When Forecasting Fails in Retail
Accurate demand forecasting is essential in retail for both generating revenue and minimizing costs. Assortment planning, inventory management, pricing, and logistics all benefit significantly from the ability to estimate trends as accurately as possible. Demand trends are rarely stable or purely seasonal; instead, they are shaped by a multitude of non-linear factors. Promotions, price changes, external events, or rapidly shifting customer preferences lead to patterns that cannot be adequately captured using traditional, predominantly linear methods.
Against this backdrop, the use of data-driven methods in forecasting is gaining importance. They promise to better identify complex relationships and patterns, thereby creating a sound foundation for operational decisions. In practice, new tools and solutions are often introduced with the expectation that they will largely compensate for existing weaknesses in planning and control—an expectation that is rarely met.
Why Projects Rarely Fail Because of Technology
The implementation of modern forecasting solutions comes with a clear expectation: advanced algorithms, improved data processing, and automated processes are supposed to lead to more accurate forecasts and more efficient management.
In reality, however, projects are delayed, results fall short of expectations, and the hoped-for impact on the P&L fails to materialize. The reason for this failure rarely lies with the technology—it is, once again, a combination of unrealistic expectations, an inadequate data foundation, and organizational shortcomings.
One problem is that the actual challenges—which already existed with traditional methods—are not properly addressed. Unclear targets or undefined decision-making processes persist under the assumption that the system will compensate for them.
As a result, projects are launched hastily without a clear vision. It remains unclear what specific improvements are to be achieved. Instead of evaluating the actual operational impact, metrics such as forecast accuracy are often relied upon. While these provide a formal assessment, they say little about whether decisions are actually improving.
Data Quality as a structural problem
In addition to expectations, the data foundation is one of the key weaknesses of forecasting initiatives. In practice, data quality is often viewed as an IT issue—or even dismissed as such—and is not addressed by business departments. As a result, data may be collected inconsistently, maintained incompletely, or updated inconsistently. Relevant influencing factors are missing or not clearly defined.
Since data forms the basis for forecasts, its quality determines the quality of the forecasts and the decisions based on them. A forecast is therefore only as good as the reality reflected in the data. If this foundation is missing, precision does not lead to a better decision, but only to a supposedly precise yet incorrect decision.
Systems recognize patterns only based on what they “see.” If the data foundation is incomplete or distorted, the result is not merely inaccurate forecasts, but systematically incorrect ones. Demand is over- or underestimated, trends are misinterpreted, and decisions do not lead to the desired outcomes.
This becomes evident with factors such as promotions or one-time events. If such factors are not accurately captured, they appear in the system as regular demand. Short-term fluctuations become supposed trends with direct impacts on inventory, availability, and markdowns.
The Organization as a Limiting Factor and a Solution
Even if the ambitions of initiatives are realistic and the data foundation is robust, the success of new solutions—as is so often the case—stands or falls with the organization. Ownership and responsibilities are unclear, and the necessary management support is lacking.
Yet the first step is relatively simple: clearly assigning responsibilities. There needs to be a person or role that takes responsibility for the relevant data. In addition to traditional master data leads, role-specific positions such as procurement data leads or assortment data leads are also conceivable. They ensure that the correct data is available in the required quality. Ideally, in addition to their subject-matter expertise, they also possess technical understanding, so that business and IT do not operate in isolation and the user perspective is consistently taken into account.
In addition to organizational adjustments, however, process alignment is also required, which is significantly more complex to implement. Forecasting is not introduced as an additional system, but rather goes hand in hand with an adjustment of the underlying decision-making logic. Impact is only achieved when forecasts are systematically embedded in decisions.
