Data-Driven Resource Allocation and Forecasting in a Marketing Agency
Makkonen, Antti (2026)
2026
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202603315387
https://urn.fi/URN:NBN:fi:amk-202603315387
Tiivistelmä
This study investigated how data-driven forecasting could be used in a marketing agency context where project scopes and effort estimates had traditionally been based on intuition and expert judgement. The case company provided access to a dataset with almost 3000 projects completed over a nine-year period, allowing a comprehensive analysis of historical project data. The main goal was to determine if machine learning models could improve the accuracy of resource allocation decisions compared to traditional methods.
The research followed a pragmatist approach and focused on addressing a practical organisational problem using real project data. Quantitative analysis of historical project records was combined with contextual insights from organisational stakeholders to develop and evaluate machine learning models for predicting project effort. Model performance was assessed using standard error metrics, including mean absolute error, and compared with historical estimates produced by project managers. To reflect realistic forecasting conditions, the dataset was divided temporally so that models were evaluated using data from later periods.
The results indicated that while data-driven models were able to improve estimation accuracy compared to intuition-based methods, the R² of human estimates was still higher than that of the machine learning models indicating the continued importance of expert judgement. The study found a hybrid approach
combining algorithmic predictions with professional judgement to be the most effective, supporting more balanced resource allocation and improved operational performance.
The research followed a pragmatist approach and focused on addressing a practical organisational problem using real project data. Quantitative analysis of historical project records was combined with contextual insights from organisational stakeholders to develop and evaluate machine learning models for predicting project effort. Model performance was assessed using standard error metrics, including mean absolute error, and compared with historical estimates produced by project managers. To reflect realistic forecasting conditions, the dataset was divided temporally so that models were evaluated using data from later periods.
The results indicated that while data-driven models were able to improve estimation accuracy compared to intuition-based methods, the R² of human estimates was still higher than that of the machine learning models indicating the continued importance of expert judgement. The study found a hybrid approach
combining algorithmic predictions with professional judgement to be the most effective, supporting more balanced resource allocation and improved operational performance.