Enterprise AI
UBS Red AI assistant reaches 100,000 employees with 25 million queries and 80-minute weekly savings
UBS has published rare production metrics for an internal generative AI assistant, giving enterprise leaders a concrete benchmark for time savings, deployment scale, and technology architecture in banking.
UBS Red is a bank-developed generative AI assistant that UBS has rolled out to roughly 100,000 employees, supporting more than 25 million queries and delivering average time savings of 80 minutes per user per week, according to UBS.
Executive Summary
UBS, the Swiss wealth manager and investment bank, has become one of the first large financial institutions to publish production-scale metrics for an internal generative AI assistant. The tool, called Red, is now used by about 100,000 employees and has supported more than 25 million queries since launch, according to UBS's innovation page. Based on user feedback, Red saves each user an average of 80 minutes per week, a figure that stands out in a sector where most generative AI deployments are still described in pilot terms or projected ROI.
The business win is twofold. First, UBS has moved an AI assistant from experimentation into daily work across a large portion of its workforce, with UBS describing hundreds of live use cases. Second, the bank has attached a quantified productivity outcome to that deployment: 80 minutes per user per week, a meaningful chunk of a knowledge worker's day. For CIOs and COOs evaluating their own AI investments, the UBS case provides one of the few publicly documented examples of a bank measuring time savings at enterprise scale.
Red was developed using Azure AI Search and Azure OpenAI Service, according to Microsoft's customer story, and it combines retrieval over curated UBS content with generative AI capabilities for summarization, drafting, and translation. The assistant is described by Microsoft as two domain-specific AI assistants collectively known as UBS Red, pointing to an architecture that separates distinct business functions rather than relying on a single company-wide model.
Why UBS built an internal AI assistant
UBS's motivation is rooted in the sheer volume of proprietary knowledge inside the bank. Research notes, financial market analysis, client insights, regulatory documentation, and internal expertise are scattered across business lines and geographies. Michel Neuhaus, Head of AI, Data & Analytics of Personal & Corporate Banking at UBS, framed the challenge in Microsoft's customer story: the bank's internal information is vast, and the institution must find ways to make it accessible to every employee.
That framing explains why UBS chose a retrieval-augmented approach rather than a general-purpose chatbot. The assistant is designed to answer questions using curated UBS content, which means the underlying model is grounded in the bank's own materials rather than open web data. For a regulated financial institution, grounding responses in controlled content reduces the risk of hallucinated answers and helps maintain consistency with internal policies and standards.
The decision to build in-house, on Microsoft Azure, also reflects a broader trend among banks: keeping model infrastructure on enterprise cloud platforms while retaining control over the application layer. UBS's published account emphasizes that Red is a UBS-built solution, not an off-the-shelf vendor product. That distinction matters to enterprise buyers because it signals that the bank treated AI as a core competency rather than a feature purchase.
The internal information that we have, the financial expertise, the insights into financial markets, what’s happening in the world, it’s vast. We have to find ways to make this content easily accessible to each and everyone in the bank, and Microsoft helps us with that.Michel Neuhaus, Head of AI, Data & Analytics of Personal & Corporate Banking, UBS
What is new: scale, usage, and measured time savings
The most consequential numbers in the UBS disclosure are the deployment scale and the query volume. UBS states that Red has been rolled out to around 100,000 employees and has supported more than 25 million queries since launch. A deployment of that size implies the assistant is embedded in day-to-day workflows, not confined to a small group of early adopters. The query count also suggests sustained usage: 25 million queries is not a pilot metric; it is a production workload.
The 80-minute weekly savings figure is based on user feedback, according to UBS. That methodology has limitations: self-reported time savings can overstate actual gains, and the figure is an average that may not hold evenly across roles. Still, it is a concrete, attributable number from the deploying organization, and it gives enterprise leaders a baseline against which to compare their own internal AI initiatives.
