Predictive Analytics and the Planning Approach in Quintet24

Predictive Analytics and the Planning Approach in Quintet 24

Planning systems and reports are hampered by the time it takes to create them and the lack of transparency in the various relevant dimensions. MobiMedia optimises planning effort and decision transparency by proactively integrating all the planning content into the process. Through intelligent and user-friendly visualisation, colour themes, emotions and the possible POS areas can be planned interactively and visually as well as quantities and values. Through data mining, MobiMedia automatically integrates all measurable influences into the planning and significantly reduces manual effort.

But how can these complex amounts of data be mastered?

Simply put, advanced analytical methods allow patterns to be detected in extensive data, which can then be used for decision making processes and forecasts.

The strategic crystal ball
What will the customer buy tomorrow and why? Only if you know what will be bought tomorrow can the inventory be kept as low as possible and yet still meet the customers’ needs. Efficient and cost-effective planning requires a precise forecast of the expected buying behaviour and the corresponding sales. Special events such as advertising campaigns, major sporting events, the weather, the start of vacations, public holidays and so on and so forth are already included in many forecast models today and seasonal fluctuations in sales are also the norm for good planning software.

Predictive planning versus classic approach or:
What do advanced analytical methods do differently?
In contrast to classic, descriptive or visual analytical techniques and the subsequent manual processing of the knowledge for extrapolations, advanced, model-based approaches are now used in predictive planning and forecasting. Correlations between the variables are identified on the basis of mathematical data analysis models in order to derive knowledge about existing patterns as well as new forecasts.

These advanced methods can recognise patterns and correlations in extensive and heterogeneous data sets and can thus be used for forecasts. There are many advantages to using predictive planning and forecasting.

The aim of using automated forecasts is to optimally support and relieve human planners of burdensome routine tasks, not to replace them – for example by supplying suggested values to be used as part of planning. Planners thus remain involved in planning processes, but are relieved of routine tasks. This eliminates key points of criticism about today’s planning processes, such as the poor quality of the planning results, the effort invested, the excessively long lead times of the planning process or the lack of resources for short-term forecasts.

Solid forecasts in shorter intervals with correct data quality

Our digitalised world has created a completely new type of consumer: a buying nomad who wants to quickly recover the respective trends from the shopping cart. A digitally networked buyer, who wanders from shop to shop, brand to brand, always seeking the best offer. It is increasingly important to recognise these trends early on and to inspire customers with the right offer as soon as possible:

With the new approaches, planning cycles can now be carried out much faster and more frequently. It is also easier to integrate sub plans and ensure they are linked to analyses and reporting. AI is used in predictive planning methods, but also in algorithms for project monitoring or early detection of deviations.  This enables companies to react more rapidly and sooner. Which saves time and money. The quality factor of the results is significantly improved and creates competitive advantages.

Robust forecasts can only be based on the right data in the right quality, the required granularity and sufficient history. Predictive models are designed for a specific application and are continuously trained, which is why they are not generally applicable. Since, among other things, markets, competitors and customer behavior are constantly in flux, it is inevitable that the model behavior and the quality of the forecast must be continuously monitored in order to make adjustments or readjustments, depending on the result.

Deep learning as a building block for robust forecasts
As the name implies: deep learning is based on the analysis of large amounts of data.
The machine literally “digs” itself through huge amounts of information from databases, e-mails, social media data and consumer purchases in order to identify small and large trends. Deep learning uses artificial intelligence and well-known neural networks to create systems that can simulate human learning behaviour using data, multi-layer algorithms and software.

Deep learning requires a high-performance specialised IT infrastructure for the analysis of huge data sets and above all for learning in real time. This also means that the underlying deep learning memory is sufficiently agile to be able to process different types of data quickly. This flexibility is one of the reasons that software-defined storage is now at the top of most companies’ wish lists.

Performance meets forecast
Nevertheless, the process cannot supply missing information. Nor can sudden changes or first-time phenomena be recognised and included in forecasts, because pattern recognition processes have their limitations: the basic operations of pattern recognition, such as data filtering or differentiation, are based on statistical, cultural, political and many other assumptions. They specify what is ultimately recognised as a pattern and what is not. The software recognises what it is trained on – an algorithm cannot “see”.

Despite all the advantages of predictive analytics and planning, a forward-looking, entrepreneurial intuition and careful monitoring by experts are still indispensable. This is, among other things, because unlike a trained expert, an algorithm cannot assess whether the correlation between two variables is purely coincidental or arises because of an indirect relationship.

My reading tip:
Predictive analytics also has its downsides in society. I recommend the following on this topic:
Hito Steyerl et al.:  Pattern Discrimination. University of Minnesota Press and Meson Press, 2018

Autor: Connie Rambold, CCO MobiMedia AG