Why is your team still using Excel to forecast?

Key ideas

• Teams export from million-dollar systems into spreadsheets because the platform is not giving them the flexibility, transparency or trust the month requires. The question is what the system is failing to deliver, not why planners are "going rogue."
• A batch job is not evidence that assumptions fit. Evaluate the treatment of heterogeneous histories, covariates and decision horizons. Global and local methods should earn their role in comparative tests.
• Automated pipelines, including Prophet and NeuralProphet, need appropriate assumptions and comparative evaluation. Inspectability helps teams review the model; manual interaction is not a proxy for forecasting quality.
• Research finds that adjustment performance varies with context, direction and size. Preserve the input and revised versions, evaluate the incremental change, and separately compare the whole process with credible baselines.
• LLMs are for language and reasoning, not numerics. Use them to capture messy context and explain assumptions; leave the math to proven forecasting and optimization methods.
• The durable pattern is a governed system of record plus a flexible decision layer, where overrides are visible and attributable, and proposed improvements are evaluated before deployment.