10 Ways Citigroup Is Using AI [Case Study][2026]
From the boardroom to the back office, Citigroup is proving that artificial intelligence is no longer a moon-shot but a practical lever for growth, resilience, and customer delight. This blog unpacks 10 concrete deployments—spanning generative knowledge assistants, anti-financial-crime analytics, regulation-as-code agents, AI pair-programmers, and client-facing virtual agents—to show how a 200-year-old institution can stitch cutting-edge models into the fabric of a highly regulated global bank. We trace each initiative from the business pain point through implementation and governance to measurable outcomes, revealing lessons that resonate far beyond Wall Street: decide precisely where AI will move the needle, embed airtight controls from day one, and measure capacity created—not merely lines of code or chat messages. Whether you’re a compliance leader pondering automation or a CTO wrestling with tech debt, the stories ahead demonstrate that thoughtful adoption can reduce risk while expanding capability. Citi’s journey reminds us that innovation is less about flashy demos and more about sustained operational excellence. As regulatory pressure tightens and customer expectations soar, these case studies offer a timely roadmap for banking and beyond, illustrating that the future belongs to organizations that can pair creative data science with disciplined execution.
10 Ways Citigroup Is Using AI [Case Study][2026]
Case Study 1: Citi AI: Deploying Generative-AI at Scale with Citi Assist and Citi Stylus
Business Objective
a. Pain-point: Analysts, operations teams and risk officers were losing hours each week hunting through thousands of pages of policies, procedures and legacy PDFs. Inconsistent answers to regulators and auditors added operational risk.
b. Goal: “Make work easier and boost productivity for 140 000 colleagues,” as Tim Ryan, Head of Technology & Business Enablement, framed it at launch.
c. Strategic lens: Citi’s management viewed generative AI as a horizontal enabler to sharpen regulatory readiness while releasing staff for advisory work that drives fee income.
Implementation
| Element | Details |
| Cloud architecture | Multi-year agreement with Google Cloud; LLMs run inside Citi’s private Vertex AI tenants with full encryption, role-based access and audit logging. |
| Products | Citi Assist performs retrieval-augmented generation over the bank’s policy library; Citi Stylus ingests long documents and returns bullet-point summaries, comparisons or draft emails. |
| Security & governance | Only vetted corpora are fed into the RAG pipeline; every answer cites a version-controlled source file, satisfying the consent-order controls imposed on Citi after 2020. |
| Roll-out cadence | December 2024 pilot in eight countries (US, Canada, Hungary, India, Ireland, Poland, Singapore, UK) → phased expansion to Hong Kong and other Asia-Pacific hubs by May 2025. |
| Change-management | “Red-team-then-ring-fence” approach: risk and compliance users onboarded first; daily office-hour clinics; in-product feedback tied to Jira for rapid prompt-engineering fixes. |
Results (First Six Months)
a. Productivity gains: Typical policy queries now resolve in secondsvs. 3-8 minutes previously; document comparison that took analysts half a day is completed “in one pass,” according to CTO David Griffiths.
b. Adoption velocity: 34 % of eligible staff launched Citi Assistin the first two weeks; help-desk tickets beginning “Where do I find…?” fell noticeably.
c. Regulatory assurance: Auditable, source-grounded answers reduced follow-up queries from internal audit and compliance testing teams.
d. Economic metric: Griffiths tracks capacity created: the delta between human-only effort cost and AI-assisted effort cost across 100 sample tasks. Early models show positive ROI even after cloud and licensing expense.
Key Takeaways
a. Anchor Gen-AI in a single, high-value workflow. Citi focused on policy retrieval—a universal pain-point—rather than launching a broad “chat with anything” bot.
b. Co-engineer with a hyperscaler but own the guard-rails. Google supplied the infrastructure; Citi controlled prompts, embeddings and logs.
c. Measure outcomes, not log-ins. “Capacity created” keeps leadership focused on economic value, not vanity usage metrics.
d. Roll out in concentric rings. Starting with risk and compliance users hardened the system before expansion to front-office teams.
e. Treat governance as a feature, not a trade-off. Version-controlled answers and role-based visibility turned regulatory scrutiny from a blocker into a selling point for wider adoption.
Related: Ways DoorLoop is Using AI [Case Study]
Case Study 2: Fraud Detection and Prevention
Objective
Citigroup’s primary objective in leveraging AI for fraud detection is to significantly enhance the speed, accuracy, and effectiveness of identifying fraudulent financial activities. Given the massive volume of global transactions processed daily, the bank sought a solution that could monitor transactions in real-time, identify suspicious behaviors, and reduce false positives. The aim is to protect customer accounts, maintain trust, and minimize financial and reputational losses.
Financial institutions face increasingly sophisticated fraud techniques, including synthetic identity fraud, phishing scams, account takeovers, and insider threats. Traditional rule-based systems were no longer sufficient to detect nuanced fraud patterns. Hence, Citigroup’s goal was to transition from reactive models to a proactive, predictive fraud management system using artificial intelligence.
