Algorithms can be used to optimise consulting services or to quickly analyse data. Companies and employees are dependent on algorithms, but should not rely on them blindly as algorithms also make mistakes - and it is human beings who are then to blame.
Algorithms in B2B:
A clever business idea and powerful algorithms are enough to upset an entire industry. The Unternehmen AppZen, for example, uses algorithms to automate and process tedious tasks such as analysing accounting data faster than employees could. During this process, the AppZen algorithm checks different data at the same time: are the details of addresses and means of payment correct? Are all of the timings plausible and do the invoicing parties actually exist? The algorithm finds out. Algorithms are instructions that computers use to solve problems. They work deterministically and always achieve the same result with the same data and conditions. The individual steps that an algorithm goes through are clearly defined and work towards a clear end. AppZen analyses expense reports worth millions of dollars each year and claims that its service has saved a single major customer approximately $40 million. Curious cases that AppZen has uncovered: Invoices for steaks worth hundreds of dollars, visits to the vet for an employee’s dog, and hotel bills for spouses who travelled with the employee. As human auditors previously only inspected receipts at random, such cases often went undetected.
German SMEs could be better positioned
Algorithms are also trained in how to deal with people. The Unternehmen Precire from Aachen uses language analysis to determine the personality profile of applicants and helps HR departments to find suitable candidates for advertised positions. The evaluation runs fully automatically using an algorithm that gives HR managers recommendations on the various candidates. Algorithmen großes PotenzialIn retail, too, expertise, patience and clever suggestions are what customers want from a good salesperson. These competencies can easily be mapped using algorithms. It no longer matters whether the digital salesperson then advises a builder on the right mortar or a company on server solutions.
SMEs in Germany face various issues: Algorithmen vor Herausforderungen Suitable experts in working with data and algorithms are rare and data protection regulations are unclear when it comes to the use of the data collected by companies. Moreover, the discourse surrounding technology is determined by the fear that algorithms could destroy jobs and make people redundant. Too seldom do people look at the Achilles’ heel of every code: how good it is depends on how well people have written it. That’s why algorithms do everything right – even mistakes.
Algorithms need guidance
“Computers and algorithms are always superior to humans when it comes to analysing large quantities of data and deriving correlations from this,” explains Martin Kleinsteuber, chief information officer at the Mercateo Group. “One area in which humans cannot be beaten by algorithms in the foreseeable future is their ability to make decisions with the help of these derivatives from the algorithms,” he says. In the financial sector, for example, a Finnish company was convicted Bank wegen Diskriminierung because its algorithm had denied a businessman a loan. The person interested in a corporate loan had stated Finnish as his native language, which was an exclusion criterion for the algorithm used. A similar case occurred in the United Kingdom: Here, the Tageszeitung The Sun and the BBC eine Versicherungsangebote. Fictitious customer details were provided for the quotation. “John” and “Mohammed”, who, apart from their names, had provided identical information regarding place of residence and income, each asked for a quotation. In the test, the insurance offered for the fictitious customer “Mohammed” was up to £1000 more expensive than for the fictitious “John”. The question that critics then raised: why does the algorithm consider the customer’s name to be an essential criterion for insurance? The reason for the decision can only lie in the source code for the algorithm. It was programmed by humans. Humans are also the ones who have to help algorithms correct the mistakes made by humans. They have to decide whether connections are really plausible or just incidental. There does not have to be a connection between two factors just because they occur together, as in the example of “John” and “Mohammed”. The accusation was quickly made that the insurance companies’ calculations had nothing to do with probabilities but were simply racist and that the algorithms had been deliberately programmed to be unjust.
Algorithms can only process what humans have taught them
Critics are therefore sceptical of the increasing use of algorithms Einsatz von Algorithmen and demand transparency about the way algorithms work. “Algorithms can only handle the data that is available to them,” explains Martin Kleinsteuber. Among other things, his team optimises the search algorithms on the B2B procurement platform Mercateo and helps users find the item that is right for them from the data volume of over 20 million items. He is acutely aware that das Problem algorithms can no longer work correctly if information is missing. “That’s why it’s still difficult for computers and algorithms to beat human opponents in some games at the moment. Especially when it comes to emotions and psychology. Algorithms would have to be able to take the facial expressions of other players or their brainwaves into account in their calculations and derive actions from the information. This is not yet possible at the moment.”
“Algorithms should only prepare decisions” - Kai Nowosel
Kai Nowosel, chief procurement officer at the management consulting firm Accenture, also believes that this is why people will be supported by algorithms rather than replaced by them in the future. He is a specialist in B2B procurement and has a clear opinion on algorithms: “Algorithms can prepare decisions. After all, an algorithm can only function within the framework of what it has been taught. There may be dependencies within decisions that are not known to the algorithm, but which make it necessary to choose another supplier. These could be business dependencies, these could be agreements or political constraints that are not known to the algorithm. I would therefore always speak of decision preparation, not of decision-making power.”
The bottom line, says Nowosel, is that “people are needed to make decisions. Just as automation requires people to train these machines. There is no process that I can hand over in which the computer teaches itself. After all, it must understand connections.” Even at AppZen, expense accounts cannot be checked without the intervention of human auditors: once the system identified an expense claim for a snake as incorrect. Only after detailed examination by a human colleague did it turn out that Schlange zu recht gekauft it had been purchased as a mascot for a project.
Who writes here?
My name is Sebastian Prill and I work as an editor at Mercateo. Digital topics fascinate me, especially how data can contribute to value creation. I find it exciting how big data makes hidden things visible in business and journalism and how digitalisation is changing everyday life.