AI in Data Management: Top Use Cases You Need To Know

Introduction

Data management today presents significant challenges, with organizations facing vast datasets, fragmented systems, and the demand for real-time, accurate insights. AI is emerging as a powerful solution, driving automation, intelligence, and actionable insights. This guide explores how AI is transforming data management, offering practical benefits while addressing key implementation challenges.

What is AI and Why Does It Matter in Data Management?

Definition of AI and Its Core Technologies

Artificial Intelligence (AI) refers to computer systems that simulate human intelligence by performing tasks such as learning, reasoning, and decision-making. Its foundational technologies include machine learning, natural language processing (NLP), and computer vision, which enable systems to process data, recognize patterns, and adapt over time. These capabilities allow AI to handle complex tasks traditionally performed by humans with greater speed and scalability. 

In data management, AI plays a vital role by automating core processes like data ingestion, classification, and validation. Machine learning models can detect anomalies or predict trends across large, complex datasets, reducing the need for manual oversight. NLP enables systems to extract and structure information from unstructured sources such as documents, emails, or logs, improving data accessibility and governance.

The Growing Role of AI in Transforming Data Management

AI is transforming how organizations manage their data by automating repetitive tasks and improving overall accuracy. Tasks such as data ingestion, cleansing, and validation, once reliant on manual labor, are now executed faster and more reliably through AI-driven processes. This reduces the time spent on routine operations and enhances the consistency of data across systems. 

AI is also redefining data integration by intelligently recognizing patterns and relationships between datasets. Instead of manually creating complex ETL workflows, AI systems can adapt to evolving data structures and automatically map new connections. This flexibility accelerates onboarding of new sources and minimizes disruptions from changing data formats. 

Analytics and insight generation are becoming more accessible with AI-enabled platforms. These tools identify hidden trends and correlations within data, allowing teams to make decisions based on real-time, predictive insights. As a result, organizations can shift from reactive reporting to proactive strategy development with greater confidence.

Key Statistics and Trends Highlighting AI Adoption in Data Management

AI adoption in data-related functions is accelerating. McKinsey reports that 78% of companies were using AI in at least one business function in 2024, up from 55% in 2023. In IT, usage rose from 27% to 36% in just six months, reflecting growing integration into infrastructure. 

IDC projects global AI spending will reach $632 billion by 2028, driven by increased adoption across industries, including data management. Meanwhile, 45% of firms are already using AI in at least three business areas, showing deeper enterprise-wide usage. 

Despite growth, success varies. A Qlik study found only 11% of companies report successful outcomes from most AI initiatives, and 44% face a gap between AI expectations and real productivity gains.

Business Benefits of AI in Data Management

AI is driving measurable improvements in how organizations handle, access, and analyze their data. From boosting data quality to enabling predictive insights, it helps businesses operate more efficiently and make faster, smarter decisions.

1. Enhanced Data Quality

AI improves data quality by automatically identifying anomalies, duplicate entries, and incomplete records across large datasets. These tasks, once manual and time-consuming, are now handled swiftly and consistently by AI, ensuring cleaner and more reliable data. As a result, organizations can trust their data and reduce the risk of decision-making based on faulty information. 

Over time, AI systems become better at detecting and correcting errors by learning from previous patterns. This adaptive capability allows for proactive quality management, catching issues before they escalate. The result is less time spent on data cleanup and more confidence in data-driven operations.

2. Accelerated Data Integration

AI streamlines data integration by recognizing and mapping relationships between diverse data sources without requiring manual schema definitions. This significantly speeds up the process of connecting new systems and maintaining existing pipelines. Businesses gain quicker access to unified data, supporting faster analytics and reporting. 

As data structures evolve, AI can automatically adjust to changes such as new fields or formats. This self-adjusting integration reduces the burden on engineering teams and keeps data flowing smoothly. It ensures resilience in dynamic data environments where change is constant.

3. Self-Service Data Access

AI makes data more accessible by enabling users to search and query information using natural language. This empowers non-technical teams to independently find the data they need, reducing reliance on IT support. With faster access to relevant insights, teams can make decisions more quickly and confidently. 

Additionally, AI enhances discovery by suggesting related datasets and frequently used resources. These personalized recommendations make it easier to navigate large data environments and encourage wider data usage across the organization. The result is a more data-driven culture with improved agility. 

