Analytics: How Modern Organizations Use Data to Make Decisions

Analytics has moved well beyond dashboards and periodic reports. Today, teams have access to predictive models, real-time analytics and conversational interfaces that let people ask questions of data using natural language.

Organizations are already putting these capabilities to work. In 2026 Snowflake-sponsored Omdia research, 59% of data analytics teams said they were using generative AI. At the same time, 96% of respondents said they still faced significant challenges putting generative AI into practice.

The disconnect between those two numbers reveals something important about modern analytics: New ways of working with data can make insights faster and more accessible, but they don’t remove the need for a strong data foundation. Reliable answers still depend on trustworthy data, consistent business definitions, the right level of freshness and an analytical approach suited to the decision being made.

As analytics continues to expand, the challenge is to connect tools to shared data, business context and data governance so that different analytical workloads can produce insight people can actually use.

Analytics uses data to measure performance, investigate changes, forecast outcomes and guide decisions. It encompasses the recurring work of preparing and modeling data, applying analytical methods and delivering results through reports, dashboards, applications and other interfaces.

The right analytical approach depends on the question being asked and how quickly the answer is needed. Descriptive and diagnostic analysis help teams understand past performance and investigate why it changed, while predictive and prescriptive methods help anticipate future outcomes and evaluate possible actions. Business intelligence (BI) gives users a recurring way to monitor and explore defined metrics, while real-time analytics supports decisions that depend on fresher data.

Increasingly, business questions are being explored through AI-assisted analytics interfaces. Natural-language querying, generated explanations and conversational follow-up can give users an easier, more flexible way to move through an analysis.

Teams use various analytics approaches to track performance over time, investigate new questions as conditions change and apply the results to operational and strategic decisions.

Watch a live demo of how Snowflake Interactive Analytics powers real-time analytics for APIs, apps and AI agents:

The terms analytics and data analysis overlap considerably in everyday use, and practitioners don’t always draw the boundary in the same place. Generally, however, data analysis refers to a specific activity, while analytics describes the broader capability that allows organizations to perform that work repeatedly and consistently.

• Data analysis involves examining data to answer a particular question. An analyst might investigate why customer churn increased last quarter, compare conversion rates across campaigns or determine which products generated the largest margin decline.
• Analytics includes the systems, processes and practices that make that work repeatable. It encompasses the data pipelines that supply information, the metric definitions that keep calculations consistent, the analytical methods used to investigate questions, the tools through which people access results and the governance required to maintain them over time.

Reporting overlaps with analytics, particularly in business intelligence, but typically focuses on presenting defined metrics and results. Analytics extends further into investigation, explanation, prediction and decision support. In practice, dashboards often include analytical functions, while analytics workflows frequently end in a report or visualization.

One common framework groups analytics into four types according to the question being asked.

• Descriptive analytics summarizes what has already happened. Revenue reports, website traffic trends, inventory levels and historical customer activity all fall into this category.
• Diagnostic analytics investigates the factors associated with an observed result. After identifying a decline in sales, for example, an analyst might segment the data by region, product or customer cohort to understand where the change occurred and what contributed to it.
• Predictive analytics uses historical data and statistical or machine learning methods to estimate future outcomes, such as demand, churn risk or equipment failure.
• Prescriptive analytics evaluates possible actions in light of objectives, constraints and expected outcomes. Optimization models, simulations and decision rules frequently support this work.

Across all four types, trustworthy data and stable business definitions provide the analytical foundation.

Analytics often starts before anyone opens a BI tool or writes a SQL query. The first task is deciding what questions need answers and how those answers will influence decisions.

A practical workflow usually includes five connected stages:

1. Define the objective: Establish the question, decision or outcome the analysis needs to support. “Why are sales down?” might turn into an investigation of whether lower transaction volume, pricing changes or product mix explain a year-over-year decline in a particular market. The objective determines which data is relevant, how fresh the result needs to be and what level of uncertainty the decision will tolerate.
2. Collect the relevant data: Depending on the question, the required information might include structured tables, semi-structured records or unstructured data such as documents and support transcripts. The source also shapes expectations around freshness, completeness and access.
3. Prepare the data: Data processing brings those sources into usable form through activities such as cleaning, transformation, joining and validation. At this stage, teams also need to understand whether fields contain the values analysts think they contain.
4. Analyze and interpret: Analysts apply queries, statistical methods, visualizations, models or other analytical techniques, then interpret the result in its business context. A mathematically correct result is still easy to misread when definitions, populations or assumptions are unclear.
5. Connect the result to a decision or action: Some analyses inform an immediate operational choice; others influence planning, refine a hypothesis or rule out an option. Either way, the result needs a destination beyond the analysis itself.

Real analytical work tends to loop through these stages. An unexpected result sends an analyst back to inspect source data, revise a definition or ask a narrower question, and a new decision creates new data, which later feeds another round of analysis.

