From “economically unfeasible” to overnight
With Exadel Colleague, what would have taken a year+ of manual FTE effort became an overnight output—making deep legacy QA coverage finally viable.
How a Leading U.S. EdTech Provider Accelerated Legacy ERP Modernization Using Agentic AIHow a Leading U.S. EdTech Provider Accelerated Legacy ERP Modernization Using Agentic AI
Introduction
Exadel deployed Exadel Colleague to generate large-scale unit test coverage inside existing workflows. The result: a regression safety baseline that made future modernization and migration decisions easier to prove—without adding delivery overhead.
Description
Label
The Risk We Had to Remove
The EdTech provider's core ERP ran on a heavily nested legacy codebase with limited documentation and few remaining original SMEs. That combination made migration risky: manual analysis was slow, expensive, and prone to regression.
The priority wasn’t “modernize at all costs”, it was to create a safer foundation for ongoing updates and future modernization by making the system’s behavior easier to verify and maintain at scale.
This is the first paper in a three-part series. Whitepaper 2 covers Operational Excellence. Whitepaper 3 covers Implementation, Roadmap, and Cloud Deployment.
Impact at a Glance
Automated test coverage to protect today’s platform—and build a baseline for what comes next.
~90% reduction in eligible testing effort
Creating 3,200+ human-equivalent hours
~7,000 unit tests generated overnight
Large-scale output with minimal token consumption
Regression safety baseline established
A reliable foundation for future modernization and migration work
Total token cost $130 - Economic quality at scale
Thousands of tests without adding delivery headcount or overhead
A real velocity proof point for modernization leadership
Proved Exadel + Colleague speed against traditional manual test creation
Agentic testing at scale, embedded in the workflow.
Fast adoption. Minimal disruption. Confidence you can build on.
Targeted Colleague deployment
Configured to automate unit test generation across the legacy codebase.
Workflow-native execution
Triggered through normal developer actions in Jira/GitHub—no new tools or habits required.
Low-friction adoption
No deep prompt engineering or specialist AI training needed to get value quickly.
Built to extend beyond testing
Deployed for unit tests, but the same integration accelerates other workstreams without modification, especially maintenance tasks.
Proven Solutions
Evidence & Performance
We measured Colleague’s test-generation run end to end, then broke results down by outcome category.
Assess
Identify AI-eligible work
Accelerate
Automated unit test creation
Control
Human review gate + routing
6,938 unit tests generated
In a single overnight run.
$129.72 total API compute cost
~$0.05 per eligible item.
3,215 human-equivalent hours recovered
From 3,587 hours total across AI-eligible work (~90% time reduction).
+$128,470 net engineering value
Recovered value across the overnight run (per model).
~402 engineering days unblocked
Recovered bandwidth redirected to core architecture vs low-level fixes.
Full pass
2,261 items • 2,967h saved
Baseline: $118,702 (2,967h)
Partial pass (60%)
472 items • 248h saved
Baseline: $24,780 (620h)
Total AI-eligible
2,733 items • 3,215h saved
Baseline: $143,482 (3,587h) • $130 total cost • +$128,470 net gain
Temporarily blocked
15,000+ items
Baseline: $787,500 (19,687h) (routed out of the AI-eligible flow)
Let’s use Exadel Colleague
to turn legacy complexity
into a regression-safe baseline.
Case Studies
