

Some say that workforce management (WFM) will no longer be necessary in the future. Others question the possible uses of artificial intelligence (AI) in conjunction with WFM. This range shows a certain lack of orientation, which is probably simply part of the discussion about WFM. Our guest authors Ender Tezel and Ralf Thomas are convinced that WFM will continue to play an increasingly important role within any omnichannel platform and therefore in companies as a whole. This is why the question of meaningful applications for AI within WFM must be considered in a differentiated way.
It starts with the fact that in customer service, omnichannel platforms, contact center suites, etc., AI is currently almost exclusively equated with just one area: generative AI.
Since the launch of ChatGPT in November 2022, the race for the best AI models has accelerated, and contact center software vendors are outdoing each other in communicating about the development of AI integration into their solutions. This is almost exclusively about bots and therefore the underlying Large Language Models (LLMs). The benchmarks for LLMs are often outdated as soon as they are published. LLMs and applications based on them are also omnipresent in customer service and decision-makers are hesitant, at least when it comes to software without the label “with AI” – certainly also to protect themselves against supposedly wrong decisions. However, there is often no in-depth examination of this area of AI.
“LLMs are used to train AI and it should be understood that this training works better the larger the database. So far, so clear and simple. But an increasing amount of data does not necessarily mean an improvement in all models, but ultimately only a more reliable prediction of a probability of occurrence. Translated, this means With every increase in the amount of data, the answers generated by LLMs become more average. It is the nature of these models to restrict themselves to the center of a (normal) distribution and thereby cut off marginal aspects below a probability of occurrence – whether you want to admit it or not.”
“Two examples illustrate this “impressively”. If you generate images with AI, the AI accesses a huge amount of data and responds to the query: “Generate an office environment in a successful company” with an image of nine middle-aged men, all fair-skinned, and a woman in the background who is also fair-skinned. Without AI, any advertising agency could do better. Another example from practice: the rather dubious result of an AI-supported adaptation of application photos for “management position, successful woman, mid-30s”. Here, too, the AI draws on a huge amount of mainstream data, in which marginal areas are always cut off. In short: as the amount of data increases, the mainstream is solved better and better by AI. For some time, this purely mathematical phenomenon will be masked by the fact that the percentage receives less attention than the number. What is currently impressive is simply the number of cases that are better – whatever “better” means – supported by AI. Customer service cases. Das wird sich zukünftig deutlich ändern, und man muss bereits jetzt die Fragen stellen, wie wohl mit der daraus stets abnehmenden Individualisierung auf Anbieter- und Kundenseite umgegangen werden wird.“
“With these impressions back to AI in customer service and in WFM: one aspect for the use of AI in a WFM can be derived directly from what has been described here. The relatively new approach of using AI-supported forecasts as the basis for subsequent workforce scheduling seems completely obvious. But that includes this misunderstanding of the learning applicability of Al. Of course, the forecast gets better the larger the amount of data is, but firstly, it will only do so in the probability of occurrence of the assumed values in the middle range, as here too an ever-increasing proportion of “margins” is being cut off, and secondly, it must be questioned what of this really AI is and what is simply the undoubtedly impressively fast, but ultimately only stochastic analysis of gigantic amounts of data. “AI generates forecasts that actually come true” is nothing more than a popular suggestion derived from this very misunderstanding. Of course, this does not mean the other way around, that the use of AI in the creation of forecasts would be in vain. A modern WFM uses machine-learning methods, which are an important aspect of AI for years. The use of these technologies in the WFM already offers the advantage of making personnel requirements planning more precise by taking historical data and future and seasonal trends into account. This leads to an optimized use of resources and reduces overstaffing and understaffing. Implemented correctly, this is already AI in the WFM. However, AI can only really support the creation of forecasts when ever-increasing amounts of data are used to determine how to deal with side effects. Analyzing large volumes of contacts in the shortest possible time and deriving patterns and predictions from them, thus enabling fast and data-based decisions and subsequently evaluating them using AI and feeding them back into the process in a quasi-learning manner, that is what AI can and should do. “
