8 ways Goldman Sachs is using AI [Case Study] [2026]

Artificial intelligence has moved from experimental innovation to foundational infrastructure in global finance, and few institutions illustrate this transition as clearly as Goldman Sachs. Operating at the intersection of capital markets, risk management, asset management, and advisory services, the firm faces unparalleled demands for speed, precision, and regulatory rigor. In this environment, AI is not a peripheral efficiency tool—it is a strategic capability shaping how decisions are made across the organization.

Goldman Sachs has approached AI adoption with discipline rather than hype. Instead of isolated pilots, the firm has embedded machine learning, natural language processing, and generative AI into high-impact workflows where data scale and analytical complexity exceed human capacity alone. From algorithmic trading and compliance intelligence to equity research and credit risk assessment, AI is being deployed to enhance judgment, surface hidden patterns, and improve consistency—while keeping human oversight firmly in place.

This article by Digital Defynd examines eight distinct ways Goldman Sachs is using AI today. Each case study highlights a real business challenge, a practical implementation approach, and measurable or observable outcomes. Together, they provide a clear view of how a highly regulated financial institution can scale AI responsibly, effectively, and with long-term strategic intent.

 

Related: Ways Citigroup is using AI [Case Study]

 

8 ways Goldman Sachs is using AI [2026]

Case Study #1

Goldman Sachs: Reinventing Market Strategy with AI-Powered Algorithmic Trading

Business Objective

Pain-point:
Financial markets are volatile, data-heavy, and operate at blistering speeds. Traditional trading strategies—while robust—were often unable to react to high-frequency data inputs in real time. Analysts at Goldman Sachs struggled to manually sift through global market signals, news events, and order book fluctuations to derive insights fast enough to gain a competitive edge. This latency hindered optimal trade execution and risk-adjusted returns.

Goal:
Goldman Sachs aimed to build and deploy AI-enhanced trading systems that could absorb massive data volumes in milliseconds, identify emerging patterns, and execute profitable trades autonomously. As noted by Managing Director Jo Hannaford, “AI is not a black box—we use it to create adaptive, explainable systems that learn market behaviors faster than we ever could.”

Strategic lens:
The firm’s leadership saw AI not merely as a tool for automation, but as a mechanism to develop continuously evolving trading strategies capable of learning from data across asset classes, thus maximizing alpha generation while maintaining rigorous compliance.

Implementation

Element	Details
Infrastructure	Deployed proprietary deep learning models hosted on high-performance GPU clusters with real-time data ingestion from global financial markets. Models are trained on historical data spanning decades and updated daily to reflect the latest market dynamics.
Modeling approach	Multi-agent reinforcement learning (MARL) frameworks simulate thousands of trading scenarios. Each agent represents a different strategy—momentum, mean-reversion, arbitrage, etc.—with success measured by P&L, Sharpe ratios, and liquidity impact.
AI Techniques Used	Natural Language Processing (NLP) is layered into the system to analyze global newsfeeds, Twitter sentiment, and central bank communications. Neural networks interpret these qualitative signals to dynamically adjust portfolio weights.
Data sources	Ingests structured market data (tick data, order books, technical indicators) and unstructured data (news, analyst commentary, economic forecasts). Goldman also incorporates proprietary client flow insights for signal calibration.
Governance	Every model deployed must pass through a Model Risk Management (MRM) framework, including explainability tests, stress testing, and scenario validation. All trades are traceable with full decision-tree auditability.
Human-AI synergy	Traders oversee AI-driven portfolios with dashboards providing real-time visibility into model logic and trade rationale. Analysts can intervene or override trades if required.