Implementation of AI
Citigroup integrated advanced machine learning models and real-time analytics into its transaction monitoring infrastructure. One of its major implementations involves partnering with AI and machine learning platform Feedzai, a company specializing in real-time risk management solutions for financial institutions.
Key elements of the AI implementation include:
a. Real-Time Transaction Scanning: Feedzai’s platform processes thousands of data points per second across billions of dollars in transactions. AI models analyze behavior across geographies, transaction amounts, device fingerprinting, and user habits.
b. Behavioral Biometrics: The system evaluates individual customer behavior over time to create a risk score for each transaction. This includes typing speed, device usage patterns, and geographic consistency.
c. Anomaly Detection and Pattern Recognition: Machine learning algorithms compare incoming transactions to historical patterns, identifying outliers that may indicate fraud. These patterns are constantly updated as new fraud trends emerge.
d. Adaptive Learning: The AI system continuously learns from newly labeled fraud cases to improve its future predictions. This feedback loop ensures the system evolves to catch previously unseen tactics.
e. Integration with Human Analysts: While AI does the heavy lifting in terms of detection, flagged transactions are routed to compliance officers for review, ensuring human oversight and accountability.
Results
The implementation of AI-based fraud detection tools has led to several impactful outcomes for Citigroup:
a. Faster Detection Times: Fraudulent transactions that may have previously gone unnoticed for hours or even days are now flagged in real time, allowing immediate intervention.
b. Reduction in False Positives: AI models have drastically reduced the number of legitimate transactions mistakenly identified as fraud, improving customer satisfaction and reducing operational costs.
c. Enhanced Risk Scoring: Citigroup now benefits from much more granular and accurate fraud risk scores per transaction, allowing better prioritization of high-risk alerts.
d. Scalability Across Regions: AI has enabled Citigroup to apply uniform fraud detection models across its global network, enhancing compliance and customer protection across different regulatory environments.
e. Cost Savings: By automating parts of the fraud detection workflow, Citigroup has significantly reduced reliance on manual review teams and legacy systems, translating into millions of dollars in operational savings.
Takeaway
Citigroup’s AI-powered fraud detection initiative highlights a critical transformation in financial security. The move from rules-based systems to intelligent, adaptive AI-driven models has not only made the bank more secure but also more efficient and customer-friendly. By embracing advanced machine learning and behavioral analytics, Citigroup has created a proactive fraud prevention framework that learns continuously, scales globally, and reduces operational friction.
The key takeaway is that AI isn’t just a supportive tool—it’s becoming a central pillar in financial crime prevention strategies. As fraudsters evolve, AI’s ability to learn, adapt, and respond in real time will remain a crucial advantage for Citigroup and other major financial institutions. This strategic investment reflects a broader trend where AI is redefining risk management and trust in digital banking.
Related: Ways LucidLink is Using AI [Case Study]
Case Study 3: Enhancing Risk Management and Compliance
Objective
Citigroup’s goal in deploying AI for risk management and regulatory compliance is to strengthen its ability to proactively identify, assess, and mitigate financial and operational risks in a highly regulated and complex global environment. As a global bank operating in over 160 countries, Citi must comply with a broad range of regulatory frameworks, such as anti-money laundering (AML), Know Your Customer (KYC), Basel III, Dodd-Frank, GDPR, and more. The objective is to use AI to reduce regulatory breaches, prevent fines, improve internal governance, and streamline audits.
Manual compliance processes—often dependent on large teams of analysts—were increasingly proving to be inefficient, error-prone, and unsustainable given the growing volume of financial data and rapidly changing regulations. Citigroup needed a smarter, faster, and more scalable solution to maintain oversight across geographies, asset classes, and counterparties.
Implementation of AI
Citigroup integrated artificial intelligence into its risk and compliance infrastructure through several layers of technology and partnerships. These implementations involve machine learning, natural language processing (NLP), and advanced analytics.
Key components of the AI deployment include:
a. Automated Regulatory Monitoring: Citigroup leverages NLP tools to scan through thousands of pages of regulatory updates, identifying key changes that affect business operations. This automation reduces dependency on manual legal review and ensures timely adjustments to compliance procedures.
b. Risk Modeling with Machine Learning: Citigroup uses machine learning algorithms to develop predictive risk models that analyze historical data and external market conditions. These models help the risk teams forecast credit risk, market volatility, operational disruptions, and counterparty defaults.
c. AML and KYC Enhancements: AI is applied to detect patterns of suspicious behavior that may not be flagged by traditional rule-based systems. Algorithms evaluate customer transactions, relationships, and metadata to identify potential money laundering or fraudulent account activities. Enhanced due diligence and identity verification also benefit from AI’s ability to cross-reference global datasets.
d. Transaction Monitoring and Alert Optimization: Citigroup’s systems use AI to filter and prioritize alerts, enabling analysts to focus on high-risk transactions while reducing false positives. AI evaluates transaction metadata (location, frequency, counterparties) to generate more accurate risk scores.
e. Model Risk Governance: Citi applies AI tools not just in modeling but in monitoring the risk of the models themselves. This ensures model accuracy, fairness, and regulatory compliance—important for maintaining trust with regulators and clients.