UBS's published account also describes Red as spanning hundreds of live use cases. That breadth is notable because many enterprise AI rollouts stall at a small number of high-profile applications. UBS appears to have distributed the assistant across multiple functions, letting employees apply it to tasks such as finding internal expertise, summarizing documents, and translating content for cross-border teams.
| Metric | Reported figure | Source |
|---|---|---|
| Deployment scope | Around 100,000 employees | UBS innovation page |
| Queries supported | More than 25 million since launch | UBS innovation page |
| Average time saved per user | 80 minutes per week | UBS innovation page, based on user feedback |
| Technology foundation | Azure AI Search and Azure OpenAI Service | Microsoft customer story |
How Red works: technical specifics
Microsoft's customer story says Red was developed using Azure AI Search and Azure OpenAI Service. Azure AI Search provides the retrieval layer that lets the assistant find relevant content within UBS's curated knowledge base. Azure OpenAI Service supplies the generative model capabilities that turn retrieved content into summaries, drafts, and translations. This is a classic retrieval-augmented generation architecture, adapted for a regulated enterprise environment.
The system is described as two domain-specific AI assistants collectively known as UBS Red. That detail suggests UBS did not build one monolithic assistant for the entire firm. Instead, the bank created tailored assistants for distinct domains, each with its own content sources and interaction patterns. Domain-specific assistants can improve response quality because the retrieval corpus is narrower and more relevant, and because the prompting and guardrails can be tuned to the language and workflows of a particular business unit.
The capabilities UBS highlights are summarization, drafting, and translation. These are not exotic use cases; they are high-frequency knowledge work tasks that consume employee time across banking. By targeting those tasks, UBS aimed at broad applicability rather than a single dramatic transformation. The design choice suggests a pragmatic strategy: give every employee a tool that makes everyday work faster, then measure whether time savings materialize.
- Summarization: Red condenses lengthy documents, research notes, and internal communications into concise summaries, reducing reading time for employees.
- Drafting: Red assists with generating drafts of emails, reports, and other written materials, accelerating the composition phase of knowledge work.
- Translation: Red supports translation across the languages used within UBS's global operations, helping cross-border teams share information more easily.
- Curated retrieval: Red uses Azure AI Search to retrieve information from curated UBS content, grounding responses in the bank's own materials.
- Domain-specific design: The UBS Red platform comprises two domain-specific assistants, built on Azure OpenAI Service, tailored to distinct business areas.
Measured outcomes and the productivity question
The 80-minute weekly savings figure is the centerpiece of UBS's productivity claim. Across a 100,000-employee deployment, that average implies a substantial aggregate gain, though UBS has not published a total hours-saved calculation in the cited materials. The absence of a grand total is itself instructive: UBS is reporting a per-user average based on feedback, not a company-wide ROI model with cost-per-hour assumptions.
Enterprise AI executives often struggle to move beyond qualitative enthusiasm to measurable outcomes. UBS's disclosure is notable because it names a metric, attributes it to user feedback, and publishes it on its own corporate site. That combination of specificity and official attribution is rare in banking, where most generative AI announcements emphasize capability rather than measured impact.
At the same time, the metric should be read with appropriate caution. Self-reported time savings can be inflated by respondent bias, and an average can mask wide variation across job functions. A trader, a compliance officer, and a wealth advisor may all use Red, but the time saved in each role could differ significantly. UBS's published materials do not break down savings by role, business line, or task type in the cited sources.
What the UBS rollout means for the banking industry
UBS's public disclosure gives other banks a reference point for what an internal AI assistant can achieve in a regulated environment. The deployment demonstrates that a large financial institution can operate generative AI at scale on enterprise cloud infrastructure, with retrieval grounded in internal content, without resorting to public web search or ungoverned consumer tools.
The choice of Microsoft Azure as the foundation also signals continued consolidation of enterprise AI infrastructure around major cloud providers. Banks face significant pressure to balance AI adoption with data sovereignty, security, and compliance. By building Red on Azure AI Search and Azure OpenAI Service, UBS placed the model layer in a controlled cloud environment while maintaining its own application and content governance.