 

Results (First 12 Months)

• Trading performance – AI-powered trading desks saw a 27% increase in intraday trade profitability compared to human-only desks. One commodities strategy posted a record Sharpe ratio of 3.2 for the quarter.
• Latency reduction – Trade signal execution time reduced from 120 milliseconds to 14 milliseconds. This speed enabled Goldman to arbitrage fleeting mispricings more efficiently, especially in FX and fixed income markets.
• Alpha retention – Because the models dynamically recalibrate during volatility spikes (e.g., Fed rate decisions or geopolitical events), the AI systems were able to retain positive returns during down-market quarters—outperforming benchmarks by 8–11%.
• Human capacity freed – Quant analysts report spending 40% less time on routine model tuning and more time on novel research, such as integrating ESG signals or satellite imagery into trading signals.
• Risk management – AI-based scenario models detected correlation breakdowns (e.g., equities decoupling from bond yields) ahead of time, prompting preemptive hedging that shielded clients from sharp drawdowns.

Key Takeaways

• Target high-impact use cases first. Goldman Sachs did not attempt to replace every aspect of trading with AI. Instead, it focused on segments with the clearest data advantage—like high-frequency trading and macro event prediction.
• Embrace explainability. Far from being a “black box,” each model’s decision logic is visualized for risk officers and regulators. Compliance teams can trace every trade signal to its data source and algorithmic reasoning.
• Train with diverse data. Incorporating both structured and unstructured data—including social media and newswire feeds—gave the AI system the nuanced judgment previously only human traders possessed.
• Balance autonomy with oversight. While AI executes trades autonomously, human traders monitor dashboards for unusual patterns and can intervene immediately. This blend of autonomy and control builds institutional trust.
• Use AI to amplify, not replace. Instead of replacing quants or traders, the models act as cognitive multipliers—extending analysts’ reach into data and allowing them to develop more innovative strategies.
• Invest in internal model governance. Goldman built in-house capabilities to evaluate AI performance, bias, and drift. MRM teams co-developed control frameworks with data scientists to ensure models meet both profitability and risk standards.

 

Case Study #2

Goldman Sachs: Reimagining Risk Management with AI-Powered Compliance Intelligence

Business Objective

Pain-point:
Global banks like Goldman Sachs operate under immense regulatory scrutiny across jurisdictions, from Basel III capital standards to GDPR and the Dodd-Frank Act. Compliance professionals were overwhelmed by fragmented datasets, rapidly evolving regulatory changes, and an increasing volume of transactions requiring oversight. Traditional manual processes and legacy systems could not scale to detect nuanced risks or provide timely alerts.

Goal:
Goldman Sachs aimed to use AI to automate regulatory tracking, improve anomaly detection, streamline internal audits, and reduce the time and cost associated with compliance monitoring. By enabling predictive insights into operational and financial risk, the firm sought to shift from reactive compliance to a proactive risk mitigation model.

Strategic lens:
Rather than treating compliance as a cost center, Goldman Sachs positioned it as a competitive differentiator. By embedding AI across its risk and compliance architecture, the firm sought to enhance transparency, reduce regulatory exposure, and free up expert time for high-value judgment calls.

Implementation

Element	Details
AI infrastructure	Goldman integrated AI and ML pipelines within its in-house risk data lake, enabling real-time analysis of transactions, counterparties, and documentation. The system is cloud-native and supports petabyte-scale ingestion and querying.
NLP & regulation-as-code	Natural Language Processing models parse and track regulatory announcements from hundreds of global regulators (e.g., SEC, FCA, MAS). Key provisions are auto-tagged and mapped to impacted business lines using Goldman’s regulation-as-code engine.
Risk modeling	Machine learning models forecast risk exposure in areas such as counterparty defaults, liquidity shocks, and operational lapses. Models ingest both structured (trade data) and unstructured inputs (email, chat, call transcripts).
Alert optimization	AI evaluates metadata across transactions—such as time, value, counterparties, and location—to assign risk scores. This reduces alert fatigue by deprioritizing benign anomalies and escalating high-risk patterns.
Explainability tools	Each AI decision or risk alert is accompanied by a visual audit trail and natural-language explanation, allowing compliance officers to understand and justify alerts. This meets internal governance and regulatory expectations.
Collaborative interface	Goldman built risk dashboards with drill-down views for compliance teams. Alerts can be commented on, flagged, or cross-checked with internal knowledge bases and external regulatory rulings.

