# Deutsche Bank Scales Enterprise AI With Publicis Sapient Across Code, Research, and AML

> The bank's AI foundation moved from proofs of concept to production in code development, research automation, and AML compliance, with more than 15,000 developers using AI coding assistants.

*Published 2026-11-09 · By Diane Okafor*

Deutsche Bank's enterprise AI/ML foundation is a governed platform, built with Publicis Sapient, that now supports three production use cases — AI-assisted software development, research and reporting chatbots, and AML and regulatory compliance — across the bank's businesses.

## Executive Summary

Deutsche Bank has moved its generative AI program from isolated proofs of concept to a production footing. In partnership with Publicis Sapient, the bank built an enterprise AI/ML foundation designed to scale AI safely across investment, corporate, private and retail, and asset management businesses. The partnership, which began in 2023, produced three use cases that are now delivering value: augmenting software code development, creating research and reporting chatbots, and applying AI to anti-money laundering and regulatory compliance.

The measured results are visible in adoption data. More than 80% of Deutsche Bank's workforce now use generative AI tools and digital assistants such as dbLumina and Microsoft Copilot, according to Deutsche Bank's technology transformation page. More than 15,000 developers use AI coding assistants, and each saves 1.5 to 2.5 hours per week. Those figures make the bank one of the more concrete examples of enterprise AI adoption in financial services, because they tie infrastructure investment to daily production usage.

This article examines what was deployed, why Deutsche Bank chose a governed platform approach rather than scattered pilots, how the three use cases were built and scaled, and what CIOs and CTOs in other regulated industries should take from the engagement. It draws on Publicis Sapient's published case study, HFS Research's analysis, and Deutsche Bank's own technology reporting.

## Why Did Deutsche Bank Need an AI Foundation, Not Just AI Tools?

Banks face two competing pressures. Generative AI can cut costs, speed up research, and improve client service, but financial institutions must also explain decisions, preserve audit trails, and satisfy regulators. Those requirements make model choice the least of the challenge; the harder work is building the infrastructure and controls around the model. Deutsche Bank's Technology, Data and Innovation division has been running a central programme to lay the foundation for safe and value-generating adoption and scaling, according to Deutsche Bank's technology transformation page.

Publicis Sapient's published case study on the engagement says the firm built and proofed an AI platform and infrastructure along with use cases, proofs of concept, operating models, and adoption and communication plans for Deutsche Bank in 2023. The proof-of-concept phase was not about demonstrating a chatbot in a lab. It was about testing whether the bank's data environment, risk controls, and engineering culture could support AI at production scale. The output was an operating model and an adoption plan, not just a model.

Deutsche Bank's scale makes governance a practical necessity. More than 15,000 developers and more than 80% of the workforce are now touching AI tools, so a mistake in one business unit could quickly become a bank-wide incident. Without guardrails, a code assistant could leak proprietary source code, a research chatbot could produce a plausible but false market summary, and an AML model could generate alerts that regulators cannot reconstruct. Deutsche Bank defined technical and operational standards, guardrails, and an AI platform with scalable shared services to keep those risks contained.

## Which Three AI Use Cases Are in Production at Deutsche Bank?

Publicis Sapient and Deutsche Bank identified three AI use cases that are delivering value to the bank, according to Publicis Sapient's case study. The first is augmenting software code development with AI, including documentation and understanding old code. For a bank running millions of lines of legacy software, AI-assisted development is one of the fastest ways to capture value because the benefit shows up in a developer's daily workflow: faster documentation, faster code explanation, and faster refactoring.

The second use case is creating chatbots for transcription, translation, summarization, research generation, insight generation, and automated reporting. These tools address the research and reporting burden that sits on top of every investment decision and client communication. A chatbot that produces a first draft of a research note or a client report does not replace an analyst's judgment; it compresses the time spent on retrieval and formatting, which frees analysts to focus on interpretation and risk assessment.

The third use case is applying AI in anti-money laundering and regulatory compliance, including conversation transcription and market abuse detection. This is the highest-stakes use case of the three. In regulated markets, AI outputs must be explainable, and their inputs must be traceable. Conversation transcription, for example, means processing trading-floor communications at scale, and market abuse detection means flagging behavior that may violate regulations. Both rely on natural language processing, and both require the auditability that the governance layer provides.

## How Did the Partnership Move From PoC to Production?

Publicis Sapient's case study describes the sequence Deutsche Bank followed to move from proofs of concept to production. The sequence explains why the bank treats governance as part of the platform rather than as a separate compliance exercise.

- Built and proofed an AI platform and infrastructure with Publicis Sapient.
- Defined technical and operational standards, guardrails, and shared services for safe AI adoption.
- Developed proofs of concept, operating models, adoption plans, and communication plans.
- Identified three use cases delivering value: code augmentation, research chatbots, and AML and regulatory compliance.
- Targeted scaling across investment, corporate, private and retail, and asset management businesses.