Discover how foundational data management practices fuel long-term growth in our guide to data-driven success for small businesses.

4. Intelligent Data Governance

AI strengthens governance by automating the tracking of data usage, lineage, and compliance with internal and external policies. This ensures organizations meet regulatory standards like GDPR and CCPA while minimizing manual oversight. By automating governance tasks, companies reduce the risk of non-compliance and increase audit readiness. 

AI also helps prioritize governance efforts by highlighting areas of high risk or unusual behavior. This allows data teams to focus on critical issues rather than sifting through every alert. The result is more efficient, focused governance that protects data without slowing operations.

5. Advanced Analytics and Insights

AI enhances analytics by quickly detecting trends, patterns, and anomalies in vast datasets. This allows businesses to uncover valuable insights that would be difficult or time-consuming to find manually. Decision-makers benefit from real-time intelligence that supports proactive strategies. 

Beyond diagnostics, AI enables predictive analytics – forecasting future events such as customer behavior, inventory needs, or financial outcomes. These insights inform long-term planning and help organizations respond to challenges before they arise. Ultimately, AI transforms analytics from reactive reporting into a strategic advantage. 

Learn how AI can extract value from unstructured sources like emails and documents in our ultimate guide to unstructured AI.

Challenges Facing AI Adoption in Data Management

Despite its advantages, AI adoption in data management presents practical hurdles. These challenges can slow implementation and limit the value organizations gain without the right infrastructure, skills, and governance in place.

1. Fragmented Data Ecosystems

Many organizations store data across legacy systems, cloud platforms, and disconnected spreadsheets, making it difficult for AI to access a complete and unified dataset. This fragmentation undermines AI’s ability to deliver reliable results, as insights drawn from incomplete or inconsistent data are inherently limited. Without centralized access, automation and analytics cannot operate at full potential. 

Integrating siloed systems requires cross-functional alignment and standardization of formats and metadata. Without these foundations, AI initiatives often fail to scale effectively.

2. Poor Data Quality

AI systems are only as effective as the data they’re trained and operated on. Inaccurate, outdated, or poorly labeled data can lead to incorrect predictions, flawed insights, and failed automation efforts. When foundational data is unreliable, AI’s outputs become just as questionable. 

Fixing data quality issues requires clear ownership, governance practices, and continuous monitoring. Many organizations lack the discipline or resources to maintain this consistently.

3. Skills and Expertise Gaps

Deploying AI in data environments requires a blend of expertise in machine learning, data engineering, and domain-specific knowledge. Unfortunately, this combination is rare and often expensive to acquire. As a result, companies struggle to build or scale AI teams internally. 

Even when tools are available, lack of understanding breeds distrust in AI outputs. Organizations must invest in both technical training and user education to drive adoption.

4. Compliance and Regulatory Risks

AI tools often automate data processing at scale, increasing the risk of violating privacy or security regulations. Without clear oversight, these systems might unintentionally expose sensitive data or bypass established access controls. Regulatory non-compliance can lead to fines, reputational damage, and operational disruption. 

To stay compliant, AI deployments must be built with strict governance, audit trails, and role-based access. This adds complexity and requires close collaboration with legal and compliance teams. 

Discover strategies for aligning digital systems with regulatory demands in our guide to navigating compliance in digital finance.

5. Measuring ROI and Business Impact

Implementing AI in data management involves significant investment in tools, infrastructure, and training. Yet many organizations struggle to quantify the return on these investments, especially in early stages. Without clear metrics, stakeholders may hesitate to commit further resources.  

Organizations need focused pilots and measurable outcomes to demonstrate value. Without them, AI projects risk stalling or losing executive support.

Specific Applications of AI in Data Management

1. Augmented Analytics

Augmented analytics leverages AI and machine learning, including natural language processing, to automate the preparation, visualization, and interpretation of data. It empowers business users to extract insights without relying on technical experts. By processing both structured and unstructured data at scale, it enables faster, more informed decision-making. 

The system pulls data from diverse sources, applies algorithms for anomaly detection and forecasting, and enables natural language queries. AutoML helps identify the most accurate model for a given scenario, with outputs visualized through platforms like Tableau or Qlik. This streamlines the entire analytics process and enhances accessibility. 