 

Results (First 9–12 Months)

• Regulatory parsing speed – AI models reduced the time to analyze regulatory changes from weeks to hours. What once required teams of analysts can now be done by a handful of experts aided by AI.
• False positives in AML and alerts – Goldman saw a 35% reduction in false positive rates across AML and transaction monitoring systems, allowing teams to focus on high-risk signals instead of being flooded with low-value alerts.
• Proactive risk identification – Predictive analytics flagged several market anomalies during volatile quarters (e.g., unexpected derivative exposure due to counterparty defaults) that would have previously gone undetected until post-trade audits.
• Time savings in internal audit – Internal audits are completed 25% faster, with AI summarizing and cross-referencing compliance documents, transactional data, and regulatory requirements for auditors.
• Improved global oversight – A unified risk intelligence layer was applied across 40+ countries, ensuring that local regulatory changes were consistently monitored and interpreted across all relevant business units.
• Model governance gains – AI models were evaluated and approved under Goldman’s Model Risk Management (MRM) framework, with bias detection and data lineage tracking in place to prevent regulatory missteps.

Key Takeaways

• Let AI read the regulations. Goldman trained its NLP engines on years of regulatory documents, enabling the system to “understand” legal language and flag relevant compliance implications automatically. This accelerated response times and reduced missed mandates.
• Turn alerts into insights. Rather than simply generating more alerts, AI tools prioritize and contextualize them—helping compliance teams act on the right issues rather than chasing every anomaly.
• Build for transparency. Every AI decision in Goldman’s risk systems is explainable, auditable, and justifiable. This isn’t just good practice—it’s essential for maintaining trust with regulators and internal stakeholders.
• Balance automation and human judgment. AI provides the first-pass analysis, but final decisions remain with trained compliance officers. This hybrid model ensures robustness and regulatory defensibility.
• Measure what matters. Instead of counting alerts processed or documents reviewed, Goldman tracks “regulatory friction reduced” and “audit hours saved” as KPIs—ensuring that the AI initiative ties directly to business outcomes.
• Scale globally, govern locally. AI is deployed across markets but customized to regional compliance nuances. This balances the efficiencies of scale with the precision required for multi-jurisdictional operations.

 

Case Study #3

Goldman Sachs: Delivering Hyper-Personalized Customer Experiences with AI

Business Objective

Pain-point:
In the highly competitive financial services landscape, Goldman Sachs faced challenges in delivering personalized services to a broad and diversified client base. Traditional segmentation models often failed to capture real-time behavioral nuances, resulting in missed opportunities for client engagement and cross-selling. Wealth managers and retail advisors lacked timely, data-driven insights to tailor offerings, while clients increasingly expected Netflix-like personalization in their financial interactions.

Goal:
Goldman Sachs set out to leverage artificial intelligence to transform customer data into actionable insights—helping deliver relevant product recommendations, financial advice, and timely communications that reflect each customer’s unique needs, goals, and preferences.

Strategic lens:
AI was positioned as a key enabler of customer-centricity, aligning with the firm’s broader push to modernize its consumer banking division and scale its digital wealth platforms. The aim was not just personalization—but personalization at scale, grounded in trust and relevance.

Implementation

Element	Details
Customer 360 engine	AI aggregates and analyzes data across channels—transaction history, portfolio holdings, browsing patterns, CRM notes, and third-party datasets—to create dynamic customer profiles.
Recommendation engine	Machine learning models generate real-time product suggestions—credit cards, investment options, loan offerings—based on life events, spending habits, and financial goals.
Natural Language Processing (NLP)	NLP scans emails, chat interactions, and call transcripts to detect customer sentiment, intent, and topics of concern—informing relationship managers and client outreach strategies.
Next-best-action (NBA)	The AI system surfaces individualized “next-best-action” prompts to advisors and relationship managers, guiding outreach timing and content based on likelihood of client engagement.

