The Due Diligence Process
CFOs and controllers evaluating AI agents conduct thorough due diligence before committing budget. Understanding the questions finance leaders consistently ask helps frame evaluation and preparation.
These seven questions appear in virtually every finance leader discussion about AI agent implementation.
Question 1: What Is the Complete ROI Including Hidden Costs?
What Finance Leaders Want to Know
Surface question: "Will this pay for itself?"
Actual concern: Are there hidden costs that will make actual ROI worse than projected ROI? Implementation overruns, ongoing platform fees, staff time requirements, integration costs.
The Complete Answer
Visible implementation costs:
- Consulting and configuration: $20,000-$30,000
- Platform setup fees: $2,000-$5,000
- Staff time (60 hours): $5,000
- Total Year 1: $32,000-$45,000
Ongoing annual costs:
- Platform subscription: $3,600-$6,000
- Usage fees (calls, messages): $1,200-$2,400
- Staff oversight (40-60 hours): $2,000-$3,000
- Total annual: $6,800-$11,400
Typical savings (60 exceptions monthly):
- Staff time reduction: $14,400 annually
- Working capital improvement (collections): $60,000-$100,000 annually
- Total annual benefit: $74,400-$114,400
Payback timeline: 4-7 months
Hidden costs to watch:
- Scope expansion beyond initial implementation
- Data quality cleanup before implementation
- Integration complexity if ERP APIs limited
- Change management if staff resistance significant
Finance Leader Satisfaction
This question gets satisfactory answer when:
- All cost categories disclosed upfront
- Ongoing costs clearly separated from implementation
- ROI calculation includes working capital not just time savings
- Realistic payback timeline (6-12 months, not "immediate")
Question 2: What Happens If the AI Platform Provider Goes Out of Business?
What Finance Leaders Want to Know
Surface question: "Is the vendor stable?"
Actual concern: Are we locked into proprietary platform that creates dependency? What happens if vendor fails, gets acquired, or changes pricing dramatically?
The Complete Answer
Platform architecture matters:
Good scenario (standard platforms):
- AI capability from OpenAI, Anthropic, Google (major tech companies)
- Voice from Twilio, Vonage (established providers)
- Workflow orchestration from Make, Zapier (standard tools)
- Your conversation scripts and rules are portable
- Risk mitigation: Can switch implementation partners without losing IP
Bad scenario (proprietary platforms):
- Vendor uses proprietary AI models
- Custom voice platform no one else uses
- Conversation logic locked in vendor system
- Risk: Complete rebuild if vendor fails
Vendor stability indicators:
- Funding history and runway
- Customer base size
- Technology stack transparency
- Industry partnerships
Contractual protections:
- Data portability clauses
- Script and rule ownership documentation
- Reasonable termination terms
- Transition assistance commitments
Finance Leader Satisfaction
This question gets satisfactory answer when:
- Implementation uses standard, established platforms
- You own conversation scripts and business rules
- Contract includes data portability provisions
- Multiple implementation partners exist for same platforms
Question 3: What If AI Makes a Mistake That Damages Customer Relationships?
What Finance Leaders Want to Know
Surface question: "What is our liability exposure?"
Actual concern: Will AI say something inappropriate, make incorrect commitments, or handle situations poorly enough to damage valuable customer relationships or create legal exposure?
The Complete Answer
Risk types and mitigation:
Inappropriate communication:
- Risk: AI uses wrong tone, makes offensive statement, handles sensitive situation poorly
- Mitigation: All calls recorded, reviewed sample weekly, scripts refined continuously, escalation for emotional situations
- Frequency: Less than 1% of calls have tone issues with proper script development
Incorrect commitments:
- Risk: AI commits to payment terms, discounts, or policies outside authority
- Mitigation: Clear authority boundaries programmed, any non-standard requests escalate to humans
- Frequency: Virtually zero with proper rule configuration
VIP account mishandling:
- Risk: AI contacts relationship-critical account without personal attention
- Mitigation: VIP accounts flagged for human-only handling, no AI contact
- Frequency: Zero with proper account classification
Comparison to current risk:
Manual handling risks:
- Staff says inappropriate things occasionally (everyone has bad days)
- Documentation inconsistent (cannot review what was said)
- Follow-up forgotten (no systematic tracking)
- High-pressure creates mistakes
AI handling:
- Consistent tone and messaging
- Complete call recordings for review
- Systematic follow-up
- No pressure-induced errors
Legal exposure: Current manual processes have same exposure plus less documentation. AI adds recording and review capability.