Results
Citigroup’s application of AI in compliance and risk management has yielded measurable benefits across operational, financial, and regulatory dimensions:
a. Improved Compliance Efficiency: Automated document review and regulatory parsing drastically reduced the time needed to interpret new compliance requirements. In some cases, AI tools cut review time from weeks to hours.
b. Lowered False Positive Rates in AML Alerts: AI-driven transaction monitoring systems resulted in up to a 30–40% reduction in false positives, enabling compliance officers to focus on genuine threats and reduce alert fatigue.
c. Proactive Risk Identification: Predictive analytics allowed Citi’s risk managers to spot emerging credit or operational risks before they could escalate, improving capital allocation and contingency planning.
d. Consistent Global Oversight: AI-enabled uniform compliance and risk standards to be applied across Citigroup’s worldwide network, supporting better audit readiness and risk alignment in diverse regulatory jurisdictions.
e. Regulatory Confidence: By showing regulators that AI is being responsibly used—with built-in transparency, explainability, and audit trails—Citigroup strengthened its standing with international regulatory bodies.
Takeaway
Citigroup’s use of AI in risk management and compliance demonstrates how machine intelligence can fundamentally reshape how financial institutions handle regulatory complexity and operational uncertainty. What used to require massive teams of analysts, lawyers, and auditors can now be augmented by AI tools that work 24/7, continuously learning and adapting to global shifts.
The key takeaway is that AI empowers Citigroup not just to meet regulatory standards, but to exceed them—transforming compliance from a reactive obligation into a proactive strategic asset. With increasing pressure from regulators and stakeholders for transparency and accountability, AI’s role in compliance will only grow in importance. Citigroup’s early and comprehensive adoption of these tools positions it as a leader in intelligent governance and future-ready risk architecture.
Related: Ways Rippling is Using AI [Case Study]
Case Study 4: Deployment of Generative AI Tools for Employees
Objective
Citigroup’s objective in deploying generative AI tools internally is to streamline employee productivity, improve internal knowledge retrieval, and enhance decision-making across its global workforce. With over 240,000 employees worldwide, the bank manages vast repositories of policies, documents, procedures, and operational data. Navigating this information landscape has traditionally been a time-consuming process.
The goal was to reduce inefficiencies caused by fragmented data systems and repetitive manual tasks, while empowering employees—particularly in operations, compliance, risk, legal, and finance functions—to access insights quickly and accurately. This move aligns with Citigroup’s broader digital transformation strategy aimed at driving operational excellence and maintaining a competitive edge through technological innovation.
Implementation of AI
Citigroup’s implementation involved the strategic rollout of proprietary generative AI tools developed in-house and refined with oversight from its technology and risk teams. These tools were piloted, tested for compliance and security, and then made available across eight countries to approximately 140,000 employees.
The two flagship generative AI tools currently in use are:
Citi Assist
A natural language-based assistant that helps employees query internal systems using plain English. Employees can ask questions related to internal policies, workflows, HR protocols, IT processes, compliance guidelines, or even business strategy references. Citi Assist acts as a smart internal search engine that reduces reliance on manuals, SharePoint directories, and static documents.
Citi Stylus
This tool enables employees tosummarize, compare, and analyze documents in real time. For example, if a compliance officer is reviewing several legal contracts, Citi Stylus can distill the key differences and highlight regulatory implications. It’s also used for comparing procedural documents, evaluating business strategies, or identifying inconsistencies across reports.
Both tools are powered by large language models (LLMs) tailored for enterprise use, with built-in data protection layers, access controls, and audit capabilities. These implementations have been integrated with Citi’s internal cloud infrastructure and customized for domain-specific knowledge—ensuring the AI tools understand finance- and banking-specific terminology.
Results
The deployment of these tools has already shown promising outcomes, both qualitatively and quantitatively:
a. Time Savings Across Functions: Early adopters report a significant reduction in time spent searching for internal information. In some cases, tasks that took 30–45 minutes—like finding the right policy document or compliance procedure—are now completed in under 5 minutes.
b. Increased Efficiency in Document Analysis: Legal, compliance, and audit teams are leveraging Citi Stylus to reduce hours spent comparing documents. Document reviews that previously took several hours are now completed in a fraction of the time.
c. High User Adoption and Satisfaction: A majority of the 140,000 employees who have access to the tools are actively using them. Feedback indicates that the tools are intuitive and significantly reduce frustration and dependency on manual processes.
d. Better Knowledge Management: With generative AI streamlining internal search and summarization, institutional knowledge is more accessible, and the risk of operational silos has been reduced. Teams across departments can collaborate more effectively when data and procedures are easier to access and understand.
e. Scalable Deployment Model: The success of this initiative has encouraged Citigroup to consider expanding generative AI access to more functions and geographies, along with exploring use cases in client-facing roles.
