Audit-ready in weeks, not months - while shipping models 25% faster.
The Challenge
As a regulated lender, Meridian already had a model risk management function, but the problem was its speed. Every model change waited in a validation queue managed through spreadsheets, shared drives, and email approvals. The backlog stretched to weeks, so business teams either waited or shipped around the process. Meanwhile the EU AI Act classified the firm's credit-scoring models as high-risk, adding a second regulatory regime on top of existing model risk guidance, with evidence requirements the manual process had no way to meet.
The Solution
Thndr AI was integrated directly into the model registry and CI/CD pipelines the teams already used, with no migration and no new workflow to learn. Validation evidence is now generated from pipeline metadata as models are built, risk-tiering routes each change to the right level of review, and approval became an automated gate that opens in minutes for low-risk changes. Validators stopped processing a queue and started reviewing the exceptions that genuinely need human judgment.
Meeting Model Risk Where It Lives
The validation team's documentation standards didn't change; what changed is who assembles the evidence. Data lineage, test results, performance benchmarks, and approval records are captured automatically from the pipelines and composed into the validation artifact validators already expect. The team reviews substance instead of chasing attachments, and every artifact is reproducible on demand.
From Queue to Gate
The old process treated every change identically, which is why everything queued. Risk-tiering inverted that: a threshold tweak on a low-risk model passes the automated gate in minutes with full audit trail, while a new credit model still gets the complete human validation it always required. Throughput went up precisely because scrutiny became proportional to risk.
Regulatory Readiness as a Byproduct
When the EU AI Act's high-risk obligations landed, the firm didn't start a compliance project. It ran an export. The same governance records that satisfy model risk management mapped onto the Act's documentation and logging requirements, and the credit-scoring portfolio's conformity evidence was assembled from data that already existed. Two regulatory regimes, one system of record.
Results
Models reach production faster because approval effort is proportional to actual risk.
From first integration to producing complete validation evidence on demand.
Automated gates cleared the queue of low-risk changes that used to wait behind high-risk reviews.
Every production model carries a documented risk classification spanning both regulatory regimes.
See what Thndr AI can do for your team
Talk to our team about your specific AI governance challenges.
