The key figures illusion - what is actually the "key" of key performance indicators (KPI)?

Artificial intelligence and machine learning offer new opportunities to fundamentally improve corporate management and controlling. Are we currently working with yesterday's methods to solve tomorrow's problems?

Digitization gap studies repeatedly indicate that the control and KPI systems in many companies have fundamental problems:

• Insufficient database: control systems are still based on too much financial information and far too little external information.

• Inadequate effect: KPIs are sometimes selected rather randomly. At the same time, the key figure systems lack connections between the individual indicators.

• Inadequate strategic alignment. Successful management requires key figure systems that map the company's business model or support the innovation of business models.

On the other hand, the technological advances in recent years have been breathtaking. This applies not only to the acquisition and development of completely new data sources, but also to the possibilities of storing huge amounts of data uncompromised and analyzing them for regular patterns and dependencies to understand cause-effect relationships, derive measures and make predictions or carry out simulations. However, such approaches require the use of new analysis methods and above all, a radically different view of data and decision models. In addition to greater automation the integration of more advanced analysis methods based on machine learning and artificial intelligence plays a central role.

Inadequate suitability of traditional analytical methods

Using the classic approach would be to try to manually identify relationships by means of descriptive analyzes, to present them and to interpret them in order to derive recommendations. BI systems offer a wide range of options for this. The potential problems with this approach, which is widespread today, are evident. With descriptive analysis, typical for classical BI systems, there is a great deal of leeway for misinterpretation. This also depends to a large extent on who carries out these manual analyzes.

On the other hand, models that provide the following information are worth looking at

• Root-Cause-Analytics: Multivariate cause-effect models, including the mapping of non-linear relationships and indirect effects
• Predictive Analytics: Prediction options based on these cause-and-effect models
• Prescriptive Analytics: Automatic how-to-achieve analyzes of how a specific goal can be achieved

Machine learning methods, such as ANN or GBM algorithms, enable the extraction of such models on the basis of large, uncompressed data spaces. In the sense of Explainable AI, selected model parameters and results can also be displayed transparently and thus made more usable for the business. This combination of ML / AI and BI, so-called AI enabled BI, enables a completely different approach to corporate management.

Implications for corporate management / controlling - building up analytical methodological skills.

AI / ML offers excellent opportunities to fundamentally improve corporate management / controlling, increase the value contribution for the company and redefine the role of data in the company. Here are few examples:

Use of new, uncompressed data as a basis for learning new KPIs in discovery.

The focus should not only be on optimizing existing target values, but rather on learning what actually needs to be optimized and how. It's not about simple if-then rules. AI / ML enables the identification of complex relationships on the basis of uncompressed data. In this way, for example, not only direct / indirect effects can be identified and quantified. On top of this, for example, the regional or customer group-specific validity of these effects. On the basis of AI / ML, new target values ​​that are important for the strategic direction of the company can be identified and new KPIs proposed.

Identification of (fake) KPIs and cleaning up the existing KPI jungle.

AI / ML enables the identification of cause / effect relationships including their quantification. In this way, previous (probable) relationships and their drivers can be validated. Useless KPIs can be cleaned up or modified if necessary.

Use of KPIs as a data basis for machine learning.

A major potential of AI / ML is to continuously learn from the data. Especially in an increasingly dynamic competitive environment, this enables control processes to be set up in such a way that the learned drivers and their effects are set as the basis for measures and the success of these measures can be tracked automatically.

Use the possibilities of AI / ML.

Not only new data sources (width), but also the granularity (depth), or the level of detail of this data, are of decisive importance. In the case of compressed data, important patterns and possibly causal drivers cannot be correctly recognized by AI / ML. The conversion and generation of new data from raw data (feature selection / engineering) also play an important role. KPIs as a separate class of data assets can play a central role as input for AI / ML.

This is a potential threshold moment for business and industry, where machine learning might weave its way further into how operations are handled, the way decisions are made, and resources get managed. It will depend on whether or not businesses collectively find real value in AI. And last, but not least the investment in the technology must prove its worth.

Related topics: Business, News: Gartner’s Top Strategic Technology Trends for 2021