Why the Model Lifecycle Matters More Than the Model Itself
In the first two articles in this series, I challenged a couple of assumptions that often shape discussions about modelling in banking.
The first was that the most persistent problems rarely sit within the algorithms themselves. The second was that governance, when properly designed and applied, is not the obstacle it is sometimes perceived to be. Taken together, these ideas lead to a fairly obvious question. If the problem is not really the model and it is not governance in isolation, then where does the friction actually come from?
In my experience, it lies in something that is often treated as secondary rather than fundamental: the model lifecycle. In many modelling conversations, attention quickly turns to model choice. Teams compare techniques, debate whether one approach outperforms another or question whether enough is being done to adopt more advanced methods. These discussions are understandable and, to a degree, healthy. They reflect how far the industry has come. At the same time, they can distort where the real challenges sit.
When you observe what happens once models enter an organisation, it becomes clear that most issues do not arise during model selection or even during development. Models are often built quickly and to a high standard. At this point, progress feels smooth and confidence is high. The slowdown tends to come later, as the model moves into validation, review and approval.
It is at this stage that a different set of questions begins to surface. Assumptions that felt obvious during development are revisited. Documentation is examined more closely. It becomes harder to trace how certain variables were constructed or why particular transformations were applied. Reproducing results can take longer than expected.
None of this necessarily suggests that the model itself is wrong. More often, it points to gaps in how the work has been carried out across the broader process. Decisions were made but not properly embedded. Context existed but was not preserved. What originally seemed like a single piece of work reveals itself to be a series of loosely connected steps, and this is where the lifecycle becomes critical.
A model is not an isolated artefact. It exists within a sequence of activities that begins well before development starts and continues long after deployment. Data sourcing, preparation, transformation, feature engineering, model development, validation, approval and ongoing monitoring are all part of the same journey.
When these stages are treated as separate activities, often owned by different teams and executed in different ways, small inconsistencies begin to accumulate. They may not be obvious early on, but they tend to surface under challenge or review. What first appeared straightforward becomes fragmented and harder to defend.
When the lifecycle is treated as a connected process, the dynamics change. Decisions made early in development remain visible later. Documentation evolves alongside the model rather than being reconstructed under pressure. Validation shifts from uncovering issues to confirming shared understanding. The process feels less like a sequence of handovers and more like a continuous flow.
One consequence of this is that it changes how model complexity is perceived. There is a common assumption that more advanced techniques will naturally deliver better outcomes. From a purely predictive perspective, that may be the case but in a banking environment, performance is only one part of the equation.
Models must also be explainable, challengeable and defensible. They are subject not only to statistical testing but to scrutiny from risk teams, auditors and regulators. In this context, complexity is not inherently problematic, but it places greater demands on the lifecycle that surrounds it. Where the lifecycle is weak or fragmented, complexity amplifies friction. Where the lifecycle is robust and well understood, complexity becomes far more manageable.
The organisations that navigate this most effectively are rarely those using the most sophisticated models. They are the ones that have invested in making their lifecycle consistent and transparent. Development, validation and governance are treated as parts of a single system rather than separate concerns. Traceability is built in, assumptions are explicit and models can move through the organisation without needing to be repeatedly reinterpreted.
Over time, this creates a different kind of capability. Delivery becomes more predictable, conversations become more focused and the overall process feels deliberate rather than reactive.
Ultimately, this requires a shift in perspective. The emphasis moves away from identifying the best possible model and towards ensuring the right environment exists for models to succeed. When the lifecycle works well, many of the challenges organisations struggle with begin to diminish on their own.
Written by Jalal Khoylou, co‑founder of Paragon Business Solutions, working across the credit industry.