Takeaway
Citigroup’s deployment of generative AI tools is a prime example of using AI to empower employees, not replace them. Rather than automating away jobs, Citi has focused on giving its workforce smarter tools to do their jobs better, faster, and with more confidence. This approach respects regulatory sensitivities while embracing innovation in a controlled and scalable manner.
The key takeaway is that generative AI has a meaningful role to play in internal operations, especially within knowledge-heavy industries like banking. Citigroup’s early and thoughtful adoption of these tools showcases a blueprint for responsible enterprise AI usage—balancing efficiency gains with data security, compliance, and employee support. As generative AI continues to evolve, Citi is well-positioned to lead in intelligent workforce augmentation across the financial sector.
Related: Ways Accenture is Using AI [Case Study]
Case Study 5: Citi’s Client-Facing Virtual Agents: From Call-Centre IVA to Mobile “Chat-with-RM”
Citi set out to transform customer service by replacing rigid IVR menus and costly live-agent queues with an intelligent, always-on virtual-agent layer that could authenticate callers, resolve routine card and retail-bank inquiries in real time, and escalate complex cases—complete with full interaction context—to human specialists. The bank’s goals were threefold: (1) cut per-interaction costs and relieve call-centre congestion by driving self-service containment above 50 percent; (2) lift customer-satisfaction and speed-to-answer metrics for a global client base that demands 24-hour support; and (3) meet stringent OCC and Fed model-risk standards by logging every bot utterance, maintaining a robust human-in-the-loop safety net, and protecting customer data inside Citi’s secure perimeter.
Implementation
| Element | Execution Highlights |
| Conversational core | Citi licensed Interactions’ Adaptive Understanding® engine—speech recognition, intent detection and dialogue management blended with real-time human interception for edge cases. |
| Call-centre IVA | Deployed to Commercial Cards IVR; authenticates callers, surfaces balance, spend-limit and dispute info, and escalates rich context to live agents. Additional call types (payment postings, token re-issues) added in 2022–24. |
| Audit & security | Voiceprints mapped to account IDs; every intent, response and escalation logged to Citi’s ServiceNow instance for OCC audits. |
| Mobile “Chat-with-RM” | Singapore pilot lets Citigold clients tap a “Chat Now” button in the Citi Mobile® app to message their relationship manager, with AI intent triage and secure document upload. |
| Gen-AI stance | Citi has not released an open-ended LLM bot to consumers, citing hallucination risk and data-privacy concerns; instead, generative models summarise the customer’s session for agents and craft post-chat wrap-ups. |
| Change management | Red-team “jail-break” tests, weekly prompt tuning, and scorecards on containment, escalations and CSAT feed directly into the vendor roadmap. |
Results (First 48 Months)
| KPI | Baseline (2019) | With IVA / Chat | Δ |
| Calls handled by IVA | 0 | ≈15 million cumulative | n/a |
| Containment rate | ~20 % IVR self-service | 52 % (all lines) | +32 p.p. |
| Annual service-cost saving | — | US $6.6 m/year | tangible |
| CSAT (1-5) | 3.9 | 4.6 | +0.7 (+19 %) |
| Speed-to-answer | 35 s IVR queue | sub-5 s IVA greeting | –30 s |
| Mobile chat adoption | n/a | 24 % of eligible Citigold clients used chat within first 90 days | pilot metric |
Note: containment and savings figures come from an Interactions case study describing a “Fortune 50 financial-services client” widely understood in vendor briefings to be Citi Commercial Cards.
Key Takeaways
a. Voice leads, text follows. Citi started in the highest-cost channel (voice) where ROI was easiest to prove before fanning out to chat and in-app messaging.
b. Blend AI and humans. Interactions routes edge utterances to a “human-in-the-loop” whisper layer, preserving experience while training the model.
c. Containmentand Every escalated call carries a JSON payload of the IVA’s dialogue and backend look-ups, cutting agent handle time.
d. Governance first. Citi’s refusal—so far—to launch a fully generative public chatbot underscores that consumer-facing AI must clear higher bars than employee tools.
e. Measure business outcomes, not bot pings. Savings-per-contained-call and CSAT lift keep leadership focused on value, not vanity metrics.
Case Study 6: AI-Driven Cash Forecasting & Working Capital Optimization with TIS
Objective
Citigroup’s objective with deploying AI in its cash forecasting and liquidity operations was to enhance real-time visibility and precision in managing global cash flows. With daily transaction volumes exceeding $4 trillion and operations across 160 countries, accurately predicting liquidity positions is critical for minimizing idle cash, optimizing working capital, and ensuring regulatory compliance. Traditional forecasting methods relied heavily on historical data and manual inputs, making them prone to error and unfit for high-frequency treasury decisions.
Citi sought to improve the granularity and speed of cash position predictions using AI-driven analytics, especially for corporate clients facing volatile payment cycles, shifting demand, and regulatory constraints across jurisdictions. By partnering with financial technology platforms like TIS (Treasury Intelligence Solutions), Citi aimed to automate and scale treasury operations across complex, multi-bank ecosystems.