Organizations benefit from accelerated insights, broader analytics adoption, and enhanced self-service capabilities. Studies show 97% adoption rates and 87% improvement in efficiency with these tools. Ensuring model transparency and robust governance is essential to avoid misinterpretation and bias.

Real-world example:

Intuit used augmented analytics to empower analysts with auto-generated insights and guided visualizations. NLP-driven suggestions reduced dashboard creation time by 50%. Engagement and decision-making improved across teams.

2. AIOps (AI for IT/Data Operations)

AIOps applies machine learning and event correlation to automate IT operations, improving the speed and accuracy of incident management. It processes telemetry, logs, and events to detect patterns and anomalies that humans often miss. These tools are essential for managing complex, high-volume environments. 

AIOps identifies root causes, predicts incidents, and can trigger automated responses using platforms like ServiceNow or Kubernetes. Supervised learning models reduce noise and improve detection accuracy over time. This results in more efficient IT operations and fewer disruptions. 

Key benefits include shorter incident resolution times, fewer outages, and lower operational costs. The AIOps market reached $11.7 billion in 2023 and continues to grow rapidly. However, it’s critical to manage risks related to data silos, automation bias, and alert fatigue.

Real-world example:

Vitria’s AIOps solution helped a telecom firm boost availability by 60% and halve staffing needs. Automated alerts cut manual errors, enhancing SLA compliance. Performance and reliability significantly improved.

3. Dark Data Transformation

Dark data refers to unstructured, untapped information, such as emails, PDFs, images, and call logs, that makes up about 90% of enterprise data. AI techniques like OCR, NLP, and computer vision help extract valuable information from these sources. This enables organizations to transform previously overlooked data into strategic assets. 

Using OCR, the system digitizes physical documents, while NLP tools extract entities, sentiment, and context. These insights are then integrated into data lakes or BI tools, enabling deeper analysis. Ensuring data privacy and compliance is a necessary part of this process. 

By uncovering hidden insights, AI-driven dark data transformation enhances operational efficiency and decision-making. It helps reduce data silos and exposes opportunities that were previously invisible. Still, it’s important to validate AI outputs and handle sensitive information with care.

Real-world example:

A global logistics firm used OCR and NLP to process scanned records. They transformed 80% of dark data and cut delivery issues by 20%. Route planning and operations became more data-driven.

4. Metadata Automation & Management

AI automates the process of generating, tagging, and organizing metadata, which traditionally required manual input. It analyzes data structure and usage patterns to produce metadata that’s consistent and accurate. This improves data discoverability and ensures better compliance with governance policies. 

Machine learning models identify relationships, schema patterns, and contextual relevance. These capabilities are embedded into catalog platforms like Alation, Collibra, and Informatica. The result is a more navigable and trustworthy data ecosystem. 

Automated metadata tools reduce the burden on data engineers, speed up onboarding for analysts, and strengthen governance. Catalog engagement increases as users find relevant data more easily. However, the models must be regularly updated to avoid misclassification.

Real-world example: 

Pfizer implemented AI-driven metadata tagging across its R&D environment, enabling more efficient data search and lineage tracking. The system improved metadata accuracy and discoverability by 65%. This led to faster research cycles and better data compliance.

5. Autonomous Data Products

Autonomous data products encapsulate data, metadata, APIs, and governance in self-contained units managed by domain-specific teams. AI helps enforce quality, monitor usage, and ensure compliance within each product. This approach is central to the data mesh paradigm. 

Each team independently builds and maintains its data pipelines, while AI provides real-time validation and anomaly detection. These products support scalable data architecture and reduce reliance on central teams. They foster faster innovation and stronger data ownership. 

This model increases agility, improves data quality, and enhances alignment with business needs. However, it requires strict interoperability standards and federated governance. Ongoing coordination is key to maintaining consistency across domains.

Real-world example: 

JPMorgan Chase deployed autonomous data products across its business lines using AI for monitoring and validation. Delivery timelines improved by 30%, and teams reduced dependence on central IT. This shift supported more responsive and scalable data operations.

6. Data Quality & Governance Automation

AI automates the detection and correction of data quality issues such as anomalies, schema mismatches, and duplication. It monitors pipelines in real time and flags issues for review or automated resolution. This reduces manual intervention and increases trust in analytics. 