“Anyone who has an operational function in customer service knows that a forecast never materializes. Even the smallest deviations, i.e. the flap of a butterfly’s wings, cause the same problems over and over again throughout the day and ruin the service level.And decisions are made again and again on how to deal with the deviations, which usually build up. But checking the correctness of these decisions and linking their correctness to a purpose is practically never done. Too little capacity. Does re-staffing help? Is that economical? Is it the right economic approach? Does the use of a virtual queue help? Does relocating the problem improve customer satisfaction? Is that better? Does the constant processing of callback scenarios lead to increased fluctuation? Is the countermeasure right or better? Too often, however, the question behind the question of KI in WFM is still aimed at the desire to have deployment plans or day management handled by “the AI” at the touch of a button. In order to at least come close to the answer, however, both the use and the functioning I but thought differently be, than btionally to sit on an unattainable target idea. If you look at a typical application of a KI-If you think of workforce scheduling as a modern WFM system, schedules are already created by a sophisticated and therefore complex control system. Legal and operational requirements are taken into account just as much as employees’ personal preferences. With increasingly complex requirements and demands from different areas of the company, the K I also the desire to acquire further KPIs into the planningflow to be taken into account. Questions such as “Is the processing time of an individual employee also taken into account?” or “Can the system take into account the likelihood of an employee being absent at the weekend?” are examples of how managers can use a WFM system with KI would like to see. Caution is advised here, the KI-module The more time that employees with short processing times have to spend on their work, the more time they will have to spend on it. Off-peak times, when the volume is lower, are therefore reserved for employees with longer processing times. This could be balanced out by also including quality, for example the first solution rate, as a parameter. Then it could be that an employee who takes longer but delivers better quality is not only scheduled for off-peak times. No matter how you do it, the AI will develop a bias: The “right ones” get the right times, the “less right ones” get the not-so-right times. Right?“
“A WFM system should help to make processes transparent and ensure a high level of employee experience. The scenario described would not do this justice. In addition, planning consists of many rules, and thus a KI only very limited room for maneuver. The benefits are minimal and possibly counterproductive. Simply calling for AI in WFM would not do justice to this. Beyond the forecast, where AI is used in a modern WFM, it will also be a question of making the users of WFM systems – and these are increasingly all areas in a company – with the help of AI. with the help of AI in upcoming decisions. AI is capable, not only to calculate permutations and effects in advance, but also to help assess the effects of decisions in order to then take the most expedient path at the respective fork in the road after all considerations have been made. AI in the WFM must help to make decisions and show the respective decision-maker the probabilities of the effects of the decisions. To put it colloquially: AI in WFM, properly thought through, will help to answer more steadfastly whether a decision is a good idea. Whether and for what correctness it is a good idea to remove employees from a Skill on the other or prefer one inbox to the other? KI will enable WFM to weigh up possible combinations of decisions and make them reliable based on data.“
“The KI will therefore not replace WFM managers in the foreseeable future, but will do what AI is needed for: support better decisions and at the same time always ask the question of the definition of right over and over again to the different areas of the company. There is no other way to explain the dramatically increasing importance of WFM cannot cope with much more than operational customer service for all areas of the company. The question is therefore not whether AI will be used in a WFM system, but rather how the use of AI will be implemented.I in a WFM from the manufacturer’s from the manufacturer’s point of view becomes. The restriction to the purely operational areas of customer service is far too limited. To put it bluntly: does anyone really believe that the management board of a company, which as company target, establishes a resilient connection to the degree of compliance with the service level in advance and retrospectively? But perhaps that is exactly what is wanted. KI in the WFM will be able to help with this andalso show that it is “more correct” in the overall view to follow the WFM, which declares a different decision to be the preference.“