Are you feeling anxious about AI again?

COMMENT. It has now been three years since the initial AI hype. AI is more powerful and widespread than ever, but how has it changed jobs and productivity in the financial sector?

Recently, that moment came around again. During an extremely impressive presentation of a new AI tool by a major data provider, I found myself wondering how much longer my role as an analyst in the financial sector would last. The tool was able to precisely aggregate data from a multitude of sources, statistically analyse it and interpret it correctly – better than any models I had tested before. Will I still be needed soon?

I immediately thought back to the spring of 2023, when ChatGPT first went viral as a language model and fuelled the narrative of widespread job losses. But what happened to those fears we had back then? What's happened since then in the field of AI, its use in the financial sector and in terms of its impact on jobs and productivity?

Rapid technological progress

Since spring 2023, AI has developed rapidly. Generative AI and language models, in particular, are now capable of significantly more than they were just a few years ago. The performance gains in recent years are attributed to the massive expansion of computing power and algorithmic advances. According to Epoch AI, the computational effort required to achieve a certain level of performance has already halved roughly every eight months.¹ In 2026, the best AI models were able to significantly outperform human performance in ‘Humanity’s Last Exam’, a test that is particularly difficult for AI but easily solvable. What was originally intended to be a challenge spanning several years was achieved in just a few months.² By integrating internet searches and specialised knowledge databases, quality was improved and hallucinations – i.e. the invention of plausible but false content – in generative models were reduced. The range of AI models has also become significantly broader, meaning that ChatGPT’s near-monopoly in the language model segment has fallen from almost 90 per cent to 65 per cent.³

Yet as impressive as the technology is, it still has clear limitations that can be easily replicated by anyone using a simple test. Generative models can effortlessly create realistic-looking images of analogue clocks. However, they fail to correctly read the time from an analogue clock. As of September 2025, even the most powerful models failed this seemingly simple task in 9 out of 10 cases.⁴

The US and China dominate the development of AI models, whilst Europe plays only a minor role. The US benefits from low levels of regulation, the presence of many top-tier research institutions, the availability of and investment in data centres, and low energy costs, whilst China benefits from massive state investment in infrastructure.⁵ A dangerous bottleneck is emerging in the global supply of the necessary high-performance chips. Almost all leading AI chips are manufactured by just one Taiwanese company, TSMC.⁶ Due to very high barriers to entry, it is highly unlikely that this situation will change in the foreseeable future.

What are the current use cases in the financial sector?

There is no clear evidence of how much financial institutions globally have invested in the application of AI systems since 2023. Studies estimate annual global investment at between USD 35 billion and USD 87 billion, with more having been invested in the financial sector than in healthcare or the media industry.⁷ As such, the financial sector is currently one of the industries with the highest adoption rates.⁸

According to PwC, by 2025, 73 per cent of surveyed companies in the German financial sector will already be using AI, particularly in back-office operations, IT, marketing and sales. Summarising and analysing texts, using AI as a knowledge base, and creating draft reports and texts are the most common use cases for generative AI models.⁹ In a 2026 survey by Wolters Kluwer, risk management and fraud detection were cited as the most common areas of application.¹⁰

Overall, however, surveys suggest that the industry remains in the exploratory phase for AI applications and that large-scale implementation has not yet taken place. A study by the Wow!Banking initiative and FHDW Hannover cites unclear objectives, a lack of standards, rigid IT infrastructures and processes, and insufficient skills as the main obstacles to the application of AI in the German financial sector.¹¹ For example, there are many applications that allow AI-powered voice memos, enabling sales staff to make productive use of downtime whilst driving after client meetings. However, it is challenging to integrate this AI functionality into existing data management systems whilst complying with all data protection and data security regulations.

Fear of job loss: mission or challenge?

