Escalate if quote margin < $100"

• Quoting agent finds a quote but margin is too thin → flags for approval before accepting

"Escalate if carrier has fraud flag"

• Vetting agent finds ownership concern → waits for your approval before approving carrier

"Escalate if customer mentions refund"

• Customer service agent detects "refund" in message → routes to your team instead of auto-responding

5. Performance Monitoring

Set targets for agent behavior:

• Quote volume: "Average 8 quotes per load" — if below, re-examine network size
• Escalation rate: "< 10% of carriers escalated for manual review" — if above, threshold too strict
• Resolution rate: "75%+ of customer inquiries resolved without escalation"

Monitor actual performance vs. targets in Analytics → Agent Health.

Role-Based Agent Configuration

Team members can have different agent permissions:

• Admins: customize all agents, set company-wide rules
• Brokers: use agents with default settings; can't change config
• Dispatchers: use agents (tracking, customer service only); no configuration access

Go to Settings → Team → select member → Agent Permissions to restrict access.

A/B Testing Agents

Test different configurations on different loads:

1. Go to Settings → Agents → Create Test Version
2. Name it: "Aggressive Quoting (low FSC threshold)"
3. Modify settings: Min FSC score = 60 (vs. default 70)
4. Select 20% of loads to test on (randomly assigned)
5. Run for 1 week
6. Compare results: quote volume, carrier quality, margin impact
7. If better, promote to default; if worse, discard

This is how you fine-tune agent behavior with data.

Learning from Agent Decisions

Check Analytics → Agent Logs to see:

• Every decision an agent made
• What information it considered
• What action it took
• Outcome (success, failure, escalation)

Example Agent Log Entry:

Agent: Quoting
Event: Load posted
Load ID: load_xyz123
Carriers queried: 42
Quote result: 15 quotes received, top 3 shown (FSC 82, 78, 75)
Agent decision: Recommended top quote (ABC Trucking, $2.12/mi, FSC 82)
Your action: Accepted top quote
Outcome: Load delivered on-time

Use these logs to understand agent reasoning and tune behavior.

Advanced Customization

Enterprise customers can build custom agents:

1. Define a new agent's purpose (e.g., "Pricing optimization agent")
2. Specify inputs (load data, historical margins, competitive rates)
3. Define outputs (suggested rate, confidence level)
4. Set thresholds (only activate if revenue > $5K/load)
5. We implement and deploy

Contact matt@freightconnect.ai to request custom agents.

Best Practices

1. Start conservative — high thresholds, human-in-loop. Gradually loosen as you trust the agent.
2. Monitor metrics — check agent health dashboard weekly
3. Provide feedback — mark agent decisions as correct/incorrect; it learns
4. Test changes — use A/B testing for major config changes
5. Document decisions — keep notes on why you configured agents a certain way

Troubleshooting

Agent too aggressive: Lower autonomy level to "Human-in-Loop". Increase decision thresholds (e.g., min FSC score).

Agent too passive: Raise autonomy level to "Full Autonomy". Lower thresholds.

Agent keeps making mistakes: Provide feedback (mark as incorrect) and review the logs to understand its reasoning. Adjust thresholds or escalation rules.

Need to revert to default: Go to Settings → Agents → Reset to Defaults for any agent.

Next Steps

• Agent Architecture — overview of all agents
• Quoting Agent — detailed quoting configuration
• Tracking Agent — shipment monitoring setup