What Good Forecast Model Design Actually Looks Like

A forecast model that only its creator can navigate is a liability. Bad model design shows up as stakeholders who distrust the outputs, scenarios that aren’t run, and forecasting teams that spend time maintaining something nobody interrogates. Drawing on J+D Forecasting expertise, this article sets out what good model design looks like and where it most commonly goes wrong.

Share:

Download today








Drawing on the expertise of J+D Forecasting, this article examines forecast model design: how to build a model that supports decision-making across the full planning cycle, and the design failures that undermine even technically capable forecasting functions.

The article covers:

  • Why models built for analysts erode trust and get abandoned
  • How to balance model complexity against usability
  • What scenario functionality requires, and where communication breaks down
  • How to build version control teams actually sustain
  • How to tell whether it’s a model or process problem

Also included: an expert Q&A with Abigail Spedding, Implementation Director at J+D Forecasting, on design horizons, the audit diagnostic, and the organizational habits that most often sit behind a forecasting function that has stopped working.

Be the first in line for every new episode.

Frequently Asked Questions

A forecast model is fit for purpose when decision-makers can use it, not just the analysts who built it. That means a clear structure, transparent assumptions, output charts that respond visibly to changed inputs, and scenario functionality that is genuinely easy to run. The technical quality of the underlying methodology matters, but a model that stakeholders cannot interrogate will not drive decisions regardless of how well it is constructed.

The right level of complexity is whatever is needed to model the components that matter most for the brand in question. Pharmaceutical markets are inherently complex, but that is not a reason to replicate every dimension of that complexity in the model. Models that accumulate features incrementally tend to end up as one-user tools that overwhelm anyone who did not build them. The practical test is whether someone who was not involved in building the model can pick it up, change an assumption, and understand what happened to the forecast.

Different functions need different things from the same scenario outputs. A finance team, a commercial team, and senior leadership are asking different questions and making different decisions. A model that presents scenario results identically to all of them tends to create disengagement across at least some of those groups. Good model design builds in the ability to present the same underlying analysis in formats relevant to each audience.

They tend to show up differently. Model problems look like assumptions that cannot be challenged, a structure only its original builder can operate, or outputs that cannot answer the decisions being made. Process problems look like a technically sound model that nobody is using consistently, or scenario outputs that get produced but never feed into decisions. The diagnostic question is whether the function has the right model, with appropriate methodology, transparent assumptions, and documented sources. If yes but those standards are not being followed, rebuilding the model will not fix it.

AI leaves the judgment calls central to good forecasting to people: what a variance actually means, whether a risk is real, whether the market is shifting in a way that changes the story. It handles the repetitive manual work that slows planning cycles down, including updating actuals, running scenario variations, flagging deviations beyond a defined threshold, and synthesizing competitive intelligence before each cycle. The teams that get the most from AI use it to create more time for analysis that requires human judgment, not just to accelerate output production.

Related Content