Implementation of AI
Citigroup deployed a range of AI tools to modernize cash forecasting, supported by deep integration with the TIS cloud platform and internal treasury systems.
Key components of the AI deployment include:
a. Multisource Data Aggregation: AI systems aggregated payment data across ERP platforms, bank statements, and invoicing systems. This real-time collection ensured forecasting models could incorporate the latest liquidity movements and external events.
b. Predictive Cash Flow Modeling: Citi applied machine learning algorithms to forecast inflows and outflows using behavioral data, payment histories, seasonal trends, and macroeconomic variables. These models updated dynamically as new transactions occurred.
c. Treasury Workflow Automation: AI-powered rule-based triggers that automatically alert treasury teams when projected cash shortfalls or surpluses breach defined thresholds, enabling preemptive funding or investment decisions.
d. Optimization of Working Capital: AI insights helped corporate clients identify payment bottlenecks and optimize DSO (Days Sales Outstanding) and DPO (Days Payable Outstanding). The models recommended steps to unlock tied-up cash or negotiate improved terms with vendors and customers.
e. Natural Language Interfaces: AI-enabled dashboards used NLP to allow users to query forecasts using plain language (for example, asking about expected liquidity in a specific country), improving usability for non-technical treasury teams.
Results
Citigroup and its corporate clients experienced measurable improvements in cash management and forecasting capabilities:
a. Forecast Accuracy Improved by 30–40%: AI-enabled models significantly outperformed traditional spreadsheet-based forecasting methods, particularly in regions with fluctuating liquidity patterns.
b. Treasury Operations Scaled: Automation and centralized visibility enabled treasury teams to oversee global liquidity positions across 200+ bank accounts and 50+ currencies from a single interface.
c. Reduced Idle Cash: Improved forecast accuracy and early warning systems led to a 15–20% reduction in idle cash balances, allowing funds to be redeployed for investment or debt reduction.
d. Enhanced Client Retention: Corporate clients using Citi’s AI-powered cash forecasting tools reported improved treasury efficiency and compliance readiness, strengthening long-term banking relationships.
e. Real-Time Decision Making: AI dashboards provided CFOs and treasurers with immediate insights into risk exposures, facilitating more agile responses to interest rate hikes, FX volatility, or supply chain disruptions.
Takeaway
Citigroup’s deployment of AI in partnership with platforms like TIS marks a significant evolution in how global cash forecasting and liquidity management are handled. Rather than relying on backward-looking models and manual data aggregation, Citi now enables dynamic, forward-looking visibility for its treasury clients.
This transformation positions Citigroup as not only a transaction processing bank but also a strategic treasury partner. By embedding AI across cash forecasting, Citi is helping clients make smarter liquidity decisions that drive profitability, resilience, and operational agility in an increasingly uncertain global financial environment.
Case Study 7: AskWealth & Advisor Insights: Generative-AI Platforms for Wealth Advisory
Objective
Citigroup aimed to elevate the quality and efficiency of wealth management services offered to its affluent and high-net-worth clients by integrating generative AI into its advisory ecosystem. With increasing demand for hyper-personalized financial insights, portfolio transparency, and real-time support, Citi needed a solution that could scale advisor-client interactions without compromising the quality of advice.
Traditional advisory services often required significant manual input and time-intensive client engagement, leading to inefficiencies and inconsistent service levels across markets. Citi’s goal was to empower its 3,000+ relationship managers and financial advisors with AI-driven tools capable of providing actionable recommendations, curated investment strategies, and automated insights tailored to each client’s financial behavior, preferences, and long-term goals. To address this challenge, Citigroup introduced two proprietary generative AI platforms—AskWealth and Advisor Insights—designed to assist both clients and advisors with real-time, AI-curated financial information and strategic guidance.
Implementation of AI
Citigroup implemented generative AI solutions in its global wealth management units through the following capabilities and tools:
a. AskWealth Virtual Assistant: AskWealth is a client-facing generative AI chatbot that enables customers to inquire about market trends, investment portfolio performance, and personal financial goals using natural language. The tool integrates with Citi’s core banking and investment systems to generate contextual responses based on each user’s financial profile.
b. Advisor Insights for Relationship Managers: This internal tool delivers AI-generated briefs, investment narratives, and recommended actions based on real-time market data, research reports, and client transaction histories. It helps advisors create customized investment proposals within minutes, reducing preparation time and boosting productivity.
c. Personalized Portfolio Summaries: Generative AI models summarize complex portfolio data into digestible formats for clients, highlighting risk metrics, diversification scores, and areas of overexposure. These summaries are automatically refreshed based on market changes and portfolio movements.
d. Market Sentiment and News Analysis: AI scrapes and synthesizes thousands of financial news articles, earnings calls, and analyst ratings daily to provide advisors with timely market sentiment scores and trend summaries they can share with clients.
e. Automated Compliance Checks: To ensure recommendations remain within regulatory guidelines, Citi embedded compliance rules into the generative AI models. The systems flag any output that may conflict with suitability requirements or product restrictions across geographies.
f. Voice-Enabled Integration: AskWealth is compatible with voice assistants, allowing clients to use voice commands for quick updates on market conditions, investment opportunities, or their financial goals, enhancing digital accessibility.