These capabilities are integrated into data quality platforms like Talend, Bigeye, and Informatica. AI also supports governance by identifying sensitive data and enforcing access controls. This ensures regulatory compliance and protects business-critical data. 

Automated governance enhances data reliability, reduces error rates, and saves operational costs. Enterprises report up to 80% fewer data quality issues after implementation. Still, human oversight is essential to validate corrections and maintain accuracy.

Real-world example:

Vodafone implemented AI-based data quality checks across their billing and CRM systems, detecting and resolving 95% of anomalies. Billing errors dropped by 50%, and customer satisfaction rose by 20%. This greatly improved operational efficiency and data trust.

Examples of AI in Data Management

Real-world case studies demonstrate how leading organizations apply AI in data management to solve complex challenges. These examples highlight measurable gains in efficiency, governance, and analytics innovation.

Real-World Case Studies

1. Procter & Gamble: Scaling AI for Master Data & Compliance

P&G – operating 48 SAP instances globally – adopted AI-driven data governance and quality tools to streamline master data management and audit processes. These systems use machine learning to detect duplicates, enforce standards, and trace lineage across P&G’s enterprise data estate. The strategic outcome: faster audit readiness, reduced data errors, and increased organizational efficiency. 

In a collaboration with Harvard Business School’s D^3 Institute, P&G measured a 12% increase in cross-functional collaboration speed when using AI tools across workflows. P&G also reported significant improvements in data traceability and task automation across supply chain and R&D. Their broader “Data Excellence Journey” further highlights AI’s role in bolstering consumer insights and analytics.

2. Avista Utilities: Transforming Metadata Management with Alation

Avista, a major utility provider, implemented Alation’s AI-powered Data Catalog to index and surface thousands of AMI and DER datasets. By leveraging NLP and usage analytics, analysts now discover relevant datasets in minutes – formerly taking hours. This shift has enabled data-led initiatives in EV planning, weather forecasting, wildfire protection, and energy efficiency. 

Analysts, like Harrison Fine, affirm this transformation: “Now spend 80% of their time doing analysis” after integrating Alation. Avista credits metadata democratization for catalyzing new analytics, such as pinpointing EV charger concentrations and enhancing grid resiliency. These targeted insights have measurably improved forecasting accuracy and operational readiness.

3. VAST Data: Real-Time AI Insights with InsightEngine

VAST Data launched InsightEngine, merging fast flash storage, vector indexing, and live data embedding to support AI analytics at scale. This system ingests and semantically indexes structured, unstructured, and streaming data in real time, powering RAG workflows with secure, AI-ready datasets. 

Built on NVIDIA AI Enterprise and VAST’s unified architecture, InsightEngine enables live document retrieval, multimodal analysis, and consistent security enforcement across data types. Enterprises deploying it achieve faster AI inference, on-demand insights, and a data layer purpose-built for real-time decision-making.

Innovative AI Solutions

AI is driving a shift toward decentralized, agent-based data architectures that enable secure, real-time access without duplicating data. These intelligent agents execute queries live at the source, preserving access controls and improving data freshness, scalability, and governance. This approach addresses limitations in traditional centralized pipelines and enhances enterprise data strategy. 

Simultaneously, unified AI-native platforms are integrating storage, compute, and orchestration layers to support high-performance data operations. These platforms provide built-in intelligence for indexing, search, and access control, streamlining AI data readiness. Standards like the Model Context Protocol (MCP) are also advancing interoperability and secure tool integration across the AI ecosystem.

AI-Driven Innovations Transforming Data Management

Emerging Technologies in AI for Data Management

AI is rapidly redefining how businesses approach data management by introducing automation, intelligence, and speed at every level of the data lifecycle. One of the most transformative innovations is generative AI, which now assists in generating metadata, summarizing content, and classifying datasets with minimal human input. This streamlining not only reduces manual workload but also ensures that data is consistently tagged, searchable, and ready for advanced analytics. 

Another crucial advancement lies in the use of machine learning to identify patterns and relationships within massive datasets. These models continuously learn from incoming data, improving the accuracy of insights and enabling real-time decision-making. As a result, organizations can maintain cleaner data environments and respond faster to emerging trends or anomalies.

AI’s Role in Sustainability Efforts 

AI plays a growing role in making data management more sustainable by reducing unnecessary resource consumption. Through predictive analytics, AI can forecast data usage and automate decisions around storage, backup, and archiving. This reduces duplicate or unused data, leading to lower energy consumption and storage costs. 

