Following a major application migration, a leading Canadian bank faced a challenge familiar to many large organizations: critical application knowledge was difficult to access, while quality engineering continued to require significant manual effort.
The challenge: Building a stronger foundation for modernization
The bank needed to strengthen its understanding of a complex application environment comprising more than 500,000 lines of code.
Documentation was incomplete, making business logic, dependencies and API behaviour more difficult to trace. Important application knowledge was also concentrated among a limited number of subject-matter experts. As a result, assessing changes and transferring knowledge could require significant time from experienced technical resources.
Quality engineering presented a related challenge. While artificial intelligence (AI) was beginning to support some aspects of test design, execution, debugging, defect analysis and reporting continued to involve substantial manual effort.
The opportunity was therefore broader than accelerating individual tasks. The bank wanted to create trusted, reusable knowledge and quality engineering capabilities that could support modernization across its application environment while maintaining human accountability, traceability and quality.
CGI’s approach: Creating reusable application knowledge with agentic AI
We deployed an agentic AI-based documentation approach to analyze the application codebase and reconstruct the knowledge embedded within it.
Using intelligent multi-agent orchestration, prompting and validation, the solution analyzed the code and generated structured technical documentation, API documentation and visual process maps for both technical and business audiences. The resulting documentation was published directly to the bank’s Confluence environment, making application knowledge easier to search, maintain and reuse.
This enabled the bank to transform fragmented application knowledge into a reusable asset that could support knowledge transfer, onboarding, impact assessments and future modernization planning, without requiring manual code walkthroughs.
Human validation remained an important part of the process. Subject-matter experts reviewed AI-generated outputs against defined quality expectations, helping ensure that greater speed did not come at the expense of accuracy or trust.
Strengthening quality engineering with AI-powered test intelligence
In parallel, we implemented an AI-powered testing approach spanning test design, execution, intelligent debugging and reporting.
Predictive failure analysis and self-correcting capabilities helped teams identify issues more quickly, reduce repetitive QA activities and gain greater visibility into testing progress, results and quality trends.
Together, these capabilities addressed two interconnected modernization needs: understanding how an application works and validating changes to it efficiently and confidently.
Delivering measurable results: Eight months of documentation work completed in six weeks
The engagement generated measurable improvements across application documentation and quality engineering:
- 95%+ first-pass acceptance: AI-generated documentation achieved more than 95% first-pass acceptance from the bank’s subject-matter experts, with minimal revision.
- Eight months to six weeks: Documentation work the bank estimated would require approximately eight months using a traditional approach was completed in six weeks.
- 80%+ reduction in documentation effort: Compared with the manual baseline, the approach reduced the person-hours required for documentation by more than 80%, freeing specialized technical resources for higher-value engineering activities.
- 500,000+ lines of code analyzed: More than half a million lines of application code were analyzed and converted into structured knowledge assets.
- Zero manual code walkthroughs: Documentation was generated without manual code walkthroughs, reducing dependency on scarce subject-matter
expertise.
Quality engineering results: 40% less manual QA effort per release
The quality engineering work also delivered measurable improvements, including a 40% reduction in manual QA effort per release, 50% faster diagnosis and resolution of test failures, a 25% reduction in end-to-end testing effort and 25% faster regression testing cycles.
From one application to a repeatable modernization capability
One of the clearest indicators of the engagement’s value came after the initial work was completed. Based on the results, the bank expanded the documentation approach to additional applications.
Rather than treating agentic AI as a standalone experiment, we established together an approach that could be reused as modernization extended across the application environment, combining AI acceleration with human validation and defined quality thresholds.
Extending AI-enabled delivery to credit-card modernization while reducing costs
As AI adoption expanded within the bank’s broader modernization portfolio, we also helped address the economics of AI-enabled software delivery (SDLC).
In a companion credit-card modernization program, AI token costs had reached approximately $1,000 per developer per month. We worked with the bank to optimize how AI was applied across the development workflow, combining task-based model selection, stronger AI literacy and conventional automation where AI was not required.
For the first use case delivered using this approach, monthly token costs decreased from approximately $1,000 to $200 per developer. A representative activity that previously required six to seven hours was completed in approximately 50 minutes.
Based on these results, the bank approved applying the approach to a more complex use case.
Building a foundation for ongoing modernization
For organizations managing large application environments, modernization depends on having a reliable understanding of how applications work, how they connect and how changes can be validated.
By combining agentic AI, quality engineering and human expertise, we helped the bank turn application knowledge into a reusable asset, reduce manual effort and establish AI-enabled capabilities that could extend beyond the initial engagement.
The experience also reinforced an important principle: AI creates the greatest value when it is applied to the right activities, with human judgment, quality controls and cost discipline built into the approach.