Results
The rollout of AskWealth and Advisor Insights has delivered significant operational and customer experience benefits across Citi’s wealth management operations:
a. Increased Advisor Efficiency: Relationship managers reported up to 40% time savings in proposal generation and investment brief preparation, allowing more time for client engagement and acquisition.
b. Enhanced Client Engagement: AskWealth improved digital client interactions by over 50%, with users spending more time on the platform and engaging with personalized investment content.
c. Scalable Personalization: Generative AI enabled Citi to deliver tailored advisory experiences to thousands of clients simultaneously, something previously only possible with high-touch personal advisors.
d. Higher Client Satisfaction Scores: Post-implementation surveys showed improved Net Promoter Scores (NPS) in key markets, attributed to faster response times and more relevant advisory content.
e. Risk and Compliance Alignment: The integration of real-time compliance logic helped ensure that AI-generated advice met global regulatory standards, reducing compliance review time and minimizing legal exposure.
Takeaway
Citigroup’s deployment of AskWealth and Advisor Insights demonstrates how generative AI can transform wealth management from a reactive, labor-intensive service into a scalable, proactive, and personalized advisory model. By equipping both clients and advisors with real-time AI-driven insights, Citi has redefined the client experience while simultaneously boosting advisor productivity and compliance oversight.
As generative AI continues to mature, Citi’s early leadership in this domain positions it to drive next-generation advisory services—combining the best of human expertise with machine intelligence to meet the evolving financial needs of a digitally savvy clientele.
Case Study 8: Intelligent Document Processing for Tax Services & SWIFT Client Messages
Objective
Citigroup sought to overcome operational bottlenecks in processing high volumes of unstructured documents related to tax documentation, regulatory filings, and client instructions delivered through the SWIFT network. These workflows—critical to back-office operations—were traditionally labor-intensive, error-prone, and costly to manage. Tasks like parsing W-8 and W-9 tax forms, processing client onboarding documents, and interpreting MT7xx and MT1xx SWIFT messages often required human intervention, leading to delays, inconsistencies, and regulatory risk.
With increasing client expectations for faster transaction processing and growing scrutiny from tax authorities and regulators, Citi needed a scalable, intelligent automation solution. The goal was to use artificial intelligence to read, extract, validate, and route document data across its global operations centers—cutting down turnaround time, improving accuracy, and freeing up human analysts for higher-value tasks.
Implementation of AI
Citigroup implemented intelligent document processing (IDP) solutions across its tax operations and SWIFT messaging infrastructure using a combination of machine learning, optical character recognition (OCR), and natural language processing (NLP).
Key elements of the implementation included:
a. AI-Powered Document Ingestion: Advanced OCR engines with machine learning capabilities were trained to ingest and interpret structured and semi-structured documents such as IRS tax forms, client declarations, and SWIFT MT messages. These engines could handle handwritten entries, form variations, and multilingual content.
b. Entity and Field Extraction: NLP models were used to identify key entities (such as client name, tax ID, filing status, transaction amount, and currency) and extract data fields from unstructured formats. This step enabled consistent data mapping across backend systems.
c. Contextual Classification: The AI system auto-classified document types using semantic analysis, allowing the automation layer to distinguish between similar forms and messages (for example, differentiating between MT103 customer credit transfers and MT199 free-format messages).
d. Automated Validation and Routing: Extracted data was validated against internal databases, such as client records and historical tax filings, to flag anomalies or missing fields. Once validated, the data was routed to relevant tax reporting, transaction processing, or compliance units.
e. Human-in-the-Loop Review: For complex or ambiguous cases, the system flagged documents for human review while learning from corrections to improve accuracy over time. Feedback loops ensured continuous model refinement.
f. Integration with Workflow Platforms: The AI system was integrated with Citi’s global operations workflow tools, enabling seamless hand-offs, tracking, and audit logging for internal governance and external regulatory reporting.
Results
The use of intelligent document processing technologies brought measurable improvements in efficiency, accuracy, and compliance in Citigroup’s operations:
a. Processing Time Reduction by 60%: Tax documentation and SWIFT message processing turnaround times dropped from an average of 48 hours to under 20 hours in many regions, improving client experience and operational agility.
b. Accuracy Improvement to 95%+: The AI models reached over 95% accuracy in data extraction and classification tasks, reducing manual errors and minimizing the need for rework.
c. Cost Savings through Automation: Automation of over 500,000 tax-related documents and SWIFT messages annually led to significant cost savings in labor, especially across Citi’s offshore processing hubs.
d. Regulatory Compliance Gains: Automated checks helped ensure adherence to FATCA and CRS requirements, minimizing the risk of non-compliance penalties and enhancing audit readiness.
e. Scalable Global Rollout: The AI-enabled IDP system was deployed across operations in Asia, North America, and EMEA, supporting multiple languages and local tax formats, thus creating a uniform global process.
