Routing Rules vs. AI Screening in Intake: Decision Guide (Without Tool Chaos)

The Core Question

Every law firm with intake volume eventually asks: Should we use simple routing rules or invest in AI screening?

The answer is not "AI is better" or "rules are enough." The answer depends on your volume, your data quality, and your tolerance for false positives.

What Routing Rules Actually Do

Routing rules are deterministic: IF condition THEN action.

Example:

• IF practice_area = "employment" AND urgency = "high" THEN route to Partner A
• IF source = "referral" THEN priority = high
• IF matter_value > 100000 THEN route to senior team

Strengths:

• Predictable behavior every time
• Easy to explain to partners
• No training data required
• Works from day one
• Easy to audit and modify

Weaknesses:

• Cannot handle nuance in free-text descriptions
• Requires well-structured input fields
• Breaks when data is messy or incomplete
• Adding new conditions = manual work

What AI Screening Actually Does

AI screening uses machine learning to classify inquiries based on patterns in historical data.

What it can detect:

• Practice area from unstructured case descriptions
• Urgency signals in language (legal deadlines mentioned, emotional distress)
• Quality indicators (detail level, coherence, spam patterns)
• Potential conflicts (entity recognition)

Strengths:

• Handles messy, unstructured input
• Improves with more data
• Catches patterns humans miss
• Scales without adding rules

Weaknesses:

• Requires training data (hundreds of labeled examples minimum)
• Black box problem: hard to explain specific decisions
• Can be confidently wrong
• Needs ongoing monitoring and retraining
• Higher implementation cost

Decision Framework: 5 Questions

Question 1: What is your monthly inquiry volume?

Question 2: How structured is your intake form?

Question 3: Do you have historical data?

AI needs training examples. If you have:

• < 100 labeled historical inquiries: Start with rules
• 100-500 labeled examples: AI possible but limited

• 500 labeled examples: AI can be effective

No historical data? Rules are your only option until you build a dataset.

Question 4: How costly are routing mistakes?

Question 5: What is your technical capacity?

The Hybrid Approach (Usually Best)

Most firms benefit from combining both:

Layer 1: Deterministic Rules (Always Runs First)

• Route known referral sources directly
• Flag high-value matters based on explicit indicators
• Apply practice-area routing from dropdown selections
• Enforce compliance rules (conflicts, jurisdiction)

Layer 2: AI Screening (For Ambiguous Cases)

• Classify practice area from free-text descriptions
• Detect urgency signals in language
• Score lead quality
• Suggest routing when rules cannot decide

Layer 3: Human Review (For Edge Cases)

• AI confidence below threshold
• High-stakes matters
• Novel case types
• Potential conflicts flagged

Implementation order:

1. Start with rules covering 60-70% of cases
2. Identify where rules fail or create bottlenecks
3. Add AI for those specific gaps
4. Monitor and refine both

Common Patterns That Work

Pattern A: Rules + AI Triage

• Rules handle clear-cut routing
• AI scores lead quality for follow-up prioritization
• Human reviews AI-flagged edge cases

Pattern B: AI First, Rules Override

• AI classifies all incoming inquiries
• Rules override AI for specific conditions (VIP sources, compliance requirements)
• Reduces manual work while maintaining control

Pattern C: Parallel Processing

• Rules and AI both process every inquiry
• Discrepancies flagged for human review
• Useful during AI testing phase

Red Flags: When AI Is Oversold

Be skeptical if a vendor promises:

"Our AI handles everything automatically"
Reality: You still need fallback rules and human review.

"No training data required"
Reality: Generic models work poorly for legal intake. Your data is different.

"100% accuracy"
Reality: No classification system is 100% accurate. What matters is error handling.

"Set it and forget it"
Reality: AI models degrade over time. Ongoing monitoring is required.

Implementation Checklist

For Rules-Based Routing

• Map all routing scenarios (who handles what)
• Define clear conditions (no ambiguity)
• Build escalation paths for unmatched cases
• Test with historical data
• Document all rules for maintenance
• Set up monitoring for unrouted inquiries

For AI Screening

• Audit historical data quality
• Define classification categories
• Label training data (minimum 100+ examples per category)
• Set confidence thresholds
• Build fallback to rules for low-confidence cases
• Plan for ongoing retraining
• Document model behavior for compliance

Cost Comparison

Next Step

Before deciding on AI, answer these questions:

1. What percentage of inquiries do current rules handle correctly?
2. Where specifically do rules fail?
3. Do you have historical data to train AI?
4. What is your tolerance for AI errors?

If rules handle 80%+ correctly and you lack training data: optimize rules first.

If rules struggle with free-text classification and you have data: explore AI for that specific gap.

Guide: Intake Routing for Law Firms

Related:
Status Model in Intake: 2 Variants