A Plea for Software Intelligence

Artificial intelligence has become the topic of the hour in 2023. Since ChatGPT and GPT-4, boards have been discussing language models, and McKinsey and Goldman Sachs have both published forecasts this year that credit generative AI with substantial productivity gains. At the same time, only a small share of companies use artificial intelligence at all, and where it is used, it usually remains a tool at the margins. This article is a plea to regard artificial intelligence not as a product but as software intelligence: as a property of the entire enterprise software that changes decisions, organizational forms, work, and the use of resources.

Why I speak of software intelligence

This year's forecasts share one feature that gets lost in the coverage. In June 2023, the McKinsey Global Institute estimated the economic potential of generative AI across business functions; in March 2023, Goldman Sachs assessed its effect on growth and labor markets. The specific figures differ, and I consider it pointless to weigh them against each other. Both studies, however, rest on the same condition: the gains do not come from employees opening a chat window, but from the technology being built into processes, data, and decisions. As early as 2020, the European Parliament, in an overview of the opportunities and risks of artificial intelligence, cited a possible increase in labor productivity of 11 to 37 percent by 2035, likewise assuming broad use across the entire economy.

That is exactly why I prefer to speak of software intelligence rather than artificial intelligence. A language model or a forecasting model is, on its own, only a building block. A company becomes intelligent in the business sense only when its software learns as a whole: when data from sales, production, procurement, and finance converge in a connected system, algorithms decide on that basis, and the results of those decisions flow back into the system. In my book "Das selbstfahrende Unternehmen" (Springer Gabler 2021, in German), I put it this way: the enormous lever does not come from faster processors or from artificial intelligence applied in isolated spots, but from the fact that, for the first time, everything is fully integrated. What this software must deliver in order to carry decisions, I described in the article Cognitive Software Is More Than Artificial Intelligence. This article is about the consequences for the company.

Software intelligence is not a technology you buy but an organizational principle you introduce.

Decisions: errors are the raw material of learning

For learning algorithms, "making decisions" and "making mistakes" are inseparable. A learning algorithm processes large amounts of data, follows its programmed sequence, and at first frequently makes wrong decisions. Humans or other algorithms detect these errors and report them back. The algorithm corrects its decision parameters, and the changed configuration is already in use on the next run. The same error does not happen twice. This feedback cycle is the basic principle of every learning system and explains why such systems must be trained before they are entrusted with anything.

The comparison with the hierarchical organization is sobering. Anyone who has worked for some time in a larger organization divided into departments knows the pattern: each department develops its own logic, information is withheld, errors are covered up or passed on in distorted form. Such errors often run through the company for years and only come to light when the people responsible have long since left. In a connected software landscape, by contrast, an error becomes visible where it arises, and the correction takes effect immediately on all subsequent decisions.

Add to this what distinguishes software from people: it processes many times more data, works faster and more precisely, does not tire, and knows neither weekends nor night shifts. The next step, which I described in the book, is already visible: individual learning algorithms are trained on a company's special cases and begin to interact with one another on the basis of shared data. The result is a decision network that runs through the company. Which decisions software makes better than people within this network, and where people remain indispensable, I set out in the article How Self-Driving Companies Decide.

Rethinking organizational forms

Companies face growing demands from outside and inside: changing market conditions, new technologies, competition from startups, individual wishes of customers, and a shortage of skilled staff. The disruptive events of the past three years, pandemic, war, and energy crisis, have further increased the pace of change. Classic, strictly hierarchical organizational forms cannot respond at the required speed. Their departments are poorly connected with one another, every decision passes through several levels, and the potential of agile ways of working gets stuck at departmental boundaries.

Software intelligence changes the starting position. When routine decisions and coordination are taken over by software, the hierarchy loses its most important reason for existing: the distribution and control of tasks. It is replaced by teams with a broad mix of expertise that make decisions decentrally and against the background of clear shared goals. They complete tasks faster than departments because information and decision-making authority sit in the same place, and they pull in the same direction because the shared goal is visible to everyone. The structure behind this, from the vertical product cut to leadership through objectives, I described in the article The Organizational Structure for True Agility.

