1. Abstract
Consistent, well-documented data preparation is the foundation of every successful analysis in Board Foresight. Following a structured workflow ensures that models are built on reliable inputs, reduces downstream rework, and improves forecast accuracy and explainability.
2. Context
Use this workflow whenever internal or external datasets are prepared for use in Foresight, including before running Discover, building a Workbench, or executing Predict jobs. This applies to all time-series data, regardless of industry or use case.
3. Content
3.1 Why It Matters
Even the most advanced modeling techniques cannot compensate for poor data quality. Incomplete histories, inconsistent units, undocumented adjustments, or misaligned timeframes frequently lead to unstable models, misleading correlations, and lost confidence from stakeholders.
A structured workflow shifts effort upstream, allowing analysts to spend more time interpreting insights and less time troubleshooting avoidable issues later in the process.
3.2 How to Apply
Follow this six-step workflow for every dataset entering Foresight:
- Assess – Understand the data context
- Identify what the metric represents and how it is calculated
- Confirm frequency, coverage, and ownership
- Verify that sufficient history exists for modeling
- Clean – Address obvious data issues
- Resolve missing values and duplicates
- Check for unit, currency, or scale inconsistencies
- Flag extreme values for further review
- Transform – Standardize structure and behavior
- Apply seasonal adjustment where appropriate
- Normalize frequency (monthly vs. quarterly)
- Apply smoothing or controls for anomalies
- Validate – Confirm statistical and business logic
- Review trend behavior visually
- Check that values fall within reasonable business ranges
- Confirm relationships make intuitive sense
- Discover – Identify external drivers
- Run Discover to test correlations with external indicators
- Review results for plausibility, not just strength
- Deploy – Finalize and document
- Save transformed indicators with clear naming
- Document assumptions, adjustments, and known limitations
- Confirm readiness before modeling
3.3 Example
A manufacturing team consolidates regional production data into a single monthly series. During preparation, they standardize units to metric tons, apply seasonal adjustment to account for annual shutdowns, and document a one-time production halt. The cleaned dataset is then used in Discover to identify energy and labor cost drivers with confidence.
3.4 Common Pitfalls
- Jumping directly into Discover or Predict without validating inputs
- Over-cleaning data and removing meaningful business variation
- Applying transformations without documenting why they were used
- Treating data preparation as a one-time activity rather than a repeatable process
3.5 Expected Results
- Fewer modeling errors and substandard Predict results
- More stable coefficients and cleaner residuals
- Faster collaboration and easier handoffs between analysts
- A transparent audit trail that supports trust and explainability
