1. Abstract
Selecting the appropriate historical start date ensures that analysis reflects relevant business conditions while preserving sufficient statistical depth. A well-chosen time window improves the interpretability, stability, and credibility of insights generated in Discover and Predict.
2. Context
Apply this best practice when creating a new Workbench or revising an existing one before running Discover or Predict. The selected historical window directly influences correlations, coefficient behavior, and forecast reliability.
3. Content
3.1 Why It Matters
The time horizon used in analysis shapes the relationships the system detects. Including too much historical data may introduce structural regimes that no longer reflect the current business environment. Conversely, using too short a history reduces statistical robustness and increases the risk of overfitting to recent trends.
Common risks of poor start-date selection include:
- Capturing outdated business models or legacy operating structures
- Inflating correlations driven by temporary macroeconomic conditions
- Overfitting to short-term volatility
- Reducing the number of usable observations for modeling
The goal is to balance relevance (reflecting current operating conditions) with depth (capturing enough history to identify stable patterns).
3.2 How to Apply
When selecting or revising a Workbench start date:
- Define the business question clearly.
Identify what decision the analysis is supporting and the timeframe most relevant to that decision. - Identify structural shifts in the business.
Examples include:- Mergers or acquisitions
- Major pricing model changes
- Distribution channel shifts
- ERP or reporting system transitions
- Significant regulatory changes
- Assess macroeconomic regime changes.
Periods such as financial crises, pandemic disruptions, or major policy shifts may distort relationships if included without consideration. - Maintain sufficient historical depth.
Aim for at least 60 monthly observations where possible to preserve statistical stability. - Confirm indicator coverage.
Ensure all explanatory variables span the full selected time window. - Document the rationale.
Record why a particular start date was chosen to support transparency and future reviews.
3.3 Example
A company that transitioned from wholesale to direct-to-consumer sales excludes data prior to the transition. Including earlier data would introduce relationships that no longer reflect current revenue dynamics. By starting the Workbench after the shift, correlations better align with the present operating model.
3.4 Common Pitfalls
- Using the maximum available history by default without assessing relevance
- Selecting overly short windows to artificially improve correlation strength
- Ignoring structural or macro regime changes
- Failing to revisit the start date as the business evolves
3.5 Expected Results
- Stronger alignment between analysis and current business conditions
- More stable coefficients and correlations
- Improved interpretability of relationships
- Greater confidence in insights and forecasts
- Clear documentation supporting governance and review