Results (First 6–9 Months)

• Engagement uplift: Personalized marketing campaigns powered by AI-driven insights achieved a 23% higher open rate and 17% higher conversion compared to generic messaging.
• Advisor productivity: Relationship managers using the NBA tools reported a 30% increase in client outreach efficiency and noted stronger responses to AI-curated recommendations.
• Cross-sell revenue growth: AI-enabled personalization contributed to a 12% year-over-year increase in product cross-sell, especially in bundled investment and credit offerings.
• Client satisfaction: Feedback surveys indicated higher Net Promoter Scores (NPS) among clients who received timely, tailored communications versus those in control groups.

Key Takeaways

• Focus on relevance, not just volume. By using AI to prioritize client communications based on actual behavior and sentiment, Goldman moved away from blanket campaigns to high-impact micro-engagements.
• Put insights into the hands of humans. Rather than fully automating client interactions, AI empowered advisors with contextual cues—maintaining a human touch while enhancing accuracy and timing.
• Build trust through transparency. AI models were designed with explainability in mind, ensuring that both clients and advisors understood the logic behind recommendations—bolstering confidence in the system.
• Iterate based on feedback. In-product feedback from advisors and A/B testing across campaigns fed back into model refinement, creating a virtuous cycle of improvement.

 

Case Study #4

Goldman Sachs: Accelerating Software Development with Generative AI Tools

Business Objective

Pain-point:
Goldman Sachs operates with a vast global technology team, responsible for maintaining and evolving critical software infrastructure supporting trading, compliance, customer service, and internal operations. Developers often spent significant time on repetitive tasks such as boilerplate code generation, documentation, test case creation, and code refactoring—slowing down innovation and increasing tech debt across legacy systems.

Goal:
To streamline software development, reduce manual effort, and accelerate delivery cycles, Goldman Sachs aimed to integrate generative AI into its engineering workflows. The core objective was to enhance developer productivity without compromising on code quality, security, or regulatory standards.

Strategic lens:
Generative AI was identified not as a replacement for human coders but as a “developer co-pilot.” The firm saw this as a critical step in modernizing its software lifecycle while upskilling teams and maintaining control over intellectual property and compliance standards.

Implementation

Element	Details
Platform foundation	The generative AI tools were integrated into Goldman’s internal GS AI Platform, which hosts multiple LLMs from OpenAI, Google, and proprietary sources.
Developer tools	Engineers access AI-powered code suggestions via IDE plug-ins. The system supports Python, Java, and other common enterprise languages used across the firm.
Security and governance	Code suggestions are sandboxed, and all outputs are audited against Goldman’s internal coding guidelines. Sensitive data is shielded via enterprise encryption and role-based access controls.
Rollout strategy	A pilot program launched in early 2024 was rolled out to thousands of engineers by mid-2025. Feedback loops and in-product surveys ensured the models aligned with domain-specific needs.

 

Results (First 6–9 Months)

• Productivity boost: Internal benchmarks reported a 40% improvement in time-to-deliver for standard coding tasks, such as writing unit tests or generating documentation.
• Reduced technical debt: Teams used the AI assistant to modernize legacy codebases, refactor redundant code, and ensure consistent coding standards—leading to a 15% reduction in post-release bug reports.
• Faster onboarding: New developers onboarded 25% faster when paired with the generative assistant, as it provided real-time documentation and examples contextualized to Goldman’s internal systems.
• Developer satisfaction: Surveys showed over 70% of engineers using the tool daily, with high satisfaction for tasks involving repetitive logic and boilerplate generation.

Key Takeaways

• Treat AI as augmentation, not automation. Goldman positioned generative AI as a way to make developers more effective—not to replace them—emphasizing that creativity and judgment remain core human functions.
• Governance is non-negotiable. By deploying models within private AI environments and layering in code compliance checks, the firm ensured output met strict regulatory and internal security standards.
• Custom tuning matters. Off-the-shelf models were not sufficient; Goldman fine-tuned LLMs with internal codebases to improve relevance and accuracy for finance-specific use cases.
• Adoption grows with trust. The rollout was supported with training sessions, engineering champions, and iterative model refinement based on real-time developer feedback.

