The Problem
Across 25+ digital media client campaigns at Stele Media, decision-makers lacked timely, structured access to revenue and performance data. Revenue patterns were buried in disconnected spreadsheets, reporting was reactive, and planning cycles were driven by intuition rather than evidence.
The core issue was a visibility problem — teams were working hard but had no reliable model to forecast what was coming or understand what was actually driving performance. Every strategic conversation started with the same question: what does the data actually say?
The Approach
The goal was to build a robust forecasting and visibility system — combining regression modelling in Python with structured KPI dashboards — to surface revenue drivers and enable confident planning decisions.
- Exploratory data analysis (EDA) — performed structured EDA using Python (Pandas, NumPy) to identify patterns, anomalies, and key revenue drivers across historical campaign data.
- Regression modelling — built regression models to quantify the relationship between campaign inputs and revenue outputs, enabling forward-looking performance forecasts.
- Visual insight outputs — generated charts and visualizations translating model outputs into clear, stakeholder-ready insight for business planning and monetization strategy.
- KPI dashboard design — standardized reporting workflows into Power BI dashboards with automated data refresh, replacing manual ad-hoc report generation.
- Review cycle embedding — implemented monthly and quarterly structured review cadences to shift decision-making from reactive to forward-looking.
Outcomes
The forecasting model and reporting system delivered measurable improvements in visibility, efficiency, and decision quality. Performance visibility improved by approximately 30%, ad-hoc reporting requests dropped by ~40%, and campaign outcomes improved by 15–25% across engagements where the system was fully embedded.
Forecasting outputs directly informed business planning and monetization strategy — improving revenue projection accuracy and giving leadership the confidence to make faster, evidence-based decisions rather than waiting for post-campaign analysis.
Key Learnings
- EDA before modelling is non-negotiable — understanding the shape and quality of your data determines whether your regression outputs are trustworthy or misleading.
- Forecasting models are most valuable when embedded in decision workflows, not used as one-off analyses. Recurring use compounds their impact.
- Reducing ad-hoc reporting requests is a leading indicator of trust in the system — when people stop asking for one-off reports, the dashboards are answering the right questions.
- Consistency beats sophistication — a simple, well-maintained model reviewed weekly outperforms a complex model reviewed quarterly.