Forecast Accuracy Collapse After a Planning Tool Migration
You are the FP&A director at a $1.2 billion consumer goods company. Six months ago, the team replaced Excel-based rolling forecasts with a cloud EPM tool. Since then, category-level forecast accuracy has deteriorated: mean absolute percentage error (MAPE) rose from 6.5% to 11.2%. The symptom log shows 38% of SKU-to-GL mappings were inherited from a legacy file with no validation; demand planners still override system forecasts in personal spreadsheets, and 14 month-end reports continue to pull from old Excel workbooks. Close now takes two extra days. The implementation partner recommends starting Phase 2-adding driver-based workforce and trade-spend planning-in eight weeks. The steering committee is split: the CIO wants to maintain momentum, the controller wants to pause, and the commercial VP says planners will resist anything that removes their spreadsheets. You have one data steward, two analysts, and a fixed budget. The next forecast cycle begins in three weeks.
1. Diagnose the most likely root causes of the accuracy decline. Distinguish tool issues from process/governance issues.
Read the worked answer
MAPE doubling from 6.5% to 11.2% is too large to blame on the forecasting engine alone. Likely process/governance causes: 38% of SKU-to-GL mappings are unvalidated, so even a good forecast lands in the wrong category; planners override in personal spreadsheets, creating version control and double maintenance; 14 reports still pull from old Excel, so month-end reconciliations absorb time and hide errors; close takes two extra days, a symptom of data validation. Tool issues could include seasonality settings, missing hierarchy logic, or poor user training, but those are secondary until master data and overrides are controlled.
Evidence to test: compare MAPE for overridden versus non-overridden SKUs; sample 50 high-value mappings and trace to GL; count active override files and their owners; compare system forecast to a naïve seasonal baseline. If the system beats overrides, the process is the problem; if not, the model needs tuning.
2. Recommend go/no-go for Phase 2 and the conditions required before go-live.
Read the worked answer
No-go on the proposed eight-week Phase 2. Adding workforce and trade-spend planning before stabilizing the core forecast would compound mapping and override problems. Conditions for go-live: (1) validate 100% of mappings for SKUs representing 80% of revenue and freeze the mapping table with change control; (2) retire or formally document the 14 Excel reports; (3) move overrides into the tool with a reason code and approver; (4) run one parallel forecast cycle and achieve category MAPE below 8.5%, with close back to baseline; (5) train planners and publish a simple override policy. If the CIO needs momentum, approve a limited pilot for one category, not full Phase 2. Revisit in one quarter with evidence.
3. Propose three controls or KPIs to prevent recurrence.
Read the worked answer
- Mapping validation rate: percent of revenue covered by validated SKU-to-GL mappings, target 100% for the top 80% of revenue, reviewed quarterly.
- Override rate and override accuracy: percent of SKU forecasts overridden, with a reason code; compare override MAPE to system MAPE-if overrides do not beat the system, remove them.
- Forecast accuracy by category against a fixed baseline (MAPE), reported monthly with a threshold: any category above 10% requires a root-cause memo. Add a close-days KPI and a report-adoption KPI (percent of month-end reports sourced from EPM). Assign owners: data steward for mappings, FP&A for accuracy, controller for close.