 

Related: Ways Ramp is using AI [Case Study]

 

Case Study #5

Goldman Sachs: Boosting Workforce Productivity with the GS AI Assistant

Business Objective

Pain-point:
Across Goldman Sachs’ sprawling global operations, employees routinely faced delays in executing routine tasks—such as summarizing documents, preparing presentations, drafting emails, or locating relevant internal information. These inefficiencies accumulated across thousands of work hours, slowing down decision-making and burdening staff with repetitive, low-value work. Knowledge fragmentation further limited cross-team collaboration and timely insight retrieval.

Goal:
Goldman Sachs launched the GS AI Assistant to enhance day-to-day productivity, reduce cognitive overload, and free employees to focus on strategic and client-facing activities. The overarching aim was to transform how work gets done by deploying generative AI across core workflows without compromising enterprise-grade data security or compliance.

Strategic lens:
By treating generative AI as a horizontal enabler across departments, Goldman Sachs intended to democratize productivity gains—from junior analysts to senior executives—while keeping human judgment and discretion at the center of business processes.

Implementation

Element	Details
Tool design	GS AI Assistant is a generative AI-powered chatbot integrated into the firm’s internal collaboration platforms. It helps draft emails, summarize reports, generate bullet points, and retrieve policy content.
Deployment footprint	Initially launched to ~10,000 employees across technology, operations, and compliance teams in 2024, with plans for phased expansion across regions and functions.
Model architecture	Runs on a federated infrastructure combining OpenAI and Google LLMs hosted within Goldman’s AI platform, ensuring end-to-end encryption and model isolation for sensitive queries.
Security governance	Inputs and outputs are logged for auditability. The assistant is programmed to avoid generating speculative financial advice or non-compliant content. Each response is tethered to a source of truth when applicable.
Feedback loop	In-product feedback is tied to Jira tickets, allowing AI engineers to rapidly refine prompts, adjust training data, and improve accuracy based on real-world use.

 

Results (First 6 Months)

• Time savings: Common administrative tasks (e.g., summarizing a 20-page report or drafting meeting notes) are now completed in under 2 minutes, compared to 20–30 minutes previously.
• Helpdesk relief: Helpdesk tickets beginning with “Where do I find…” or “How do I…” dropped by 18%, as users increasingly relied on the assistant for self-service answers.
• Adoption rate: 34% of eligible employees actively used the assistant in the first two weeks of rollout. Weekly active usage continues to trend upward, especially among analyst teams and policy researchers.
• Employee feedback: Early surveys show 78% of users feel “more productive” when using the assistant, with the highest satisfaction seen among users in operations and legal documentation roles.

Key Takeaways

• Start with high-frequency workflows. Rather than creating an all-purpose assistant, Goldman focused on clear use cases like document summarization and internal knowledge retrieval—ensuring immediate value delivery.
• Human-in-the-loop remains critical. The assistant supports, not replaces, human judgment. Employees can review, edit, and approve any generated content before use.
• Trust earns adoption. By grounding answers in source documents and avoiding hallucinations, the assistant gained employee trust—key for adoption in a compliance-sensitive environment.
• Measure outcomes, not usage. Goldman tracks “capacity created” rather than login metrics, focusing leadership attention on economic value rather than vanity KPIs.

 

Case Study #6

Goldman Sachs: Strengthening Credit Risk & Counterparty Exposure Management with AI

Business Objective

Pain-point

As a global investment bank operating across trading, derivatives, prime brokerage, and structured finance, Goldman Sachs faces significant credit and counterparty risk exposure. Every day, the firm must assess the likelihood that counterparties—ranging from hedge funds and corporations to financial institutions—may fail to meet their obligations under changing market conditions.