The negative impact of AI on labour markets has fallen far short of the original fears. In 2023, there was intense debate about which jobs were particularly at risk and likely to be replaceable in the near future. Two-thirds of all jobs in the economy were identified by Goldman Sachs in 2023 as being at risk, and a quarter were even deemed replaceable.¹² The problem with the discussion at the time was that a distinction was often not made between mission and task. For example, radiologists will not become redundant simply because AI systems can analyse medical images more effectively. Patients still want to communicate with a human being who takes responsibility and with whom trust can be built. The mission of helping and healing people remains unaffected by AI and is reserved for humans.¹³

In the financial industry, AI has so far established itself primarily as a support technology to increase efficiency, but not as a replacement for humans. According to a 2025 survey by the consultancy EY-Pantaleon, the focus is on the transformation and redesign of jobs, but not on complete automation.¹⁴ A report by the World Economic Forum from the same year concludes that AI in the financial sector not only automates tasks but also, to a large extent, complements and enhances work.¹⁵

However, there are also sectors where a decline in employment is on the horizon. These sectors are not, however, confined solely to the financial industry. According to a Stanford study from 2025, there was a decline in employment in the US for certain highly standardisable roles in customer support or software development, which particularly affected young, inexperienced career starters. For example, in July 2025, employment among young software developers was 20 per cent below the peak reached in 2022. It is noteworthy that software development is a professional field particularly closely linked to the development of AI, yet demand for more standardisable tasks within software development is falling.¹⁶

Has AI led to productivity gains?

Whilst there is clear evidence of the successful use of AI tools in the financial sector, concrete productivity gains are harder to demonstrate. Forrester Research cites significant time and efficiency savings in the banking sector and a shift of staff towards value-adding tasks.¹⁷ The AI consultancy Databricks expects AI automation to reduce operating costs at financial institutions by up to 20 per cent.¹⁸ The 2025 EY-Pantaleon survey shows that more than half of the US banks surveyed already see clear benefits, in the form of better decision-making or an improved customer experience. They expect to see further positive effects even more frequently in the future.

Deloitte, on the other hand, estimates that a clear, measurable value contribution is achieved through full-scale AI implementation in only one in five cases. It is believed that only a fraction of the profitability potential in the German banking sector has been realised so far.¹⁹ At best, therefore, one can speak of an initial indication of productivity gains in the financial sector.

A common narrative remains that productivity gains will lead to job losses in the financial sector in the long term, which should be reflected, among other things, in a decline in new hires. However, this does not necessarily have to lead to an overall decline in jobs in the financial sector. According to Jevons’ Paradox²⁰, the use of AI can boost productivity, causing product prices to fall and thereby increasing demand for those products. Consequently, capacity can be expanded, meaning that more new jobs are created than are lost elsewhere.

What are the current AI narratives for the sector that need to be examined in the future?

As is well known, stock markets trade on expectations of the future. At the start of 2026, parts of the software industry began to feel that the business models of specialist software providers and pay-per-user models were being challenged by AI. For financial services providers and banks, however, this narrative is not currently evident on the stock markets. The difference between mission and task is particularly stark here. Financial services are trust-based goods in which people are an essential component.²¹ There may well be a market for fully automated, human-free investment in the future, but its size is likely to remain modest. Large-scale waves of redundancies are therefore not currently expected in the financial sector. Forrester Research even reports that many companies regret AI-driven redundancies and are having to fill the staff gaps again.²² For employers and employees in the financial sector, this means that skills and specialist knowledge are in greater demand than ever.

This realisation puts my mind at ease.

My mission is clear, and AI systems are available for many tasks – I have made extensive use of them for this article too: Perplexity for researching current studies and DeepL for translations. ChatGPT and MS Office have helped to refine the wording. The AI function of the Google search engine has provided new insights into specific aspects. And ultimately, it was even the new AI tool mentioned at the start that motivated me to write this article.

I think this is ample evidence of a productivity gain, as it would hardly have been possible for me to complete this within the timeframe without these tools.

______________________________________________________

¹ See Epoch AI: “Algorithmic progress in language models” (2024).

² See Stanford: “Technical Performance | The 2026 AI Index Report | Stanford HAI” (2026).