Beyond that, I expect software intelligence to produce organizational forms we do not yet know today. Digitalization, with the online startups, created a generation of companies that achieved very large reach without branch networks, without classic sales forces, and with very small teams. The startups that have been emerging since the arrival of large language models go one step further: they build their value creation around learning software from the outset and staff functions that established companies cover with entire departments with a handful of people. Established companies will have to measure themselves against this benchmark.

Shifting activities: from reports to competence

When software makes decisions and handles routines, human capital shifts to a different level. The repetitive activities, merging spreadsheets, preparing reports, searching for receipts, documenting processes for quality management, are taken over by software. Employees concentrate on their actual competence: on customers, on products, on problems that demand creativity and empathy. According to the thesis of my book, middle management, which spends its time preparing figures, will largely be replaced within the next five to ten years by algorithms that perform these tasks faster and better.

Leadership remains but changes its character. Teams are led by leaders who motivate through authentic communication of their vision, not through instructions and control. The "why" is decisive for motivation: why do we pursue this goal, why is this task important? Those who know the why identify with the goals, and that feeds back into performance.

With the shift to creative and interpersonal activities, it also matters less when and where work is performed. Rigid working-time models, long outdated from the perspective of motivational psychology, break open, and everyone can work according to their own performance curve. In the book, I described the observation that part-time staff in most professions deliver about 80 percent of the workload of full-time staff in half a day because they manage their time better. Creativity rarely arises under time pressure in a cramped office; it usually arises when, after intensive engagement, some distance from the problem emerges.

Using resources more sustainably

The productivity gain from software intelligence arrives at a time when sustainability requirements are rising, and these requirements concern not only ecological aspects but all resources deployed.

• Human capital. The knowledge and skills of employees are deployed where they make the greatest difference. While people fail at the flood of information, software creates value from the same volumes of data, and it performs repetitive tasks with fewer errors than people can under time pressure.
• Production resources. Demand-driven planning, optimized lead times, and make-to-order production reduce inventories and waste. Robotics and learning control systems realize what previously failed in planning for lack of data.
• Business processes. The handling of orders, invoices, applications, and approvals is taken over by software; people intervene only in exceptional cases. This shortens lead times and reduces the effort spent on control.
• Demographics. Demographic change will leave us with significantly fewer workers by 2035. The productivity gain is therefore not merely a matter of maximizing profit but the precondition for maintaining prosperity with fewer people.

Companies can thus operate more successfully and more sustainably at the same time. The efficiency gain affects profit and at the same time reduces the impact on the environment and climate.

Where we stand: data before models

How far is the German-speaking region from this? According to a 2022 report in Handelsblatt, around one in ten companies in Germany used artificial intelligence. Interest has surged since then; broad adoption has not. In the conversations we are having this year, the topic is almost always models: which language model, which cloud, which vendor. That is the wrong first question.

Today's state of the art permits the widespread use of software intelligence, on one condition: the data must be available in machine-readable form and be connected. The path to the self-driving company therefore does not begin with buying a model but with saying goodbye to paper and connecting previously isolated data silos. In the book, I defined the target as follows: one can speak of a self-driving company when about 80 percent of business functions and decisions are automated and controlled by sufficiently intelligent algorithms. The four fields of action in this article can be measured against that.

In this company, people take on a more important role than ever before. They are freed from spreadsheet jobs, from heavy lifting in production, from searching for misplaced receipts, and from tedious documentation. Theirs is a creative, empathetic role that moves the company forward on the basis of big goals. That is the core of my plea: software intelligence is not a cost-cutting program but the condition for people in the company to do what only people can do.

The next step

How software intelligence changes decisions, organization, value creation, and staffing in the company of the year 2035 is described in my book "Das selbstfahrende Unternehmen" (Springer Gabler 2021, in German): About the book. If you would like to know how far your data and systems are today from connected use, we would be glad to discuss it: Book an expert meeting.