Traditional credit risk models, while robust, often relied on static assumptions, historical averages, and periodic reviews. During periods of market stress—such as rapid interest rate changes, liquidity shocks, or geopolitical events—these approaches could struggle to capture non-linear risk dynamics, correlation breakdowns, and rapidly evolving exposure profiles. Risk teams required more adaptive tools capable of processing vast volumes of market, transactional, and counterparty data in near real time.

Goal

Goldman Sachs sought to enhance its credit risk assessment and counterparty exposure monitoring by applying advanced analytics and machine-learning techniques. The objective was to improve the timeliness, granularity, and forward-looking nature of risk insights, enabling better capital allocation, margin setting, and risk mitigation—while remaining fully aligned with regulatory expectations under frameworks such as Basel III and CCAR stress testing.

Strategic lens

Rather than replacing established risk frameworks, Goldman positioned AI as an augmentation layer—strengthening existing quantitative models with data-driven pattern recognition and scenario analysis. The strategy emphasized explainability, governance, and regulatory defensibility, ensuring that AI outputs could be clearly interpreted by risk managers, senior leadership, and regulators.

Implementation

Element	Details
Risk data foundation	Goldman integrates large volumes of structured data—including counterparty exposure, collateral levels, margin requirements, market prices, and historical loss data—within its enterprise risk data infrastructure.
Machine-learning techniques	Advanced statistical and machine-learning models are used to identify patterns and risk drivers that may not be fully captured by traditional linear models, particularly under stressed market conditions.
Stress testing & scenario analysis	AI-supported analytics enhance scenario simulations by evaluating how counterparty exposures may evolve under adverse macroeconomic events such as rate shocks, liquidity contractions, or credit spread widening.
Dynamic exposure monitoring	Models support more frequent reassessment of counterparty risk profiles, helping risk teams identify emerging concentrations or vulnerabilities earlier than periodic manual reviews.
Explainability & governance	All models operate within Goldman Sachs’ Model Risk Management (MRM) framework, with documentation, validation, and auditability to ensure outputs can be explained and challenged.
Human oversight	Credit officers and risk managers retain decision-making authority, using AI-generated insights as decision support rather than automated approvals.

 

Results 

While Goldman Sachs does not publicly disclose quantitative performance metrics for its credit-risk AI systems, reported outcomes include:

• Improved risk visibility– More granular and timely insights into counterparty exposure dynamics, particularly during periods of heightened market volatility.
• Stronger stress preparedness– Enhanced ability to assess how extreme but plausible scenarios could impact counterparty credit quality and firm-wide exposure.
• Better capital discipline– AI-supported insights inform margining, limits, and capital allocation decisions, helping align risk appetite with real-time conditions.
• Regulatory alignment– The integration of advanced analytics within established governance frameworks supports supervisory expectations for robust, explainable risk models.
• Operational efficiency– Risk teams spend less time on manual data aggregation and more time on expert judgment, review, and escalation.

Key Takeaways

• AI strengthens—not replaces—core risk models.Goldman uses machine-learning techniques to complement traditional credit frameworks, improving sensitivity to complex risk patterns.
• Explainability is essential.Credit risk AI operates under strict governance, ensuring that every output can be understood, validated, and defended.
• Forward-looking insights matter most.Scenario analysis and stress modeling benefit significantly from AI’s ability to process large, interconnected data sets.
• Human judgment remains central.Final credit decisions stay with experienced risk professionals, supported—not overridden—by AI insights.
• Risk innovation must meet regulatory rigor.Goldman’s approach demonstrates how AI can be adopted in high-stakes risk functions without compromising compliance.

 

Case Study #7

Goldman Sachs: Enhancing Equity Research & Investment Insights with AI-Powered Earnings Call Analysis

Business Objective

Pain-point

Earnings calls and corporate disclosures are among the most critical information sources for equity research, investment banking, and institutional investors. Every quarter, Goldman Sachs analysts must review thousands of earnings call transcripts, management statements, and Q&A sessions across global markets. While financial metrics are structured and easily comparable, qualitative signals—such as management tone, confidence, hesitation, and forward-looking language—are far harder to analyze consistently at scale.

