Citi консолидировала данные клиентов в едином облачном хранилище для внедрения искусственного интеллекта

Citi has taken a decisive step toward preparing its global wealth management operations for artificial intelligence by consolidating client and market data that had long been trapped in separate regional data centers. Over an 18-month project, the bank moved information from five distinct regional facilities into a single cloud-based repository built on Snowflake and Amazon Web Services. The results have been immediate: reports that once required weeks of manual effort now appear in a matter of days. The head of data at Citi Global Wealth has been clear on the underlying principle: without one authoritative copy of the data, meaningful artificial intelligence applications remain out of reach. The project addressed a problem familiar to many large financial institutions. For years, Citi Global Wealth maintained client portfolios, transaction histories, market research, and compliance records in isolated systems across North America, Europe, Asia, Latin America, and the Middle East. Each region operated its own data infrastructure to satisfy local regulations, language requirements, and legacy technology choices. While this arrangement allowed regional teams to work efficiently on their own, it created significant friction when analysts or portfolio managers needed a global view. Pulling together a complete picture of a high-net-worth client’s holdings across jurisdictions could take days or weeks, involving multiple requests to different teams and subsequent reconciliation of inconsistent formats. By deciding to create one central copy in the cloud, Citi Global Wealth aimed to eliminate that fragmentation. The bank chose Snowflake as the core data platform because of its ability to handle structured and semi-structured data at scale while supporting multiple concurrent workloads. Amazon Web Services provided the underlying compute and storage layer, offering the elasticity needed to accommodate spikes in demand during market events or reporting cycles. The migration involved careful mapping of data fields, standardization of definitions, and the establishment of governance processes to ensure ongoing quality. Engineers and data architects spent the first several months cataloging every source system and documenting how each region captured similar information differently. Client identifiers, for example, followed distinct numbering conventions in different countries. Investment product classifications varied according to local regulatory terminology. Currency conversion methods and timestamp conventions also differed. The team created a master data model that preserved regional nuances while establishing a single source of truth for enterprise-wide analysis. Once the model was agreed upon, automated pipelines began extracting data from the regional data centers and loading it into the Snowflake environment. The process was staged to avoid disrupting day-to-day operations, with parallel runs conducted to validate accuracy before regional systems were gradually retired from analytical use. The payoff became visible quickly. Previously, a quarterly performance report for a global ultra-high-net-worth family office might require two weeks of coordination between London, New York, Singapore, and Dubai teams. Data would be exported into spreadsheets, manually adjusted for currency and classification differences, then combined in yet another spreadsheet that required further validation. With the new single copy, the same report can now be generated in hours through a self-service analytics layer that sits on top of the Snowflake instance. Portfolio managers can filter across asset classes, geographies, and client segments without waiting for extracts from multiple systems. The reduction in manual effort has freed analysts to focus on interpretation rather than data wrangling. Beyond faster reporting, the consolidated repository sets the foundation for more advanced applications. The head of data emphasized that artificial intelligence models require consistent, high-quality inputs across large volumes of information. A recommendation engine designed to suggest portfolio adjustments based on client risk profiles and market conditions cannot function reliably if it must reconcile conflicting definitions of “equity exposure” or “liquidity score” every time it runs. Similarly, natural language processing tools that scan research notes and news feeds for sentiment signals need a single reference point for entity recognition across all client records. Without that unified foundation, models either produce inconsistent results or require constant manual overrides that defeat the purpose of automation. Citi Global Wealth is now piloting several artificial intelligence use cases on the new platform. One involves predictive analytics for client behavior, identifying patterns that suggest when a client might be considering a major withdrawal or asset reallocation. Another experiment uses machine learning to detect anomalies in transaction flows that might indicate compliance issues or fraud. Both applications depend on the ability to join client data, market data, and external feeds without repeated cleansing steps. The bank has also begun experimenting with generative artificial intelligence tools that can draft customized market commentary for private banking clients, pulling from the unified research repository and tailoring language to each client’s stated preferences and risk tolerance. The project required more than technology investment. Significant attention went to change management across regional teams. Some analysts initially worried that centralization would reduce their influence over data definitions or slow response times for local regulatory requests. Citi addressed these concerns by implementing a federated governance model. Regional data stewards retain authority over local compliance interpretations while contributing to a central data catalog that documents every transformation rule. A self-service portal allows approved users to access the cloud repository with role-based permissions that respect both global standards and local privacy laws. This balance has helped maintain trust while still delivering enterprise-wide benefits. Data privacy and security received equal emphasis. Because the consolidated repository contains sensitive client information spanning multiple jurisdictions, Citi worked closely with legal and compliance teams to implement granular access controls and encryption both at rest and in transit. The