Day 1 Module 2: GenAI Use Cases for Finance

The GenAI Application Lifecycle

Every successful GenAI project follows this cycle:

Phase	What you do	Key question
1. Define	Identify the use case and business problem	Is this a good fit for GenAI?
2. Select	Choose the right foundation model	Which model fits the task, budget, and latency needs?
3. Improve	Optimize performance (prompt engineering, RAG, fine-tuning)	Is the output quality good enough?
4. Evaluate	Measure quality with rubrics and testing	How do we know it's working?
5. Deploy	Put it into production with guardrails	Is it safe, compliant, and monitored?

When GenAI IS the Right Solution

Generative AI excels when the task involves unstructured data, repetitive content creation, or personalization at scale.

Category	Finance Use Case	Example
Content Creation	Reports, narratives, summaries	Auto-draft merchant risk assessments from transaction data — structured GREEN/AMBER/RED reports in seconds
Personalization	Tailored financial communications	Generate personalized PayLater limit explanations, customized collection reminders based on customer history
Automation	Document processing at scale	Extract data from thousands of vendor invoices, classify and route customer complaints
Data Augmentation	Synthetic data for testing	Generate realistic test transaction data for new market launches without using real customer data
Creative Exploration	Ideation for process improvement	Brainstorm automation opportunities, draft policy change impact assessments
Localization	Multi-market content	Adapt compliance notices and internal policies across SG, MY, ID, TH, VN, PH

When GenAI is NOT the Right Solution

Not every problem needs GenAI. These scenarios require different approaches:

Scenario	Why Not	Better Approach
Real-time fraud scoring	Requires sub-millisecond deterministic decisions at massive scale	Traditional ML models (XGBoost, neural nets)
Interest rate calculations	Requires exact math, zero tolerance for approximation	Deterministic financial formulas
Regulatory capital calculations	Must be auditable, reproducible, and precisely calculated	Rule-based systems with audit trails
Credit scoring (final decision)	Regulatory requirements for explainability and fairness	Traditional ML with SHAP/LIME explainability
Transaction settlement matching	Exact matching logic, penny-accurate	Deterministic matching algorithms
AML final determinations	High-stakes regulatory decisions requiring human accountability	Rule-based detection + human investigation

⚠️ Rule of thumb: If the task requires deterministic, penny-accurate results OR makes irreversible high-stakes decisions — use traditional ML or rule-based systems. If it involves generating text, summarizing documents, or creating content — GenAI is likely a good fit.

GenAI Use Cases by Finance Function

Here's how GenAI maps to the key finance processes your team manages:

💰 Procure-to-Pay (PTP)

Use Case	What AI does	Impact
Invoice Data Extraction	Extract vendor, amount, line items, dates from PDF invoices	Hours → seconds per batch
PO Matching & Validation	Match invoices to purchase orders, flag discrepancies	Catch mismatches before payment
Vendor Communication Drafts	Generate payment status updates, dispute responses	Consistent, professional communications
Exception Narratives	Write explanations for flagged invoices	Faster exception resolution
Spend Analysis Reports	Summarize spending patterns by vendor, category, market	Executive-ready insights from raw data

📊 Record-to-Report (RTR)

Use Case	What AI does	Impact
Journal Entry Narratives	Generate explanations for manual journal entries	Audit-ready documentation
Reconciliation Exception Reports	Summarize unmatched items with suggested actions	Faster period close
Monthly Finance Review Decks	Generate presentations from KPI spreadsheets	Hours of deck-building → minutes
Variance Analysis Narratives	Explain budget vs actual variances in plain language	Board-ready commentary from numbers
Regulatory Filing Drafts	Draft sections of regulatory submissions	Consistent, compliant language

🛡️ Risk & Compliance

Use Case	What AI does	Impact
Merchant Risk Assessment	Analyze transaction patterns → structured risk report	Consistent assessments at scale
Regulatory Impact Analysis	Scan new circulars → assess impact on operations	Rapid response to regulatory changes
KYC Document Parsing	Extract fields from identity documents	Faster onboarding, fewer errors
Compliance Audit Reports	Generate audit-ready documentation from transaction scans	Weeks → days for audit prep
Policy Q&A (RAG)	Answer compliance questions grounded in policy documents	Consistent, cited answers

💳 Payments & PayLater

Use Case	What AI does	Impact
Credit Risk Narratives	Convert PD/LGD/DSCR into plain-language narratives	Faster credit decisions
Customer Dispute Responses	Draft chargeback dispute responses with evidence	Faster resolution, reduced backlog
Collection Communications	Generate personalized, empathetic reminders	Better recovery rates
Transaction Anomaly Summaries	Summarize flagged patterns for investigation	Faster triage
Product FAQ Generation	Generate customer-facing FAQs for PayLater terms	Consistent info across channels

📢 Regional Operations & Reporting

Use Case	What AI does	Impact
Management Report Narratives	Generate commentary for regional dashboards	Consistent voice across 6 markets
Cross-Market Comparison	Analyze performance differences across markets	Identify trends automatically
Project Status Summaries	Extract tasks from email threads → visual timeline	Consolidate scattered info
Board Presentation Drafts	Generate executive summary slides from data	Reduce prep time for reviews

Use Case Scenarios

Evaluate these scenarios — is GenAI the right approach?

Scenario 1 ✅ Invoice Processing Automation

Use case: A financial services company processes 5,000+ vendor invoices per month across 6 Southeast Asian markets, each in different formats.

Challenge: Manual data entry takes 15-20 minutes per invoice. Mismatches between invoices and POs are caught late.

