WHAT MAKES ANALYTICS INITIATIVES SUCCESSFUL?
Translating actual process problems into the analytics world, building horizontal data competencies and the ability to scale and industrialise use cases are critical for sustainable success and broad impact of AI initiatives. Interview mit Ulrich Krieg, Partner, ROI-EFESO
DIALOG: Mr. Krieg, how do you experience the entry into the topic of data analytics and AI in practice? In which areas is it gaining importance?
UK: At the moment, we often see the topic in SCM projects, for example when it comes to optimising sales planning. In this environment - predictive demand planning, demand sensing - the intelligent use of data offers great potential. The aim is to combine different data reservoirs and to generate more precise demand forecasts from past, order and environmental data. If it is possible to use all the data that is available, demand patterns can be better forecast, resources can be better secured and inventories and delivery times can be optimised.
DIALOG: How critical is data availability in this context?
UK: Often there is more data available than you think, you can actually always find a start. It becomes interesting when, on the one hand, you include data that is not yet available in structured form, for example from communication processes, CRM systems, from the point of sales, and on the other hand also external macroeconomic data and forecasts or industry indices. We often discuss the question of which statistical data can be used in projects. Integrating such different data would be extremely time-consuming in the conventional world. In the meantime, however, very good tools are available that make working with restructured data enormously easier.
DIALOG: With data analytics, you enter an area in which the majority of companies have little experience to build on. How do you ensure that the entry is successful?
UK: There are several factors. If we stay with the example of demand planning, the first questions are very classic. Because regardless of new analytics possibilities, a robust backbone for planning processes is needed first. The basis must be right and correspond to established methods and best practices. Then AI tools and analytics solutions can build on it. And then it is a matter of cutting the projects as small as possible and tackling the topic in small steps, in sprints. This is about proof of concept, about trying things out and also discarding them in order to create empirical values. Of course, this also requires a certain mindset in the organisation: the willingness to deal with a topic where the result is not completely predictable, where you don’t know which approach will work. In many use cases, you find that you can’t learn anything from the compiled data, that the algorithm can’t do anything with the result. Then you just have to move on.
DIALOG: What remains after a failed sprint?
UK: Lessons learned. You learn from each attempt what works and what doesn’t, what you have to pay attention to, which sub-areas have potential. You can act more precisely in the next attempt. You gain experience and create the basis for future projects - that is also part of dealing with new issues. The art lies in designing these learning loops efficiently. In our analytics projects, it takes an average of six weeks for the team to achieve resilient results.
DIALOG: How should a project team be composed in order to increase the likelihood of success of analytics initiatives?
UK: If we stay with the planning example, then the representatives of the affected areas must be involved - sales controlling, marketing, product management, supply chain management, IT. They are supported by data experts. It is crucial that the two groups work together effectively. In analytics projects in the area of production, for example, it is important that so-called Citizen Data Scientists are involved, process engineers from the specialist departments who have basic qualifications in the analytics topic. On the one hand, they know their own processes exactly and on the other hand, they can understand the analytics experts. If you focus too much on data expertise, you approach the topic too technically, analyse large amounts of data and often end up with findings that are trivial. You absolutely need the process expertise to focus on the right question from the outset and on pre-selecting the data that could have an impact. Trial and error does not mean looking for a needle in a haystack.
So, the art lies in translating the actual process problem into the analytics world. Building this bridge between the process world and the data world is a key task and it is important to also use the projects to strengthen the horizontal data competence in the company. The algorithms, the technical implementation - there are more and more standardised tools and services on the market for this. That is not the hurdle at which initiatives fail.
DIALOG: What follows a successful analytics project? What needs to be done to build on what has been achieved?
UK: Above all, it is important to industrialise the topic of analytics. Setting up an algorithm once and getting it to work is a start, but not a solution. Because the process landscape continues to develop, the process parameters change. Therefore, you must train and change the models again and again. This requires a defined process, professional change management and an organisation that manages and follows up. Unfortunately, the opposite can be observed again and again. A big analytics initiative is launched, numerous use cases are identified, and some of them are tackled. But the overall framework is missing - and then there is no traction in the organisation, the project plans in the individual areas peter out, competences are not built up. The result is then a one-off, resource-intensive project with no long-term impact. It is therefore necessary to think about the organisation from the beginning, develop clear governance, demonstrate the benefits on the ground and initiate a continuous improvement process. Successful initial projects are also an opportunity to bring about a change in thinking so that a self-sustaining process emerges. The topic - like any change topic - is partly risky and thankless. You have to change roles and responsibilities, the way of working together, which is often unpleasant. And that is why management’s far-sightedness, responsibility and leadership are particularly in demand.
