On 8 May 2026 Commerzbank presented first-quarter results together with an updated strategy called Momentum 2030. It contains a group-wide reduction of around 3,000 gross roles by 2030, alongside an announcement on artificial intelligence (AI): agents are to carry entire processes. Bettina Orlopp, chief executive officer, spelled out the connection to Deutsche Presse-Agentur. A “very large part” of the job cuts, she said, falls to AI.
A chief executive has thereby tied a workforce reduction to a technology with unusual directness. The statement deserves testing against the bank's own filings, and in both directions: how solid is the arithmetic, and how far along is the technology?
What: reduction of around 3,000 gross roles by 2030 under the Momentum 2030 strategy
Who: Commerzbank; remarks by Bettina Orlopp, chief executive officer
When: announced 8 May 2026 with the first-quarter results
Background: around 3,900 full-time roles by 2028 had already been announced in February 2025
Instruments: early retirement schemes, natural attrition, demographics; a transformation agreement with employee representatives is in place
The AI case: around €500m annual value contribution from 2030, roughly 10 per cent of capacity freed
What the filing says and what the chief executive says
The written release carries both subjects in adjacent paragraphs without connecting them. First the technology: agents are to support complete processes, from switching accounts through know-your-customer (KYC) and document checks to contract creation. Then the expectation: around €500m of annual value contribution from 2030, and roughly 10 per cent of capacity freed, expressly in order to redeploy part of it so that sales staff have more time for advice.
Only in the following paragraph do the cuts appear: the continued transformation of the bank comes with a group-wide reduction of around 3,000 further gross roles. No causal conjunction joins the two at that point. Bettina Orlopp supplied it verbally, and she narrowed it down: “Wir gehen zum Beispiel an die Kapazitäten bei externen Call-Centern ran. Das Gleiche gilt für das IT-Umfeld, wo wir noch viele Externe einsetzen.” In substance: external call centre capacity comes first, followed by the external IT workforce. The effect, she added, is larger than assumed a year ago.
The reduction therefore hits externally contracted capacity before the core workforce. That qualification is lost in the headline, and it explains why the bank believes it can avoid compulsory redundancies.
How far the technology has actually come
The filings draw a clean line here, and it is more sobering than the coverage. Commerzbank names exactly one process as agentic AI in production: an optimised complaints handling process in retail banking, described as launched and running with agentic AI.
Everything else sits in the future tense. Switching accounts, know-your-customer checks, document checks and contract creation are announced as the next development step, with no date and no pilot evidence. Read the four use cases as work in progress and you are reading an announcement as a stocktake.
Two frequently cited building blocks moreover do not belong in the agentic category. The banking avatar Ava has run on a Microsoft Azure foundation since 2025, handles around 30,000 enquiries a month and resolves 70 to 75 per cent of them independently on more recent figures. Christiane Vorspel, chief operating officer, describes agentic AI in her Börsen-Zeitung interview expressly as the step after Ava. Hawk AI, which scans large data volumes for money laundering and financial crime in risk management, is a pattern recognition model without autonomous process control. The partnership with the Munich firm Hawk has been public since 10 March 2026, and the product complements the existing rules-based compliance infrastructure.
What the numbers support, and what they do not
Three figures carry the case, and all three need framing. The €500m of annual value contribution applies from 2030, at the end of the strategy period. The roughly 10 per cent of freed capacity comes with no disclosed basis of calculation; the release names neither the reference base nor the method. And the phrase about freeing capacity in order to redeploy part of it describes a reallocation rather than a straight removal.
More importantly, the bank nowhere equates the 10 per cent with the 3,000 roles. Building an equation out of that is arithmetic on your own account. What holds is Bettina Orlopp's statement that a very large part of the reduction falls to AI, and that share carries no number.
The history matters just as much. The 3,000 roles are expressly “further” ones. In February 2025 Commerzbank had already announced around 3,900 full-time roles by 2028, then under the original Momentum strategy. The two programmes run to different target years on different counting bases; only the 3,000 are expressly labelled gross. Adding them into 6,900 conceals that difference.
Why the case matters beyond Commerzbank
For other institutions the construction is more interesting than the number. Commerzbank is promising the capital market an efficiency dividend from AI that largely still has to materialise, and pairing it with a reduction target that is already fixed. Four years separate the promise from the proof.
That creates a burden of evidence which transformation programmes rarely carry cleanly. Demonstrating in 2030 that the saving came from AI rather than from attrition, interest rates or portfolio effects requires an attribution per process starting today. With complaints handling, the bank has one case against which this could be measured.
What this means in practice
Four starting points for leaders in operations, transformation and HR.
Now: Record for each use case in your own house whether it runs in production, sits in a pilot, or has merely been announced. Commerzbank names exactly one agent process in production against four announced. Without that split in internal reporting, you cannot evidence maturity to a board or a supervisory board.
Now: Anyone announcing an efficiency dividend from AI has to separate it later from attrition and market rates. An attribution per process, set up before rollout, costs little. Reconstructed afterwards it is barely achievable, and that is precisely where evidence in transformation programmes tends to fail.
In planning: Bettina Orlopp names external call centres and external IT capacity as the first targets. That travels well: volumes there are clearly attributed, contracts can be ended, and co-determination questions are smaller. Starting your automation planning at that point delivers effect sooner than working through the core workforce.
Ongoing: An avatar in customer service, a pattern recognition model in anti-money laundering and an agent that runs a process autonomously are three different things. Blur them in reporting and you create a maturity on paper that operations does not have. Commerzbank separates them cleanly in its own filings; many retellings do not.
Christian Schablitzki
Strategy & Management Consultant · Agentic AI expert for financial institutions
More than 20 years in investment banking and derivatives trading, followed by over 10 years advising financial institutions. Currently Partner at Infosys Consulting in Germany. Certified in Google AI, Generative AI Leader (Google Cloud) and IBM RAG and Agentic AI.
LinkedIn profile →Keep reading – every fortnight in your inbox.
Capital markets insights, regulatory updates and AI trends. Concise, well-founded, free of charge.
GDPR-compliant. Unsubscribe at any time.