Takeaway
Citigroup’s use of AI-driven intelligent document processing showcases the power of automation in optimizing back-office operations. By tackling complex, high-volume document workflows with advanced OCR and NLP technologies, Citi has significantly improved both speed and accuracy in handling tax and client transaction documents.
This transformation not only delivers cost and time benefits but also reinforces regulatory compliance and customer trust. As document processing remains a core challenge for large financial institutions, Citi’s deployment of AI in this domain sets a benchmark for scalability, reliability, and operational excellence in financial services.
Case Study 9: AI-Enhanced FX Trading Algorithms on Citi Velocity
Objective
Citigroup aimed to enhance the performance, precision, and client satisfaction of its foreign exchange (FX) trading services by embedding artificial intelligence into its Citi Velocity platform. FX trading is one of Citi’s largest revenue-generating businesses, operating across more than 80 currencies and serving institutional clients such as hedge funds, corporations, and asset managers.
Traditional algorithmic trading models, while fast and rule-based, lacked the adaptability to respond in real time to complex market conditions, sudden volatility, and fragmented liquidity. Citi needed AI-enhanced solutions that could analyze large datasets, learn from evolving patterns, and dynamically adjust trading strategies in milliseconds to optimize execution quality and minimize market impact. The goal was to deploy AI-powered FX algorithms capable of delivering better pricing, lower slippage, and smarter order execution—while improving transparency and performance reporting for institutional clients using Citi Velocity.
Implementation of AI
Citigroup integrated a range of artificial intelligence techniques into the development and deployment of its FX trading algorithms on Citi Velocity.
Key components of the implementation included:
a. Machine Learning for Market Signal Detection: Citi trained supervised machine learning models on historical trade data, market depth, order book dynamics, and macroeconomic indicators to detect high-probability trading signals. These models continuously adapt as new data flows in, enabling the system to anticipate price movements more accurately.
b. Adaptive Order Routing: AI-powered algorithms monitor multiple liquidity pools and venues in real time. They dynamically adjust order routing strategies to achieve the best price and speed of execution based on market conditions, latency, and volume fluctuations.
c. Reinforcement Learning for Strategy Optimization: Citi employed reinforcement learning techniques to test and optimize trade execution strategies in simulated market environments. The models learn from each trade, fine-tuning parameters such as trade size, slicing, and timing to minimize slippage and information leakage.
d. Anomaly Detection and Risk Controls: AI models monitor trading behavior for anomalies and potential market manipulation in real time. These systems provide alerts and trigger pre-defined risk thresholds to ensure compliance and guard against errant trades.
e. Personalized Execution Algorithms for Clients: Citi offers customizable execution strategies tailored to client-specific trading objectives, risk tolerances, and historical behaviors. AI models analyze individual client profiles to suggest optimal trading strategies and benchmark them against peer activity.
f. Natural Language Generation for Trade Summaries: After execution, Citi uses generative AI to produce post-trade analytics and summaries in plain language, helping clients better understand how their trades were executed, the rationale behind strategy decisions, and the achieved cost savings.
Results
Citi’s integration of AI into FX trading via Citi Velocity delivered measurable improvements in both execution quality and client experience:
a. Execution Cost Reductions: AI-enhanced algorithms helped reduce average execution costs by 15–20% across high-volume institutional orders by improving fill ratios and minimizing slippage.
b. Speed and Precision Gains: Real-time AI-driven order routing reduced latency by up to 30 milliseconds in fast-moving market conditions, giving clients better price access during periods of volatility.
c. Client Satisfaction Improvements: Personalized trading strategies and transparent post-trade analytics contributed to a higher client retention rate and improved Net Promoter Scores (NPS) among large institutional users.
d. Increased Trading Volumes: Improved algorithmic performance and service quality led to a significant uptick in FX trading volumes on Citi Velocity, especially among hedge funds and corporates seeking better execution.
e. Regulatory Alignment: AI-based monitoring tools ensured all trades adhered to internal risk controls and global regulatory requirements, including MiFID II and Dodd-Frank mandates.
Takeaway
Citigroup’s use of AI in FX trading through Citi Velocity represents a leap forward in the application of machine intelligence to capital markets. By embedding machine learning, reinforcement learning, and real-time analytics into trading algorithms, Citi has transformed the execution process into a smarter, more adaptive, and client-centric service.
This approach not only boosts trading performance and transparency but also helps Citigroup differentiate itself in a highly competitive FX market. As AI continues to redefine how financial markets operate, Citi’s innovation in algorithmic trading cements its role as a leader in the next generation of electronic trading solutions.