## How Was the Governance Layer Built?

Governance was not an add-on to Deutsche Bank's AI program; it was the foundation. The bank defined technical and operational standards, guardrails, and an AI platform with scalable shared services for safe AI adoption, according to Deutsche Bank's technology transformation page. Publicis Sapient's case study describes the result as an enterprise AI/ML foundation designed to scale AI across the bank. Shared services matter because they prevent each business unit from rebuilding compliance controls, data pipelines, and monitoring tools independently.

The governance layer also extends to the prompts that employees enter into AI tools. Gil Perez, Deutsche Bank's chief innovation officer, told HFS Research that organizations should, from day one, consider prompts that are unique to their company and how they capture, analyze, and harvest them. The comment frames prompts as a proprietary asset rather than as ephemeral user input. If employees are asking the AI questions about internal policy, deal structures, or client histories, those prompts reveal what the organization knows and what it does not.

For Deutsche Bank, capturing and analyzing prompts serves two purposes. First, it helps the bank understand which AI capabilities employees actually use and where the tools fail. Second, it feeds the governance process: prompts that surface sensitive information can be flagged, and prompt patterns can be used to train future models or refine guardrails. That is a more mature posture than the typical enterprise approach, which treats prompts as disposable.

> People undervalue and underestimate the interaction and knowledge of their company and employees. Organizations should, from day one, consider prompts that are unique to their company and how they capture, analyze, and harvest them.Gil Perez, Chief Innovation Officer, Deutsche Bank

## What Do the Adoption Numbers Show?

The clearest evidence that Deutsche Bank's AI foundation is working is the adoption data published by the bank. More than 80% of the workforce now use generative AI tools and digital assistants such as dbLumina and Microsoft Copilot, according to Deutsche Bank. For a bank of Deutsche Bank's size, that adoption rate is unusually high for a regulated financial institution, where AI adoption is often slowed by compliance reviews and access restrictions.

The developer statistic is even more specific. More than 15,000 developers at Deutsche Bank use AI coding assistants, according to Deutsche Bank, and the bank says those assistants save 1.5 to 2.5 hours per week. At two hours per developer per week, the stated adoption would represent roughly 30,000 developer hours weekly, though Deutsche Bank has not published that aggregate figure. Even at a conservative loaded cost per developer hour, that scale of savings is material to the bank's technology budget.

Those numbers should be read carefully. Deutsche Bank has not published a dollar-denominated ROI figure in the materials reviewed for this article, and time saved does not automatically convert to cost saved unless the time is redeployed. Still, the adoption rate and the time-savings range are concrete operational metrics, and they are unusual in a sector where many AI programs are still described in pilot terms.

## What Changed From PoC to Production?

| Use case | Before the foundation | After the foundation |
| --- | --- | --- |
| Software code development | Ad hoc tooling and manual documentation | AI coding assistants used by more than 15,000 developers, saving 1.5 to 2.5 hours per week |
| Research and reporting | Manual transcription, translation, summarization, and report drafting | Chatbots automate transcription, translation, summarization, research generation, and reporting |
| AML and regulatory compliance | Manual review of conversations and market activity | AI applied to conversation transcription and market abuse detection with auditable guardrails |
| Governance and scaling | Business-unit-specific pilots with inconsistent controls | Technical and operational standards, guardrails, and shared services on one AI platform |

The before column reflects the workloads described in Publicis Sapient's case study, and the after column reflects Deutsche Bank's published adoption data. The table shows that each use case moved from manual or ad hoc work to a shared, governed AI capability.

## What Are the Market and Stakeholder Implications?

Deutsche Bank's disclosure matters beyond the bank itself. It gives the market a rare look at what a large, regulated financial institution can achieve when AI is treated as an infrastructure program rather than a collection of pilots. The bank's decision to publish adoption statistics suggests that it considers the program a competitive advantage, not a compliance burden. That framing is intended for investors, regulators, and clients who want evidence that AI spending produces operational change.

For system integrators, the engagement is a signal about where the market is heading. Publicis Sapient's case study emphasizes the AI/ML foundation, operating models, adoption plans, and communication plans, not the specific models used. That is the language of enterprise transformation, and it reflects a shift in how banks buy AI services: they are buying the ability to govern and scale AI, not just a model deployment. HFS Research's analysis of the partnership places it in the context of Deutsche Bank's broader GenAI agenda.

For Deutsche Bank's business units, the foundation creates a common path to production. The partnership targets scaling solutions across the bank's investment, corporate, private and retail, and asset management businesses, according to Publicis Sapient's case study. A shared platform means a use case proven in one division can be reused in another without restarting the compliance review. That is the economic logic of a foundation: the first use case is expensive, but each additional use case gets cheaper.

Regulators are another stakeholder group with a direct interest in the engagement. Deutsche Bank's central programme and its published guardrails give supervisors a clear map of how AI decisions are made and where accountability sits. In an environment where financial regulators are issuing guidance on model risk management and AI governance, a bank that can document standards, shared services, and prompt management is better positioned than a bank running decentralized pilots. The disclosure itself is a form of regulatory communication.

## What Does Gil Perez Say About Prompts as a Strategic Asset?

Gil Perez's comment to HFS Research is the most direct executive statement in the public record about how Deutsche Bank thinks about enterprise AI. Perez said that people undervalue and underestimate the interaction and knowledge of their company and employees, and that organizations should, from day one, consider prompts that are unique to their company and how they capture, analyze, and harvest them. The remark puts prompt data on the same footing as other proprietary information: it is something to be captured, analyzed, and harvested.

