ibs Analytics embeds a revolutionary concept within its forecast engine. It provides a "meta-engine" approach allowing fitting any situation with the best accuracy.

“every item is different and has its own profile. it theoretically requires a specific modeling”

The meta-engine approach adds a layer in the process that automatically determines the best model to be applied to any item. Each item is then modeled picking the right approach in a library of models, including:

• Soft Solutions owns ADDM modeling approach for:
  • Seasonal items
  • Planogram items
  • Introduced Items
• Standards of the information mining & signal processing, redesigned by Soft Solutions to match requirements of the retail (ARIMA, SARIMA, exponential smoothing, moving average…)

Final benefits observed on real conditions and heterogeneous sets of items / stores has shown an increase of weekly accuracy of 2.5% at the item level without any performance drawback. Moreover, performances have been optimized, using simple models for simple situations and reserve computation time for more complex item sales profiles.

THE BUSINESS WITHIN THE SCIENCE

ibs Analytics emeta-engine keeps in parallel the will to take benefits of any business knowledge and stay transparent to users.

The first step of the meta-engine process is the items clustering. This one is exposed to the business users in term of graphical profiles. They can track results by profile and validate that those profiles are consistent. They can also provide business definition to clusters (seasonal, weather dependant, pending discontinued, newly introduced…) to retrieve results quickly.

“every item is different and has its own profile. it theoretically requires a specific modeling”

For each pattern, they can then check which forecast model has been selected automatically by ibs Analytics engine, visualize expected accuracy and even test other models and update them manually if required.

This approach combining the top of the science and the feedback of the business into a single process actually increase both accuracy and efficiency of business strategies.

UNDERLYING CONCEPTS

In order to implement this meta-engine approach and provide better accuracy with better performances, ibs Analytics research laboratory has focused on three main steps to achieve ready for production engine.

• Data Sampling. In order to test each model over clusters and determine the latter, the first step was to limit the amount of data (production databases usually deployed in production handles 1 to 100 billions sales records), ibs Analytics includes a sampling step that is coupled to parameters estimators to determine the minimum amount of information required to get a consistent sample of the whole dataset. When facing performances issues, the size of the sample can also be determined based on allocated computation time.
• Clustering. With the same performance reasons for data sampling and, moreover, for transparency and efficiency reasons, the next step is a clustering for items profiles. This achievement is based on both attributes and sales patterns. It is based on X-Means clustering approach (various others are also implemented and ready for use - such as K-Means, HClust…)
• Estimation.Finally, each cluster is tested for all the models available in the library over the past weeks (usually a period of 6 weeks is reserved for this purpose) and the best approach (in term of accuracy - weekly or consolidated) is defined to be used for any item belonging to the cluster.

SERVICE-BASED ARCHITECTURE FOR FAST IMPROVEMENTS

The ibs Analytics meta-forecast engine has been thought as an evolving engine since it was designed and based on a library of forecast models, implemented as services.

Depending on the clients, the situation, the market, the technical layers, budget, different services can be deployed in order to match closely the need of our clients.

Moreover, ibs Analytics research lab can integrate new models in production environment with a complete control of the risks. Clients can include their own approaches to ensure continuity in their business.

With this meta-engine layer, ibs Analytics has projected forecast processes into a new dimension, providing the best of science ready for production and tier 1 retailer's scope.