For competitors, the UBS case raises the bar on disclosure. If a large global bank can publish deployment scale, query volume, and time-savings estimates, peer institutions may face internal pressure to produce similar metrics for their own AI programs. The absence of such numbers from other banks will increasingly look like a gap rather than a strategic choice.
Executive reaction and the role of vendor partnerships
Michel Neuhaus's comments in Microsoft's customer story emphasize the scale of UBS's information challenge and the role of the technology partnership. His quote frames Microsoft as an enabler of accessibility, not as the builder of the assistant. That division of labor is common in enterprise AI: the vendor supplies the model infrastructure, while the customer supplies the data, the use cases, and the domain expertise.
The Microsoft customer story describes UBS and Microsoft as co-creating the future of banking with Azure AI. For enterprise buyers, that framing is a reminder that successful AI deployments are rarely turnkey. UBS had to define its content architecture, curate its knowledge base, build the assistants, and integrate them into employee workflows. The cloud platform was an enabler, not the full solution.
Named-executive commentary of this kind also matters to procurement and risk teams. A senior business leader publicly tying an AI tool to measurable time savings gives internal stakeholders permission to treat the technology as a productivity investment rather than a research experiment. That cultural signal can be as important as the underlying model quality.
What should peer executives take away?
The first lesson is to ground the assistant in proprietary content. Red's value proposition depends on curated UBS material, not on general web knowledge. Enterprises that want similar results need to invest in the content layer: identifying authoritative sources, structuring them for retrieval, and keeping them current.
The second lesson is to measure from the start. UBS's 80-minute figure is based on user feedback, which means the bank built a mechanism to collect and aggregate that feedback. Any enterprise planning a large AI rollout should define its success metrics before deployment and build the instrumentation needed to report them credibly.
The third lesson is to target broad, frequent tasks. Summarization, drafting, and translation are not headline-grabbing applications, but they are tasks that touch many employees every day. By optimizing for frequency and breadth, UBS created a usage pattern that could plausibly accumulate to 25 million queries. Enterprises that chase only exotic, high-visibility use cases may miss the compounding gains of everyday assistance.
What is next for UBS Red
UBS's published materials do not specify a next-phase roadmap for Red in the cited sources. The bank's innovation page positions the assistant as part of a broader approach to AI and technology, and the Microsoft customer story emphasizes the co-creation relationship with Microsoft. Based on the disclosed metrics, the logical next steps would be expanding the assistant to the remaining employee population, adding more domain-specific assistants, and deepening integration with existing banking workflows.
The 80-minute weekly savings figure also creates a baseline for future measurement. If UBS can demonstrate that the savings persist or grow as the assistant gains more features, the case will strengthen. If the figure declines as novelty fades, it will provide an equally valuable lesson for the industry. Either way, UBS has given enterprise AI leaders something rare: a documented, attributable outcome to study and compare against their own programs.
Frequently asked
How many UBS employees use the Red AI assistant?
UBS has rolled out Red to around 100,000 employees, according to the bank's innovation page.
How much time does Red save each user per week?
UBS reports that Red saves employees an average of 80 minutes per week, based on user feedback.
What technology powers UBS Red?
Red was developed using Azure AI Search and Azure OpenAI Service, according to Microsoft's customer story.
What can UBS Red do?
Red combines curated UBS content with generative AI capabilities for summarization, drafting, and translation.
Where can I find the original UBS and Microsoft sources?
UBS published its figures on its innovation and AI page, and Microsoft published a customer story describing the Azure-based deployment.
Sources
- UBS — UBS states that Red has been rolled out to around 100,000 employees, has supported more than 25 million queries since launch, and saves employees an average of 80 minutes per week based on user feedback.
- Microsoft — Microsoft's customer story describes UBS Red as two domain-specific AI assistants developed using Azure AI Search and Azure OpenAI Service, and includes a direct quote from UBS executive Michel Neuhaus.
- McKinsey — UBS deployed its Red AI platform to around 100,000 employees, supporting over 25 million queries and delivering average savings of 80 minutes per user per week across hundreds of live use cases.