Smart systems powered by AI also help optimize the performance of data infrastructure. They can adjust computing resources dynamically, shutting down idle systems and prioritizing efficient workflows. These adjustments not only save power but also support broader environmental and sustainability goals within digital operations.

How to Implement AI in Data Management

Implementing AI in data management isn’t just a tech upgrade, it’s a strategic shift that requires careful planning across people, processes, and platforms. Here’s how to move from interest to impact, step by step.

Step 1: Assessing Readiness for AI Adoption

Before introducing AI into your data workflows, it’s crucial to understand where your business stands in terms of data maturity. Start by identifying pain points like inconsistent metadata, manual tagging, or fragmented data storage. These areas often provide high-impact, low-risk opportunities for initial AI experimentation. 

Readiness isn’t just about technology, it’s about people and processes too. Leaders need to champion the change, and teams must be open to working with new systems. If internal resistance or lack of clarity exists, even the best AI initiative can stall before it begins.

Step 2: Building a Strong Data Foundation

For AI to deliver results, it needs access to clean, well-organized, and consistently labeled data. Businesses should focus on consolidating data sources, unifying formats, and resolving duplicates to reduce noise in the system. The better your data hygiene, the smarter and more reliable your AI outputs will be. 

It’s equally important to establish clear data governance frameworks. When your teams know who owns the data, how it’s managed, and where it resides, AI can operate more securely and efficiently. A strong data foundation is the backbone of sustainable, scalable AI adoption.

Step 3: Choosing the Right Tools and Vendors

Choosing an AI solution isn’t just about features, it’s about fit. Look for vendors that align with your data strategy and understand your industry’s regulatory and operational demands. The right partner will help you customize AI tools without disrupting your existing workflows. 

Be mindful of how your data is processed and protected. Transparency around data privacy, model training, and integration is key to long-term trust. A good vendor relationship should feel like a partnership, not just a software license.

Step 4: Pilot Testing and Scaling Up

Start small with a use case that has a clear problem, measurable goal, and defined scope. Pilots could involve automating data tagging, anomaly detection, or data quality scoring – areas that benefit from AI’s pattern recognition without risking critical operations. Early results provide insight and momentum. 

Once a pilot shows promise, scale thoughtfully. Use what you learn to refine your models, improve user workflows, and expand your governance policies. Gradual expansion reduces risk while proving the value of AI in real-time.

Step 5: Training Teams for Successful Implementation

Even the smartest AI needs informed users to maximize its potential. Invest in upskilling employees with a focus on practical use cases, ethical AI practices, and how automation complements their current roles. Confidence with the tools translates directly into adoption and ROI. 

Foster collaboration between technical teams and domain experts. Cross-functional knowledge-sharing leads to better models, stronger buy-in, and more innovative problem-solving. When teams are aligned and supported, AI becomes a trusted partner – not a disruptive force. 

For practical insights on how to lead successful AI integration, explore our guide to AI adoption strategies for tech leads.

Measuring the ROI of AI in Data Management

Key Metrics to Track Success

To understand the impact of AI on your data operations, focus on how AI improves speed and efficiency. Many organizations report faster data processing, shorter tagging times, and quicker discovery – often reducing task completion time by 30–50%. These changes increase agility and allow teams to act on insights sooner. 

Cost reduction is another key indicator. AI minimizes manual work, improves data quality, and decreases errors, which lowers operational risks and compliance costs. Over time, these gains add up to significant savings.

Case Studies Demonstrating ROI

Danone applied machine learning to improve demand forecasting, cutting forecast errors by 20%, reducing lost sales by 30%, and boosting promotional ROI by 10 points. These results came from better use of structured data and AI-powered insights, enabling smarter inventory and supply chain decisions. 

In public health, an AI anomaly detection system analyzed 5 million data points daily and accelerated review speed by 54×. This shift dramatically reduced manual workload while increasing trust in data-driven decisions.

Common Pitfalls and How to Avoid Them

One of the most common mistakes is assuming that AI will generate ROI automatically once implemented. Without clear goals, quality data, and team readiness, AI projects often struggle to move beyond the pilot stage or fail to produce measurable results. Leaders must ensure that performance metrics are in place from day one and that AI efforts are focused on solving real business challenges. 