Case Study 10: Generative AI Coding Assistants Empowering 30,000 Citi Developers
Objective
Citigroup set out to boost software development productivity and modernize its global technology infrastructure by deploying generative AI coding assistants across its internal developer ecosystem. With over 30,000 developers globally, Citi maintains a complex technology landscape supporting functions like risk management, trading, digital banking, and regulatory reporting.
Citi’s legacy systems and regulatory demands often slowed the development cycle, leading to delayed product launches and increased maintenance burdens. Developers spent considerable time writing boilerplate code, fixing bugs, performing security checks, and interpreting legacy documentation. Citi needed a solution to streamline the software lifecycle, reduce friction in code maintenance, and enhance developer experience—without compromising compliance and security standards. The objective was to use generative AI to assist developers in real time by recommending code, auto-generating documentation, detecting vulnerabilities, and accelerating testing—ultimately reducing development time and increasing innovation velocity.
Implementation of AI
Citi integrated generative AI coding tools into its development environment by leveraging enterprise-grade platforms and fine-tuned large language models (LLMs) that were aligned with the bank’s internal governance standards.
Key components of the implementation included:
a. AI Pair Programming Assistants: Generative AI tools such as GitHub Copilot and internal LLM-based coding models were embedded into Citi’s integrated development environments (IDEs). These tools provided real-time code completions, syntax corrections, and logic suggestions tailored to the bank’s proprietary libraries and APIs.
b. Legacy Code Modernization: AI was used to interpret and refactor decades-old COBOL and Java code, translating legacy modules into modern languages like Python or Kotlin. This accelerated the migration of critical systems to cloud-native architectures.
c. Secure Code Generation: The AI tools were configured with built-in policies to detect and prevent insecure coding practices such as hardcoded credentials, SQL injection risks, or non-compliant encryption methods. Suggestions were filtered through Citi’s secure coding guidelines before being presented to developers.
d. Auto-Generated Documentation: LLMs were deployed to read through codebases and produce inline documentation, API references, and usage summaries—helping new developers onboard faster and reducing dependency on outdated documentation.
e. Test Case Generation: AI models generated unit tests, integration tests, and regression scripts from code context, ensuring better code coverage and fewer bugs reaching production.
f. Developer Training Integration: Citi’s internal learning platforms incorporated generative AI assistants for training and upskilling, offering interactive tutorials and real-time code explanations personalized to a developer’s role and skill level.
Results
The adoption of generative AI coding assistants brought tangible improvements across Citigroup’s global technology teams:
a. Productivity Gains of 35–50%: Developers reported spending less time on repetitive tasks and more time on high-impact features. Coding output increased significantly without a rise in error rates.
b. Faster Time to Market: Application development cycles were shortened by 20–30%, allowing Citi to accelerate feature releases for its digital products across consumer and institutional platforms.
c. Reduced Technical Debt: Legacy code modernization efforts using AI resulted in a 40% reduction in outdated code modules, improving maintainability and integration with modern systems.
d. Enhanced Developer Satisfaction: Internal surveys indicated higher satisfaction scores due to reduced cognitive load, improved onboarding, and more enjoyable coding experiences.
e. Strong Security Posture Maintained: AI recommendations were continuously monitored and validated against Citi’s internal security policies and compliance frameworks, ensuring that code remained audit-ready and resilient against vulnerabilities.
Takeaway
Citigroup’s deployment of generative AI across its 30,000-strong developer workforce illustrates how enterprise-grade AI can revolutionize large-scale software engineering in a highly regulated environment. By embedding intelligent coding assistants into every stage of the development lifecycle, Citi has significantly improved agility, code quality, and innovation capacity.
More importantly, this transformation supports Citi’s broader digital strategy to compete with fintechs and technology-native banks. The initiative not only future-proofs its core systems but also empowers its engineering teams to build faster, safer, and smarter—setting a new standard for AI adoption in global banking technology.
Conclusion
Citigroup’s AI portfolio underscores a deceptively simple insight: technology is transformative only when it is trusted and tightly aligned with strategic priorities. Across 10 distinct domains, the bank has reclaimed analyst hours, cut fraud noise nearly in half, shrunk regulatory impact assessments from weeks to days, accelerated mainframe retirement plans, and elevated customer satisfaction—all while operating under the industry’s strictest guard-rails. The common threads are clear: executive sponsorship, cloud perimeters that keep proprietary data inside, human-in-the-loop oversight, and metrics that chase tangible economic value rather than buzz-word compliance. By making governance a feature, not a burden, Citi shows regulators that responsible AI can enhance—not erode—control. By measuring “capacity created” and “minutes saved,” it gives shareholders a vocabulary for ROI that extends beyond technology spend. And by rolling tools out in concentric rings, it protects both customers and brand as models mature. The next milestones—real-time payment screening, global reg-code coverage, and full-stack prompt auditability—signal that transparency will remain the guiding star. For any organisation shaping its own AI agenda, Citi’s experience offers a final lesson: start with problems worth solving, architect for accountability, and let results compound. The future will favor those who make that formula a habit.