GenAI Solution: Extract structured data from PDF invoices, match against purchase orders, flag discrepancies, generate validation reports.

Why it works: Large volume of unstructured documents, repetitive patterns, clear validation rules, human-in-the-loop for exceptions.

Scenario 2 ❌ Real-Time Transaction Fraud Scoring

Use case: A digital payments company processes 3 million transactions per day. Each needs a fraud risk score within 50 milliseconds.

Why GenAI doesn't fit: LLMs are too slow (seconds, not milliseconds), non-deterministic (different answers each time), and can't handle the throughput. Traditional ML models are purpose-built for this.

Scenario 3 ✅ Merchant Risk Assessment Narrative

Use case: Risk analysts assess 200+ merchants quarterly, writing narrative reports with transaction analysis, chargeback trends, and recommended actions.

Challenge: Each assessment takes 30-45 minutes manually. Quality varies between analysts.

GenAI Solution: Generate standardized risk assessments (GREEN/AMBER/RED) using a reusable prompt template with structured output and RAG grounding.

Why it works: Repetitive narrative generation, clear structure, data-grounded, human review before action.

💡 You'll build this template in the Day 2 prompt engineering exercise!

Scenario 4 ✅ Regulatory Impact Assessment

Use case: When a regulator issues a new circular, the compliance team must assess its impact across all affected markets.

GenAI Solution: Use RAG to ground the AI in the new circular AND internal policy documents. Generate a structured impact assessment with affected operations, required changes, and compliance timeline.

Why it works: Document analysis and synthesis, grounded in source documents, structured output, human review for final decisions.

Scenario 5 ❌ Capital Adequacy Calculation

Use case: Calculate risk-weighted assets and capital ratios for regulatory reporting.

Why GenAI doesn't fit: Zero tolerance for approximation. Regulatory calculations require deterministic formulas, not probabilistic text generation. The audit trail must show exact computation steps.

Real-World Impact: Financial Services on AWS

Financial services companies are already seeing measurable results from GenAI and Agentic AI:

Impact Area	Result	What this means for your team
Contact center	94% reduction in wait times, $7.5M savings in 6 months	Customer support for PayLater disputes and insurance claims
Fraud operations	20% efficiency improvement in investigations	Fraud team investigates more cases with AI-assisted summaries
Reconciliation	90% reduction in manual reconciliation effort	RTR team — dramatically faster period close
Report generation	80% faster regulatory and accounting reports	Monthly reports in hours, not days
Financial close	50% faster close cycles	Period close in half the time
Research	Research time reduced from 3-4 weeks to 1 week	Planning team — competitive analysis in a quarter of the time
Analytics	80% reduction in routine analytical tasks	Data analysts focus on insights, not data gathering
Team productivity	40,000 team member hours saved	Equivalent to 20 full-time employees worth of capacity

Use Case Evaluation Checklist

When evaluating a finance use case for GenAI, check these boxes:

☐ Involves large volumes of unstructured data (invoices, emails, documents, reports)
☐ Requires human-like text generation (narratives, summaries, communications)
☐ Currently done manually and repetitively (monthly, weekly, or per-transaction)
☐ Can tolerate some imprecision (human review catches errors before action)
☐ Benefits from consistent structure and quality across analysts/markets
☐ Has clear evaluation criteria (good/bad output is measurable)
☐ Does NOT require deterministic, penny-accurate calculations
☐ Does NOT make irreversible high-stakes decisions alone
☐ Complies with regulatory requirements for auditability and explainability

Workshop Connection

The use cases above map directly to what you'll build in the workshop:

Use Case	Workshop Exercise	Day
Invoice processing & PO matching	Module 1: Invoice Processing	Day 2
Transaction monitoring & reconciliation	Module 2: Transaction Dashboard	Day 2
Monthly finance review presentations	Module 3: Finance Presentation	Day 2
Project planning from email threads	Module 4: Project Planning	Day 2
Fraud detection & investigation	Module 5: Fraud Detection	Day 2
Regulatory compliance reporting	Module 6: Compliance Report	Day 2
Merchant risk assessment narrative	Prompt Exercise 1	Day 2
Credit risk narrative for committees	Prompt Exercise 2	Day 2
Workflow automation design	Agent Design Canvas	Day 3

💬 Discussion Questions

Think about these as we go through the module:

"Think about your month-end close — where do you spend the most time on repetitive writing that AI could draft for you?"
"Where would you absolutely NOT trust an AI to make the final call in your operations?"
"If you could give an AI access to all your transaction data, what's the first report you'd ask it to generate?"
"How would your auditors react if they knew an AI wrote the reconciliation narratives?"
"What's more dangerous — a false positive that flags a legitimate merchant as risky, or a missed risk that lets a bad merchant through?"

← Workshop Home

💡 Exploring GenAI Use Cases for Finance

The GenAI Application Lifecycle

When GenAI IS the Right Solution

When GenAI is NOT the Right Solution

GenAI Use Cases by Finance Function

💰 Procure-to-Pay (PTP)

📊 Record-to-Report (RTR)

🛡️ Risk & Compliance

💳 Payments & PayLater

📢 Regional Operations & Reporting

Use Case Scenarios

Scenario 1 ✅ Invoice Processing Automation

Scenario 2 ❌ Real-Time Transaction Fraud Scoring

Scenario 3 ✅ Merchant Risk Assessment Narrative

Scenario 4 ✅ Regulatory Impact Assessment

Scenario 5 ❌ Capital Adequacy Calculation

Real-World Impact: Financial Services on AWS

Use Case Evaluation Checklist

Workshop Connection

💬 Discussion Questions