Financial modelling, flexibility, and why pre-built tools keep failing
11 January 2026
For a long time, I kept wondering whether it was possible to design a genuinely flexible financial modelling journey. Not a better template, not a smarter platform, but a way of working that could adapt to the reality of how questions actually emerge in finance. Over time, the conclusion became difficult to avoid: such a journey cannot be pre-built, because the questions themselves are neither stable nor predictable.
In real situations, modelling does not start from a clean objective. It starts from partial information, unclear intent, competing concerns, and often from disagreement about what the problem even is. A discussion that begins as a valuation exercise can quickly turn into a liquidity question, then drift into governance, timing, or risk appetite. Each answer reshapes the next question, and often invalidates the structure that was just built.
This is not an edge case. This is normal.
Any tool that assumes a predefined path, however flexible it claims to be, is already misaligned with how financial thinking actually unfolds. The problem is not poor software design. The problem is that modelling is not a workflow to be executed, but a process of inquiry that mutates as understanding evolves.
This is where Excel’s persistence starts to make sense. Excel is often criticized for being messy, unstructured, and permissive. Yet that permissiveness is precisely why it survives. Excel allows incomplete logic to coexist with provisional assumptions. It allows contradictions to exist long enough for them to be noticed. It does not require the modeler to know in advance what the final structure will look like.
In other words, Excel tolerates intellectual disorder. Most tools try to eliminate it.
Over the years, many platforms have attempted to improve on Excel by offering guidance, structure, and best-practice workflows. Almost all of them eventually disappoint in high-stakes decision settings. They are usually very good at running known models, but much less capable of supporting the discovery of what the model should be in the first place. They tend to harden structure too early, which creates a false sense of clarity and an unhelpful rigidity when the conversation inevitably shifts.
This limitation becomes particularly visible at committee level. Credit committees, investment committees, and boards are not environments where modelling exists to compute optimal answers. They are environments where judgment is exercised under accountability. What matters in those rooms is not only whether the numbers are correct, but whether the assumptions are visible, discussable, and defensible.
This helps explain why technically superior Python models almost never appear directly in such settings. The issue is not capability. It is legibility. Code concentrates logic in a form that is difficult to inspect socially. Excel, for all its flaws, externalizes logic onto a surface that can be collectively examined. Even if nobody edits the file live, the fact that it could be edited matters. It signals reversibility and shared ownership.
For a long time, this made Excel both the place where thinking happened and the place where decisions were formalized. That dual role is now starting to change, not because Excel is being replaced, but because something new has entered upstream.
AI changes the economics of exploration.
Instead of forcing early commitment to structure, AI allows vague questions to be explored without penalty. Provisional models can be generated, challenged, and discarded quickly. Alternative framings can be tested without the organizational cost that usually accompanies rebuilding a model from scratch. In this sense, AI absorbs ambiguity rather than trying to eliminate it.
What AI does not do is create a universal modelling environment. That problem remains unsolved because it is not solvable. The diversity of questions, contexts, and decision dynamics makes any single guided journey brittle. What AI enables instead is a separation of roles.
Thinking no longer has to happen inside the final artefact.
Exploration can happen in language, in sketches, in temporary calculations that are never meant to survive. Only once something meaningful has emerged does it need to be distilled into a form that is suitable for decision-making. That form often still looks like Excel, not because Excel is ideal, but because it remains socially aligned with how decisions are legitimized.
Seen this way, the future of financial modelling is not Excel versus AI, nor is it about replacing one tool with another. It is about allowing thinking to remain flexible for longer, and freezing it only when accountability requires it. AI expands the space of possible questions. Excel continues to mark the moment when a particular answer is put forward to be owned.
Pre-built modelling tools will continue to struggle as long as they promise flexibility through structure. Flexibility in finance does not come from better interfaces or more configuration options. It comes from the ability to delay commitment until understanding has matured.
