15 KiB
15 KiB
ACI JIRA AI Fixer
Executive Proposal
Date: February 18, 2026
Version: 1.1
Update: Azure OpenAI mandatory for compliance
Classification: Internal - Executive
Executive Summary
The Problem
The support team faces growing challenges in resolving Support Cases:
| Challenge | Impact |
|---|---|
| Response time | Initial analysis consumes hours of senior developer time |
| Growing backlog | Issues accumulate while team focuses on urgent demands |
| Variable quality | Dependency on individual knowledge about the code |
| Concentrated knowledge | Few specialists know all modules |
The Solution
An Artificial Intelligence system that:
- Monitors new Support Cases in JIRA automatically
- Analyzes the problem and identifies affected source code
- Proposes specific fixes in COBOL, SQL, and JCL
- Documents the analysis directly in JIRA
- Creates branches with fixes for human review
Expected Result
┌─────────────────────────────────────────────────────────────────────────────┐
│ BEFORE vs AFTER │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ BEFORE AFTER │
│ ────── ───── │
│ Issue created Issue created │
│ ↓ ↓ │
│ Dev analyzes (2-4h) AI analyzes (5min) │
│ ↓ ↓ │
│ Search code (1-2h) Code identified │
│ ↓ ↓ │
│ Investigate cause (2-4h) Cause + suggested fix │
│ ↓ ↓ │
│ Develop fix (2-4h) Dev reviews and approves │
│ ↓ ↓ │
│ Review + deploy Review + deploy │
│ │
│ TOTAL: 8-14 hours TOTAL: 2-4 hours │
│ │
│ ✅ 60-70% reduction in resolution time │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Why Now?
1. Mature Technology
Language models (GPT-4, Claude) have reached sufficient quality for code analysis and generation, including legacy languages like COBOL.
2. Competitive Advantage
Leading companies are adopting AI to accelerate development. Those who don't adopt will fall behind in productivity.
3. Manageable Volume
With 5-10 issues/month, the risk is low and the environment is ideal to validate the solution before scaling.
4. Accessible Cost
Operational cost is minimal (~$500/month) due to low volume.
How It Works
Simplified Flow
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Support │ │ AI │ │ Dev │
│ Case │────▶│ Analyzes │────▶│ Reviews │
│ (JIRA) │ │ + Suggests │ │ + Approves │
└──────────────┘ └──────────────┘ └──────────────┘
5min 5min 30min-2h
┌─────────────────────┐
│ JIRA Comment: │
│ - Root cause │
│ - Affected files │
│ - Proposed fix │
│ - Link to PR │
└─────────────────────┘
Real Example
Issue: "Transaction being declined with code 51 even with available balance"
AI Response (in 5 minutes):
📋 AUTOMATIC ANALYSIS
🔍 Identified Cause:
The ACQAUTH.CBL program is comparing the WS-AVAILABLE-BALANCE field
with format PIC 9(9)V99, but the value returned from HOST uses
PIC 9(11)V99, causing truncation.
📁 Affected File:
- src/cobol/ACQAUTH.CBL (lines 1234-1256)
💡 Proposed Fix:
Change WS-AVAILABLE-BALANCE declaration to PIC 9(11)V99
and adjust the comparison in SECTION 3000-VALIDATE.
📊 Confidence: 87%
🔗 PR with fix: bitbucket.tsacorp.com/projects/ACQ/repos/...
Security: AI Does Not Alter Production Code
┌─────────────────────────────────────────────────────────────────────────────┐
│ SEPARATION OF RESPONSIBILITIES │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ CLIENT Repository (production) │
│ ACQ-MF-safra-fork │
│ ├── AI has access: READ ONLY │
│ └── Changes: ONLY by developers │
│ │
│ AI Repository (isolated) │
│ ACQ-MF-safra-ai │
│ ├── AI has access: READ AND WRITE │
│ └── Purpose: Branches with fix suggestions │
│ │
│ Approval Flow: │
│ 1. AI creates branch in isolated repository │
│ 2. AI opens Pull Request to client repository │
│ 3. HUMAN developer reviews │
│ 4. HUMAN developer approves or rejects │
│ 5. Only then code goes to production │
│ │
│ ✅ 100% of changes go through human review │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Investment
Development (MVP)
| Item | Investment |
|---|---|
| Development (4.5 months) | $70,000 - $90,000 |
| Initial infrastructure | $3,000 |
| Total MVP | $73,000 - $93,000 |
Estimate considers dedicated team of 4-5 professionals.
Monthly Operational Cost
| Item | Cost/Month |
|---|---|
| Artificial Intelligence APIs | $30 |
| Infrastructure (servers) | $200 - $500 |
| Maintenance (10% team) | $700 |
| Total Operational | ~$1,000/month |
Low cost due to volume of 5-10 issues/month.