The implication for other enterprises is that prompt governance belongs in the AI operating model, not in a privacy policy. Companies that ignore prompt data are leaving value on the table and, potentially, exposing themselves to risk. If a bank does not know what its employees are asking an AI assistant, it cannot detect data leakage, measure tool effectiveness, or improve the models that serve the business. Deutsche Bank's approach suggests that prompt capture and analysis should be designed into the platform from day one, exactly as Perez says.

## What Should Peer Executives Take Away?

The first takeaway is that governance is a scaling mechanism, not a speed bump. Deutsche Bank built standards, guardrails, and shared services before pushing AI to 80% of the workforce. That sequence is the opposite of the common enterprise pattern, where tool adoption runs ahead of controls and governance is retrofitted after an incident. The Deutsche Bank sequence makes governance the condition that enables scale.

The second takeaway is that the highest-value use cases are often the least glamorous. Code documentation, research summarization, and compliance transcription are not the kind of AI demos that generate headlines, but they are workloads with clear owners, clear data, and clear measures of success. Those characteristics make them easier to govern and easier to scale than broad, open-ended assistant projects.

The third takeaway is that adoption numbers matter more than model names. Deutsche Bank's disclosure of an 80% workforce adoption rate and 15,000 developers using AI coding assistants gives stakeholders a concrete way to track whether the program is real. Peer institutions should be prepared to answer the same question: how many employees use AI tools, and what work is the AI actually doing?

The fourth takeaway is that system integrators are now selling foundations and operating models, not model deployments. Publicis Sapient's case study documents the infrastructure, proofs of concept, operating models, and adoption plans as the deliverables. For buyers, that means procurement should ask about governance, prompt management, and scaling plans, not just about the model's benchmark scores.

## What's Next for Deutsche Bank's AI Program?

The published materials describe the next phase in general terms. The partnership targets scaling solutions across Deutsche Bank's investment, corporate, private and retail, and asset management businesses, according to Publicis Sapient's case study. That means the three proven use cases will be extended from the teams that built them to the wider organization, using the shared services and guardrails already in place.

The bank's Technology, Data and Innovation division continues to run the central programme that laid the foundation for safe and value-generating adoption and scaling, according to Deutsche Bank. The existence of a central programme suggests that new use cases will flow through the same governance process, rather than being launched independently by business units. That is a deliberate choice in a bank where regulators will expect consistency in how AI risk is managed.

Deutsche Bank has not published a timeline for additional use cases in the materials reviewed for this article. What is clear is the direction: the foundation is built, the governance is defined, the use cases are in production, and the bank is measuring adoption and time savings. For an industry that is still debating whether generative AI is real, Deutsche Bank's disclosures are evidence that it is, at least at one institution, an operational tool.

## What Is the Bottom Line for Financial Services?

Deutsche Bank's engagement with Publicis Sapient is a case study in how to industrialize generative AI in a regulated environment. The bank did not start with a large language model. It started with an AI/ML foundation, technical and operational standards, guardrails, and shared services. It then proved three use cases, measured adoption, and is now scaling across the enterprise.

The result is a rare set of public metrics in enterprise AI. More than 80% of the workforce uses generative AI tools, more than 15,000 developers use AI coding assistants, and those developers save 1.5 to 2.5 hours per week. The metrics are attributable to Deutsche Bank's own technology reporting, and they are consistent with the use cases documented in Publicis Sapient's case study and HFS Research's analysis.

Financial services firms will watch whether Deutsche Bank extends these use cases to revenue-facing workflows, such as client onboarding, trade surveillance, and investment research. The bank's stated ambition to scale across investment, corporate, private and retail, and asset management businesses suggests those workflows are in scope. If the adoption metrics hold as the program scales, Deutsche Bank will have demonstrated that governed AI can move from a technology initiative to a business operating system.

## Sources

1. [Publicis Sapient built and proofed an AI platform and infrastructure for Deutsche Bank and identified three use cases delivering value: augmenting software code development, creating chatbots, and applying AI in anti-money laundering and regulatory compliance.](https://www.publicissapient.com/customers/stories/deutsche-banks-gen-ai-transformation)
2. [HFS Research analyzed the partnership and published the Gil Perez quote on prompts.](https://www.hfsresearch.com/research/deutsche-bank-partners-with-publicis-sapient/)
3. [More than 80% of Deutsche Bank's workforce use generative AI tools and digital assistants such as dbLumina and Microsoft Copilot; more than 15,000 developers use AI coding assistants saving 1.5 to 2.5 hours per week; the TDI division runs a central programme for safe AI adoption.](https://www.db.com/what-we-do/focus-topics/tech/index?language_id=1)

---
Source: https://aiintelreport.com/enterprise-ai/deutsche-bank-publicis-sapient-enterprise-ai-foundation
Index: https://aiintelreport.com/llms.txt · Full text: https://aiintelreport.com/llms-full.txt