Another pitfall lies in ignoring the importance of governance and training. If AI is layered on top of messy or unstructured data, its outputs will be inconsistent or misleading – wasting both time and investment. The key to long-term ROI is a disciplined approach: start small, measure everything, and scale only once the foundational pieces are solid. 

To ensure your AI initiatives deliver value, check out our guide on evaluating AI model performance and efficiency.

Future Trends of AI in Data Management

Predictions for the Next Decade

AI in data management is moving toward more autonomous, intelligent systems that require less human intervention. The rise of agent-based architectures and decentralized data products is redefining how organizations handle ownership, access, and governance. Over the next decade, we’ll see AI models embedded directly into data pipelines, making real-time decisions on quality, relevance, and lifecycle without waiting on centralized control. 

At the same time, natural language interfaces powered by large language models are becoming a new layer of access. Teams will interact with data using plain language, making insights more accessible and reducing the need for technical skills. As privacy regulations grow stricter, expect a surge in privacy-preserving AI and edge-based processing, ensuring data stays secure while still enabling powerful, automated insights.

How Businesses Can Stay Ahead of the Curve

To stay ahead in the evolving world of AI-driven data management, businesses must build a culture that values data literacy and cross-functional collaboration. This means equipping teams – not just IT – with the skills to understand, interpret, and act on data insights powered by AI. Companies that empower both technical and non-technical staff to use AI tools responsibly will see faster adoption and better results. 

It’s also essential to keep a close eye on emerging trends and evolving regulations. Partnering with vendors who prioritize scalability, security, and innovation helps ensure long-term flexibility. By continuously refining their data strategies and staying adaptive, businesses can turn AI from a competitive advantage into a core capability.

Conclusion

Key Takeaways

AI is fundamentally transforming data management by making processes faster, smarter, and more scalable. From automating metadata generation to enhancing data quality and discovery, AI allows organizations to unlock deeper insights while reducing manual effort. When implemented thoughtfully, it improves not only operational efficiency but also strategic decision-making across the business. 

Successful AI adoption in data management hinges on having clean, well-governed data, choosing the right tools, and empowering teams with the right skills. Businesses that align AI efforts with clear goals and track measurable outcomes can see tangible ROI – from cost savings to improved accuracy and speed. As the technology evolves, staying agile and forward-thinking will be key to turning data into a long-term strategic asset.

Moving Forward: A Strategic Approach to AI-Driven Transformation 

As AI reshapes the way organizations manage and utilize data, now is the moment to future-proof your data infrastructure with intelligent, scalable solutions. From streamlining data workflows to enhancing data accuracy and reducing compliance risks, AI is quickly becoming a cornerstone of operational excellence and innovation. 

At SmartDev, we help forward-thinking businesses harness the full potential of AI in data management. Whether you’re improving data quality, automating governance, or enabling real-time insights, our tailored solutions are designed to deliver measurable results that align with your goals. 

Explore our AI-powered software development services to learn how we design and implement systems that elevate your data operations and drive business value. 

Contact us today to begin your AI journey and transform your data into a true competitive advantage.

—

References:

1. The state of AI: How organizations are rewiring to capture value | McKinsey & Company
2. Worldwide Spending on Artificial Intelligence Forecast to Reach $632 Billion in 2028, According to a New IDC Spending Guide | IDC
3. How many companies use AI in 2025? Key statistics and industry trends | Hostinger
4. How do data catalogs simplify search & discovery? | Alation
5. Privacera’s Integration with Collibra Creates End-to-End Data Governance | Privacera
6. Configure and run data profiling for a data asset | Learn Microsoft
7. The Fastest, Most Scalable AI and Analytics Platform – Period. | VAST Data
8. VAST Data Adds New Capabilities to Deliver the First and Only Enterprise AI Data Platform for Real-Time Agentic Applications | VAST Data
9. The Future of Collaboration: How AI is Supplementing Teamwork and Innovation at P&G | Procter & Gamble
10. ​​RAG is dead: why enterprises are shifting to agent-based AI architectures | TechRadar
11. AI Case Study: Danone reduces forecast error and lost sales by 20 and 30 percent respectively and achieves a 10 point ROI improvement in promotions with machine learning | Best Practice AI
12. An AI-Based Public Health Data Monitoring System | arXiv