Return on Investment (ROI)
Time Savings
| Metric | Value |
|---|---|
| Issues per month | 5-10 |
| Current average time per issue | 8-14 hours |
| Average time with AI | 2-4 hours |
| Savings per issue | 6-10 hours |
| Monthly savings | 30-100 hours of senior dev |
Financial Calculation
Senior developer hourly cost: ~$40
Monthly savings: 50 hours (average)
Value saved: $2,000/month
MVP Investment: $80,000 (average)
Operational cost: $1,000/month
Payback: ~40-48 months
However, considering:
- Scale to more clients/products
- Reduction of bugs in production
- Freeing devs for innovation
- Knowledge retention
Real ROI: Hard to quantify, but HIGHLY POSITIVE
Intangible Benefits
| Benefit | Impact |
|---|---|
| Standardization | All issues analyzed with same rigor |
| Documentation | Complete analysis history in JIRA |
| Knowledge | AI learns patterns, doesn't depend on people |
| Speed | Initial response in minutes, not hours |
| Team morale | Devs focus on complex problems, not repetitive ones |
Risks and Mitigations
| Risk | Probability | Mitigation |
|---|---|---|
| AI suggests incorrect fix | Medium | Mandatory human review in 100% of cases |
| Team resistance | Low | Position as assistant, not replacement |
| Code security | ✅ Eliminated | Azure OpenAI - data stays in Azure tenant, not used for training |
| LLM cost increases | Low | Enterprise Azure contract with fixed prices |
Compliance and Security
The solution exclusively uses Azure OpenAI, ensuring:
- ✅ Code data is not sent to public APIs
- ✅ Data is not used to train Microsoft models
- ✅ Processing in Brazil South region (low latency)
- ✅ Compatible with ACI corporate policies
- ✅ Uses existing Enterprise Agreement contract
Note: The existing GitHub Copilot will continue to be used by developers in the IDE. They are complementary tools - Copilot for code autocomplete, AI Fixer for automating issue analysis.
Conservative Approach
The system will be implemented in phases:
Phase 1 (MVP): Analysis and suggestion only
AI comments in JIRA, doesn't create code
Phase 2: Code generation in isolated repository
Human decides whether to use or not
Phase 3: Automatic Pull Requests
Human still approves
Phase 4: Auto-merge (only for high-confidence fixes)
Only after months of validation
Timeline
┌─────────────────────────────────────────────────────────────────────────────┐
│ MVP ROADMAP │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ Month 1-2 Month 2-3 Month 3-4 Month 4-5 │
│ ──────── ──────── ──────── ──────── │
│ Setup + Code Fix Tests + │
│ Integrations Indexing Generation Refinement │
│ │
│ ✓ JIRA ✓ COBOL ✓ LLM ✓ Pilot │
│ ✓ Bitbucket ✓ SQL ✓ Validation ✓ Adjustments │
│ ✓ Infra ✓ JCL ✓ Output ✓ Docs │
│ │
│ │ │
│ ▼ │
│ GO-LIVE │
│ ~4.5 months │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Solution Differentiators
Why not use ready-made tools (GitHub Copilot, etc)?
| Aspect | Generic Tools | Our Solution |
|---|---|---|
| JIRA Integration | ❌ Manual | ✅ Automatic |
| ACI system knowledge | ❌ Generic | ✅ Trained with ACI rules |
| COBOL expertise | ⚠️ Limited | ✅ Optimized for mainframe |
| Support Case flow | ❌ Doesn't exist | ✅ Native |
| Security (on-premise) | ❌ Cloud only | ✅ 100% internal |
| Customization | ❌ Generic | ✅ Configurable rules |
Recommendation
Requested Decision
Approve the development of the ACI JIRA AI Fixer MVP with:
- Investment: $73,000 - $93,000
- Timeline: 4.5 months
- Scope: ACQ-MF and ICG-MF products
- Objective: Reduce Support Case analysis time by 60%+
Next Steps (after approval)
- Week 1: Define team and start infrastructure setup
- Week 2: Create AI repositories and configure integrations
- Month 1: First demonstration with real issue
- Month 3: Functional MVP for pilot
- Month 5: Go-live in production
Conclusion
The ACI JIRA AI Fixer represents an opportunity to:
✅ Increase productivity of support team by 60%+
✅ Reduce response time from hours to minutes
✅ Standardize quality of analyses
✅ Retain knowledge independent of people
✅ Position ACI at the forefront of AI automation
The timing is ideal: mature technology, controlled volume for pilot, and urgent team demand.
Document prepared for Executive presentation.
For questions, the technical team is available for demonstration.