For finance leaders, the practical implication is subtle but important. The question is no longer which tool to standardize on, but where different kinds of work should happen. Exploration and sense-making benefit from openness and reversibility. Committees require clarity, stability, and legibility. Trying to force both into the same medium inevitably creates friction.
Excel endures because it fits the second role. AI changes the first.
That division, rather than any new platform, is likely to shape how modelling actually evolves in practice.
This is not an edge case. This is normal.
Any tool that assumes a predefined path, however flexible it claims to be, is already misaligned with how financial thinking actually unfolds. The problem is not poor software design. The problem is that modelling is not a workflow to be executed, but a process of inquiry that mutates as understanding evolves.
This is where Excel’s persistence starts to make sense. Excel is often criticized for being messy, unstructured, and permissive. Yet that permissiveness is precisely why it survives. Excel allows incomplete logic to coexist with provisional assumptions. It allows contradictions to exist long enough for them to be noticed. It does not require the modeler to know in advance what the final structure will look like.
In other words, Excel tolerates intellectual disorder. Most tools try to eliminate it.
Over the years, many platforms have attempted to improve on Excel by offering guidance, structure, and best-practice workflows. Almost all of them eventually disappoint in high-stakes decision settings. They are usually very good at running known models, but much less capable of supporting the discovery of what the model should be in the first place. They tend to harden structure too early, which creates a false sense of clarity and an unhelpful rigidity when the conversation inevitably shifts.
This limitation becomes particularly visible at committee level. Credit committees, investment committees, and boards are not environments where modelling exists to compute optimal answers. They are environments where judgment is exercised under accountability. What matters in those rooms is not only whether the numbers are correct, but whether the assumptions are visible, discussable, and defensible.
This helps explain why technically superior Python models almost never appear directly in such settings. The issue is not capability. It is legibility. Code concentrates logic in a form that is difficult to inspect socially. Excel, for all its flaws, externalizes logic onto a surface that can be collectively examined. Even if nobody edits the file live, the fact that it could be edited matters. It signals reversibility and shared ownership.
For a long time, this made Excel both the place where thinking happened and the place where decisions were formalized. That dual role is now starting to change, not because Excel is being replaced, but because something new has entered upstream.
AI changes the economics of exploration.
Instead of forcing early commitment to structure, AI allows vague questions to be explored without penalty. Provisional models can be generated, challenged, and discarded quickly. Alternative framings can be tested without the organizational cost that usually accompanies rebuilding a model from scratch. In this sense, AI absorbs ambiguity rather than trying to eliminate it.
What AI does not do is create a universal modelling environment. That problem remains unsolved because it is not solvable. The diversity of questions, contexts, and decision dynamics makes any single guided journey brittle. What AI enables instead is a separation of roles.
Thinking no longer has to happen inside the final artefact.
Exploration can happen in language, in sketches, in temporary calculations that are never meant to survive. Only once something meaningful has emerged does it need to be distilled into a form that is suitable for decision-making. That form often still looks like Excel, not because Excel is ideal, but because it remains socially aligned with how decisions are legitimized.
Seen this way, the future of financial modelling is not Excel versus AI, nor is it about replacing one tool with another. It is about allowing thinking to remain flexible for longer, and freezing it only when accountability requires it. AI expands the space of possible questions. Excel continues to mark the moment when a particular answer is put forward to be owned.
Pre-built modelling tools will continue to struggle as long as they promise flexibility through structure. Flexibility in finance does not come from better interfaces or more configuration options. It comes from the ability to delay commitment until understanding has matured.
For finance leaders, the practical implication is subtle but important. The question is no longer which tool to standardize on, but where different kinds of work should happen. Exploration and sense-making benefit from openness and reversibility. Committees require clarity, stability, and legibility. Trying to force both into the same medium inevitably creates friction.
Excel endures because it fits the second role. AI changes the first.
That division, rather than any new platform, is likely to shape how modelling actually evolves in practice.