Finance Leader Satisfaction
This question gets satisfactory answer when:
- All calls recorded and reviewable
- VIP account protection demonstrated
- Escalation logic clearly defined
- Comparison shows current risk is equal or higher
Question 4: How Much Ongoing Staff Time Does This Require?
What Finance Leaders Want to Know
Surface question: "Is this really automated or just shifting work?"
Actual concern: Are we eliminating 40 hours of coordination work only to create 35 hours of oversight work? What is the net time savings realistically?
The Complete Answer
Current state (60 exceptions monthly):
- Exception coordination: 30 hours monthly
- Escalation handling: 10 hours monthly
- Documentation: 5 hours monthly
- Total: 45 hours monthly
With AI agents:
- Oversight and review: 3-5 hours monthly
- Escalation handling: 12-15 hours monthly (volume increases as agent brings more exceptions to attention)
- Rule refinement: 2-3 hours monthly
- Total: 17-23 hours monthly
Net time savings: 22-28 hours monthly (50-60% reduction)
Why not 90% reduction:
- Escalations still require human handling
- Quality oversight ensures continued effectiveness
- Rule refinement enables continuous improvement
- Increased volume from systematic handling
Breakdown by phase:
Months 1-3 (implementation): 12-15 hours monthly (learning curve)
Months 4-6 (stabilization): 8-12 hours monthly
Months 7+ (steady state): 5-8 hours monthly for oversight, 10-15 for escalations
Finance Leader Satisfaction
This question gets satisfactory answer when:
- Net time savings are 50-60% (not 90%)
- Escalation time acknowledged upfront
- Ongoing oversight explained as necessary
- Time estimates match other implementations
Question 5: What About Data Security and Customer Privacy?
What Finance Leaders Want to Know
Surface question: "Is customer data protected?"
Actual concern: Are we sending sensitive customer information (contact details, payment history, account balances) to third-party AI platforms? What compliance or data protection issues does this create?
The Complete Answer
Data flow architecture:
What gets sent to AI platforms:
- Customer name and contact information
- Invoice details (number, amount, date, due date)
- Account status (overdue days, payment history)
- Previous communication notes
What does NOT get sent:
- Credit card numbers or payment method details
- Social security numbers or tax IDs
- Banking information
- Personally identifiable information beyond name and contact
Platform security:
Major AI platforms (OpenAI, Anthropic, Google):
- SOC 2 Type II certified
- Do not train on customer data
- Data encrypted in transit and at rest
- GDPR and CCPA compliant
Voice platforms (Twilio, Vonage):
- HIPAA compliant capabilities
- PCI DSS certified for payment data
- Call recording encryption
- Access controls and audit logs
Contractual protections:
- Data processing agreements (DPAs)
- Business associate agreements if needed
- Data retention and deletion policies
- Breach notification requirements
Comparison to current state:
Current manual handling:
- Staff access same data from ERP
- Email communications unencrypted often
- Phone calls not recorded
- Less oversight and audit trail
With AI agents:
- Same data access, more controls
- Encrypted communications
- All calls recorded for compliance
- Complete audit trail
Finance Leader Satisfaction
This question gets satisfactory answer when:
- Data flow clearly documented
- Platform certifications provided
- Comparison shows equal or better security
- DPA and contractual protections in place
Question 6: Can We Start Small and Expand Based on Results?
What Finance Leaders Want to Know
Surface question: "What is the minimum implementation?"
Actual concern: Are we forced into large upfront commitment before proving value? Can we test with limited scope and expand only if successful?