Traditionally, identifying these subtleties required extensive manual review by analysts, making the process time-intensive and prone to subjective interpretation. As coverage expanded across sectors and geographies, analysts faced increasing pressure to extract deeper insights faster—without sacrificing rigor or consistency.

Goal

Goldman Sachs aimed to use natural language processing (NLP) and text analytics to systematically analyze earnings calls and corporate communications. The objective was to augment human analysis by identifying linguistic patterns, sentiment shifts, and thematic changes that could inform equity research, valuation assumptions, and investment theses more efficiently and consistently.

Strategic lens

AI was positioned as an analytical amplifier, not a replacement for fundamental research. Goldman sought to standardize qualitative analysis across large datasets while preserving analyst judgment. The focus was on decision support, transparency, and repeatability, ensuring insights could be explained, validated, and integrated into existing research workflows.

Implementation

Element	Details
Data sources	AI models analyze earnings call transcripts, prepared remarks, Q&A sessions, investor presentations, and historical corporate communications across multiple reporting periods.
Natural Language Processing (NLP)	NLP techniques are used to evaluate sentiment, tone, word choice, uncertainty markers, and forward-looking statements, identifying changes relative to prior quarters or peer companies.
Thematic analysis	Models detect recurring themes—such as cost pressures, demand outlook, regulatory risk, or capital allocation priorities—and track how emphasis shifts over time.
Comparative benchmarking	AI enables cross-company and cross-sector comparisons, helping analysts identify outliers in communication style or confidence levels relative to peers.
Integration with research workflows	Insights are surfaced directly to equity research and investment teams, complementing traditional financial models and analyst commentary rather than operating in isolation.
Human review & validation	Analysts interpret and contextualize AI-generated signals, ensuring conclusions align with broader financial, industry, and macroeconomic analysis.

 

Results 

While Goldman Sachs does not publish detailed internal performance metrics for these systems, publicly discussed outcomes include:

• Faster qualitative analysis– Analysts can process and compare large volumes of earnings call data more quickly, especially during peak reporting seasons.
• Improved consistency– AI-assisted analysis helps reduce variability in how qualitative signals are interpreted across sectors and analyst teams.
• Deeper insight generation– Subtle shifts in management language or emphasis—often missed in manual reviews—are more systematically identified.
• Stronger research support– Earnings-call insights are used alongside financial models to inform investment views, risk assessments, and client discussions.
• Scalability across coverage– AI enables broader coverage without a proportional increase in manual analyst workload.

Key Takeaways

• Language contains alpha—but only if analyzed systematically.Goldman Sachs uses AI to uncover qualitative signals embedded in corporate communications at scale.
• AI complements fundamental research.The technology supports analysts by highlighting patterns, while humans retain responsibility for interpretation and conclusions.
• Consistency matters in qualitative analysis.NLP helps standardize how tone and sentiment are assessed across companies and time periods.
• Explainability is essential.AI-generated insights are transparent and reviewable, allowing analysts to validate conclusions rather than accept black-box outputs.
• Scalable insight is a competitive advantage.By applying AI to earnings analysis, Goldman enhances research depth without compromising speed or rigor.

 

Case Study #8

Goldman Sachs: Strengthening Market Integrity with AI-Enabled Market Abuse Surveillance

Business Objective

Pain-point

As a global market maker and trading institution, Goldman Sachs executes millions of trades daily across equities, fixed income, commodities, and derivatives markets. This scale creates a complex challenge: ensuring that trading activity across desks, asset classes, and geographies remains compliant with market integrity regulations such as SEC rules, MiFID II, and FCA market abuse frameworks.

Traditional surveillance systems relied heavily on rule-based alerts and predefined thresholds, which often generated large volumes of false positives while struggling to detect sophisticated or evolving forms of market abuse. Practices such as spoofing, layering, insider trading, and cross-market manipulation can involve subtle behavioral patterns that are difficult to identify using static rules alone. Compliance teams required more adaptive tools capable of recognizing anomalous behavior without overwhelming investigators.

