Regulatory fluency
Participants develop a working vocabulary that maps technical outputs to regulatory expectations - SR 11-7, BCBS 239, and local equivalents. This makes cross-functional conversations faster and less prone to misalignment.
Live Webinars
12 sessions active
Machine learning models
banks can actually explain
Regulators ask questions that most ML teams cannot answer. These sessions are built for practitioners who need to close that gap - not theoretically, but in the specific language of credit risk, audit trails, and model governance.
Who this is for
Conditions under which this works
These webinars are not introductory. They assume you already work with data in a financial institution - as an analyst, a model risk manager, a data scientist, or someone in a compliance function who needs to evaluate what the modelling team produces.
Participants who get the most out of each session tend to arrive with a specific problem they are trying to solve: a model that failed an internal review, a regulator query about decision logic, or a deployment that stalled because stakeholders could not interpret the output.
- You work in or alongside a financial institution's modelling or risk function
- You have encountered model explainability as a real constraint, not just a concept
- You can attend live sessions or engage with recorded replays within the week
- You are prepared to bring your own examples and ask specific questions
- You are based anywhere - the format is fully online, no travel required
"I already know the theory - will this actually add anything?"
Most explainability resources stop at SHAP plots and feature importance charts. The sessions here go further: into what happens when a regulator disagrees with your explanation, how to structure a model card that survives an audit, and where LIME breaks down under distribution shift in credit scoring data.
Knowing the theory is the starting point, not the destination. The gap between understanding SHAP values and being able to defend a model's decision to a credit committee is where most practitioners get stuck - and that gap is exactly what each session addresses.
Participants regularly bring cases from their own institutions.
Sessions are not lectures - they are working conversations with a structured agenda.
After the sessions
What stays useful six months later
Immediate knowledge fades. The more durable outcome is a changed way of evaluating models - a habit of asking the right questions before a problem reaches the review stage.
Structured decision logic
Working through real cases builds a mental framework for evaluating model decisions - not just whether the output looks right, but whether the reasoning behind it can be articulated to a non-technical audience.
Practical toolchain judgment
Sessions compare tools directly - when SHAP is the right choice, when Integrated Gradients is more appropriate, and when neither gives you what a compliance team actually needs. That judgment compounds over time.
The skill that compounds
Explainability is not a one-time deliverable - it is a recurring requirement that appears at model validation, at deployment, at audit, and whenever a model decision is challenged. Practitioners who can navigate it confidently find it affects their work continuously, not just during a single project.
Sessions are designed with that compounding in mind. Concepts are introduced in a way that makes them retrievable under pressure, not just recognisable in a textbook.
Participant accounts
Specific changes, in their words
"Before these sessions, I could produce a SHAP summary plot but I had no idea how to explain what it meant to the credit committee. After working through the session on stakeholder communication, I restructured how I present model outputs entirely. The next review went through without the usual back-and-forth."
Model review cycle shortened by two weeks"I came in skeptical. I had read the papers and done the online courses. What I had not done was sit in a session where someone challenged my interpretation of a LIME output with a real counterexample from a credit portfolio. That kind of pressure-testing is what changed my thinking - not the theory."
Identified a stability issue missed in prior validation
The professional environment
Peers who are working through the same problems
Each session draws practitioners from retail banks, development finance institutions, insurance companies with credit exposure, and regulatory bodies. The mix matters - a model validator and a data scientist looking at the same explainability problem from different sides of the table produce a more useful conversation than either would alone.
Between sessions, participants share documentation templates, regulatory correspondence they have navigated, and tools they have tested. This is not a managed community with moderated posts - it is a working group that happens to meet regularly.
340+
Practitioners across Southern Africa and beyond
8
Institution types represented in each cohort
4
Live sessions per month with recorded access
72h
Average response time on technical questions
Participants based outside major cities use the platform without any
disadvantage - sessions are online, materials are downloadable, and the discussion continues
asynchronously between live meetings.
What sets this apart
Four things this does that most alternatives do not
The comparison is not about quality - it is about fit. These are structural differences that affect whether a programme is useful for someone working in a regulated financial environment.
Cases are drawn from banking, not generic ML
Every worked example uses data structures and decision contexts from financial services - scorecards, PD models, behavioural models, fraud detection. Translating from a retail recommendation example to a credit application is not a trivial step, and these sessions skip it entirely.
Regulatory framing is built in, not added as a footnote
Explainability requirements in banking are not optional. Sessions treat the regulatory context as a design constraint from the start - not as a compliance checkbox at the end. This changes which techniques get covered and how they are evaluated.
Live format allows real-time challenge and correction
Recorded courses cannot tell you when your interpretation is wrong. Live sessions can - and do. Facilitators actively question participant reasoning, not to create difficulty, but because unchallenged misunderstanding is the most common reason practitioners struggle when it matters.
The programme has been running since 2017
The content has been tested, revised, and refined through hundreds of sessions with practitioners who brought real problems. What remains is what actually proved useful - not what sounded useful when the curriculum was first drafted.
