Seed case study — illustrative, pending content-team sign-off and client approval before publication.
Challenge
A Tier-1 South African bank had no shortage of AI ideas. What it needed was a way to take one of them seriously — to stand it up in a form that risk, audit and the regulator could all live with. The bar was not “does it work in a notebook”; it was “can we defend this in production.”
Approach
We brought our agentic delivery capability — multi-agent patterns, retrieval, tool use, observability, guardrails and evaluation loops — under a Responsible-AI method, the 4 Ds: Delegation, Description, Discernment, Diligence.
The governance was the design, not an afterthought:
- Human sign-off on every production-bound change.
- Auditable agent activity — a complete record of what the system did and why.
- Segregation of duties carried into the agentic layer.
- Residency satisfying POPIA and GDPR.
Outcome
A won proof-of-concept — and, more importantly, a template for governed adoption the bank could repeat. Not “we have AI”, but speed the institution could defend to a regulator.