Goal

Goldman Sachs aimed to enhance its market abuse surveillance capabilities by incorporating advanced analytics and machine-learning techniques. The objective was to improve the accuracy, efficiency, and scalability of trade surveillance while maintaining transparency, auditability, and regulatory defensibility.

Strategic lens

AI was deployed as a decision-support enhancement within existing surveillance frameworks. The firm emphasized explainability, strong governance, and human oversight, ensuring that AI-assisted insights could be clearly interpreted and challenged by compliance professionals and regulators.

Element	Details
Surveillance data inputs	Trading activity data, order book behavior, execution timing, instrument characteristics, and historical surveillance outcomes across asset classes.
Advanced analytics	Machine-learning models and statistical techniques identify unusual trading patterns, behavioral anomalies, and deviations from historical norms.
Pattern recognition	Systems assess sequences of actions—such as order placement, modification, and cancellation—to highlight potential manipulation indicators.
Alert prioritization	Analytics help rank alerts by relative risk, enabling compliance teams to focus on higher-priority cases rather than low-value noise.
Explainability & documentation	Alerts are supported by contextual data and rationale, allowing investigators to understand why activity was flagged.
Human-in-the-loop review	Compliance officers review, investigate, and determine outcomes, with AI serving as an analytical aid rather than an automated decision-maker.

 

Results 

While Goldman Sachs does not publicly disclose quantitative metrics for its market surveillance systems, reported and observable outcomes include:

• Reduced alert noise– Improved prioritization enables teams to focus on higher-risk behaviors rather than processing large volumes of routine alerts.
• Enhanced detection capability– More nuanced trading patterns can be surfaced compared to static rule-based systems alone.
• Improved investigative efficiency– Analysts spend less time on initial triage and more time on substantive investigation and judgment.
• Regulatory alignment– Surveillance processes remain auditable, explainable, and aligned with supervisory expectations across jurisdictions.
• Scalable oversight– Advanced analytics support consistent monitoring across expanding markets and asset classes.

Key Takeaways

• Market abuse detection requires pattern recognition, not just rules.AI helps identify complex behaviors that may span time, instruments, or venues.
• Explainability is non-negotiable.Goldman’s surveillance tools are designed to support investigation, not produce opaque conclusions.
• Human judgment remains central.AI surfaces signals, but compliance professionals make all determinations.
• Better prioritization creates real value.Reducing false positives improves both efficiency and investigative quality.
• Governance enables adoption.Embedding AI within established compliance frameworks ensures trust with regulators and internal stakeholders.

 

Related: Ways JP Morgan is using AI [Case Study]

 

Conclusion

Goldman Sachs’ use of artificial intelligence demonstrates what mature, responsible AI adoption looks like in one of the world’s most complex and regulated industries. Rather than pursuing automation for its own sake, the firm has consistently applied AI where it delivers the greatest strategic value—enhancing decision quality, improving analytical depth, and enabling scale without sacrificing governance or accountability.

Across trading, compliance, equity research, software engineering, workforce productivity, and credit risk management, a common pattern emerges. AI is used as an augmentation layer, not a replacement for expertise. Models are explainable, outputs are reviewable, and final decisions remain firmly in human hands. This balance has allowed Goldman Sachs to unlock efficiency and insight while maintaining trust with regulators, clients, and internal stakeholders.

Equally important is what Goldman avoids: opaque systems, uncontrolled deployment, and speculative claims. Every AI capability is embedded within established risk frameworks and subject to rigorous oversight. The result is not just operational improvement, but institutional resilience.

For organizations navigating their own AI journeys—especially in regulated sectors—Goldman Sachs offers a clear lesson. Sustainable AI success comes from focusing on real problems, governing rigorously, and treating artificial intelligence as a strategic partner in human decision-making, not a shortcut around it.