The Complete Answer
Pilot approach:
Minimum viable pilot:
- Single exception process (AR collections only)
- Limited customer segment (exclude VIP accounts)
- 90-day trial period
- 30-50 exceptions monthly minimum for meaningful test
Pilot investment:
- Implementation: $15,000-$25,000 (vs. $35,000 full)
- Platform: $1,000-$2,000 (3 months)
- Total pilot: $16,000-$27,000
Pilot success criteria:
- 60-70% complete handling rate
- 20-30% appropriate escalation rate
- Zero customer relationship damage
- Measurable time savings
- Staff acceptance
Expansion path if successful:
Phase 2 (Month 4-6):
- Expand to full customer base for collections
- Investment: $5,000-$10,000 incremental
- Platform scales with volume
Phase 3 (Month 7-12):
- Add second exception process (vendor bills or back orders)
- Investment: $10,000-$15,000 per process
- Leverage learnings from Phase 1
Total 12-month investment if all phases succeed: $31,000-$52,000
Exit option if pilot fails:
- Pilot investment lost
- No ongoing commitments
- Return to manual handling
- Learn from experience
Finance Leader Satisfaction
This question gets satisfactory answer when:
- Pilot option clearly available
- Investment scales with scope
- Success criteria defined upfront
- Exit option if results disappoint
Question 7: What Is the Real Implementation Timeline?
What Finance Leaders Want to Know
Surface question: "How long until we see results?"
Actual concern: Vendor says "4-6 weeks" but what is the realistic timeline including all the work we need to do? When do we actually achieve the promised time savings?
The Complete Answer
Complete timeline with all activities:
Weeks 1-2: Discovery and Planning
- Staff time required: 8-10 hours
- Activities: Document current process, define decision rules, identify exceptions
- Deliverable: Implementation plan and rule documentation
Weeks 3-4: Configuration and Script Development
- Staff time required: 6-8 hours
- Activities: Review conversation scripts, approve decision logic, test scenarios
- Deliverable: Configured agent ready for testing
Weeks 5-6: Testing and Refinement
- Staff time required: 10-12 hours
- Activities: Test with sample exceptions, listen to calls, refine scripts
- Deliverable: Agent handling sample exceptions successfully
Weeks 7-8: Limited Production Pilot
- Staff time required: 8-10 hours
- Activities: Agent handles 30-40% of exceptions, staff monitor closely
- Deliverable: Proof of concept with real exceptions
Weeks 9-12: Expansion to Full Volume
- Staff time required: 6-8 hours monthly
- Activities: Gradual volume increase, continue refinement
- Deliverable: Agent handling full exception volume
Total implementation: 10-12 weeks from kickoff to full production
When benefits begin:
- Week 8: First measurable time savings (limited scope)
- Week 12: Full time savings realized
- Month 4-6: Working capital improvements visible (collections)
Why timeline matters:
- Budget planning (ongoing costs start Month 3)
- Realistic staff capacity planning
- Appropriate success measurement timing
Finance Leader Satisfaction
This question gets satisfactory answer when:
- Complete timeline including staff time shown
- Benefits timing is realistic (not immediate)
- Phases clearly defined with deliverables
- Staff capacity requirements explicit
The Pattern Across Questions
What Finance Leaders Really Want
Not looking for: Perfect solution with zero risk
Actually seeking:
- Honest assessment of costs and benefits
- Realistic timeline and effort requirements
- Risk mitigation strategies
- Ability to test before full commitment
- Exit options if results disappoint
What Builds Confidence
Transparency: Acknowledging limitations, realistic success rates, ongoing effort requirements
Specificity: Actual numbers not ranges, comparable examples, documented approaches
Risk mitigation: Pilot options, standard platforms, contractual protections, escalation logic
Peer validation: Reference customers in similar situations, documented results, industry adoption data
The Reality
Finance leaders ask these seven questions consistently because they represent legitimate due diligence concerns. Satisfactory answers require transparency about costs, honest discussion of risks, realistic timelines, and demonstrated risk mitigation approaches.
Vendors who provide vague answers, dismiss concerns, or make unrealistic promises fail to gain finance leader confidence. Implementation partners who acknowledge limitations, provide specific examples, and demonstrate risk awareness earn trust and budget approval.
The questions are not obstacles. They are opportunities to demonstrate thorough understanding of operational and financial realities.