Snowflake environment supports dynamic data masking, so analysts in one region see only the fields they are authorized to view. Audit logs capture every query, providing regulators with the transparency they require. These safeguards were essential for gaining buy-in from privacy officers who might otherwise have resisted moving data into a shared cloud environment. The 18-month timeline reflects the complexity of coordinating such a project across a global organization. Early phases focused on proof-of-concept work with a subset of high-priority data sets, allowing the team to demonstrate value before committing to full-scale migration. Once initial results showed clear improvements in reporting speed, executive sponsorship strengthened and budget for additional resources became easier to secure. Regular steering committee meetings kept stakeholders aligned on priorities and helped resolve conflicts between regional requirements. By the final months of the project, the focus had shifted from migration to optimization, tuning query performance and refining data quality rules based on real usage patterns. Financial institutions have long recognized that fragmented data limits their ability to serve clients effectively and compete with newer entrants. Citi’s experience illustrates one practical path forward. By investing the time to create a single, governed copy of core information assets, the bank has reduced operational friction and opened the door to artificial intelligence applications that would otherwise remain theoretical. The head of data’s statement captures the logic succinctly: without one copy, there can be no reliable artificial intelligence. The cloud-based repository on Snowflake and AWS now serves as that foundational layer. Looking ahead, Citi Global Wealth plans to expand the scope of the platform to include additional data types such as alternative investments, ESG metrics, and real-time market feeds. The team is also exploring tighter integration with the bank’s broader institutional businesses to enable cross-selling opportunities and holistic risk management. Each expansion will build on the same principle of data unity that drove the original project. As artificial intelligence capabilities continue to mature, the quality and consistency of underlying data will determine which organizations gain genuine advantage and which merely experiment without meaningful outcomes. The initiative also highlights the growing strategic importance of data platform decisions. Choosing Snowflake and AWS was not simply a technology preference; it represented a commitment to scalability and future flexibility. The separation of storage and compute in Snowflake allows Citi to adjust resources dynamically as query volumes grow, avoiding the over-provisioning common in traditional data warehouse environments. Meanwhile, AWS provides a global footprint that aligns with the bank’s own operations, simplifying compliance with data residency rules where they still apply. These architectural choices support both current reporting needs and the anticipated demands of more compute-intensive artificial intelligence workloads. For other financial services firms watching similar transformations, Citi’s project offers several practical lessons. First, success depends on aligning business priorities with technical execution from the outset. The decision to focus initially on reporting improvements created visible wins that sustained momentum through the more difficult standardization work. Second, governance cannot be an afterthought. Establishing clear ownership, documentation, and access policies early prevented later disputes and accelerated adoption. Third, migration should be approached incrementally. By validating each regional data set before cutting over, the team minimized risk and maintained operational continuity. The broader implication is that data consolidation is no longer optional for institutions serious about artificial intelligence. Models trained on incomplete or inconsistent information produce unreliable outputs that can damage client trust and invite regulatory scrutiny. A single, well-governed copy provides the necessary foundation for experimentation at scale. Citi Global Wealth has demonstrated that with disciplined execution, the transition from regional silos to a unified cloud platform can be achieved within a reasonable timeframe and deliver measurable returns even before advanced artificial intelligence features come fully online. As the bank continues to refine its data environment, the emphasis remains on quality, accessibility, and control. The cloud repository is not an end in itself but the starting point for a new generation of client services and internal efficiencies. Portfolio managers can now spend less time assembling data and more time advising clients. Compliance teams can run sophisticated monitoring routines without weeks of preparation. Data scientists can prototype models with confidence that their inputs reflect a consistent global reality. These shifts, taken together, position Citi Global Wealth to compete more effectively in an industry where timely insight and personalized service increasingly determine success. The project also underscores the continuing evolution of cloud adoption in banking. While many institutions began their cloud journeys with less sensitive workloads, leading organizations like Citi are now comfortable moving core client and market data into managed cloud services when proper safeguards are in place. The combination of Snowflake’s data-sharing capabilities and AWS’s security controls has given the bank both agility and assurance. Future enhancements may include tighter integration with on-premise systems for workloads that still require dedicated hardware, creating a hybrid model that balances innovation with regulatory obligations. Ultimately, Citi’s experience shows that meaningful progress toward artificial intelligence in wealth management starts with fundamental data management discipline. The 18-month effort to collapse regional silos into one authoritative cloud copy has already shortened reporting cycles, reduced operational costs, and created the conditions for more ambitious applications. As the head of data observed, the absence of a single copy makes artificial intelligence effectively impossible at enterprise scale. By addressing that gap, Citi Global Wealth has laid a practical foundation for the next phase of its digital transformation, one that places accurate, timely, and unified information at the center of every client interaction and internal decision.