“Ultimately, value shows up in decision velocity and organizational alignment,” says Josh Klahr, Snowflake’s Head of Product Management for AI-Powered BI. “The test is whether cross-functional teams spend less time debating whose metrics are right and more time making decisions from agreed business logic.”

An analytical result inherits the characteristics of the data underneath it. Missing records, stale values, inconsistent identifiers or incorrect field mappings all affect what the analysis represents, even when the query or model operates exactly as designed.

Teams typically assess data quality across several dimensions:

• Accuracy: Whether values correctly represent the underlying event or entity
• Completeness: Whether required values and records are present
• Consistency: Whether the same information agrees across systems and data sets
• Freshness: Whether the data is current enough for the decision it supports
• Validity: Whether values conform to expected formats, ranges and rules
• Uniqueness: Whether duplicate records distort counts or other calculations

Before defining rules for those dimensions, teams need to profile the current data and understand its actual distributions, null rates, formats and relationships. A freshness threshold that works for quarterly financial planning, for example, would be inappropriate for fraud detection.

Data quality alone doesn’t guarantee analytical consistency, however. Two teams might work from perfectly valid records but report different revenue figures if they define annual recurring revenue differently, for example. One dashboard might include expansion revenue immediately, another might wait until a contract start date, and a third might exclude a particular product category. Without an agreed definition, all three calculations could be technically correct while producing different answers to the same business question.

A semantic layer addresses that problem. Rajhans Samdani, Principal Software Engineer at Snowflake, explains: “Semantic models are the shared interface that lets BI dashboards and AI agents operate on the same definitions.” Metric definitions, entity relationships and other semantic context give analysts, dashboards, applications and AI systems a common interpretation of the underlying data.

A semantic layer is especially crucial for natural-language interfaces. When definitions disagree, the system can reproduce misaligned answers at greater scale, and it can be more challenging to identify the problem. Semantic consistency is playing a larger operational role as more users interact with data through conversational and agentic systems.

Business intelligence sits within the broader analytics discipline, concentrating heavily on reporting, dashboards, visualization and access to recurring business metrics.

A BI environment might give a finance team a monthly revenue dashboard, provide sales leaders with pipeline metrics and let operations teams explore inventory trends. Because those questions recur, teams generally define the underlying metrics and data models in advance, giving business users a consistent way to inspect performance without rebuilding each analysis from scratch.

Large language models (LLMs) and generative AI give users new ways to query and explore analytical data. Alongside filters and dashboards, users can interact with business data through natural-language questions, generated visualizations and conversational follow-up.

Real-time analytics is often framed around latency: milliseconds, seconds or minutes from an event occurring to its appearance in an analytical result. Whether or not a use case requires real-time analytics and what the appropriate latency target is, however, depends on the decision the result supports. If a five-minute-old answer leads to the same decision as a five-second-old one, reducing latency further provides little practical value.

Tighter freshness requirements generally demand continuously updated data, low-latency processing and analytical infrastructure designed to handle queries while new records are still being ingested.

To enable real-time analytics, the system has to manage both latency and concurrency: data needs to reach the analytical layer quickly, while many users or applications query continuously changing information without creating unacceptable delays.

For workloads whose decisions tolerate older data, batch processing remains simpler and often less expensive.

Self-service analytics moved a lot of analytical work closer to business users by giving them governed dashboards, drag-and-drop exploration, filters and ad hoc querying. AI has significantly expanded self-service capabilities by letting users formulate questions in natural language and continue the analysis conversationally.

Generative AI supports much of the experience through text-to-SQL, generated explanations and conversational querying, while machine learning supports predictive and prescriptive methods that handle tasks such as forecasting outcomes and evaluating possible actions.

AI can also assist analysts with tasks such as identifying patterns, generating explanations and summarizing results — capabilities often grouped under augmented analytics.

As more of the analytical path moves from predefined navigation to generated queries and multi-step workflows, semantic context and governance carry more of the burden of keeping results aligned with business definitions.

“Introducing conversational AI changes the work from designing dashboards and reports to managing context,” says Klahr. “The system has to understand the user’s intent and apply the right business definitions and analytical rules to the question being asked.”

Industry changes the context around analytics more than the underlying analytical methods. The important differences usually show up in the data being analyzed, the freshness required for the decision and the governance or regulatory constraints surrounding its use.

An analytics environment needs several capabilities across the path from raw data to an analytical result. Data has to be ingested and transformed, stored and processed, interpreted through consistent business definitions and made available through interfaces suited to different users and workloads.

For analytics specifically, those capabilities commonly include:

• Storage and compute: Process the structured, semi-structured and unstructured data used for analysis.
• Data ingestion and transformation: Bring source data into the analytical environment and prepare it for use.
• Semantic and metrics capabilities: Define shared business concepts, measures and relationships.
• BI and visualization: Support dashboards, recurring reporting and interactive exploration.
• Notebooks and code environments: Support statistical analysis, machine learning and more customized analytical workflows.
• Conversational interfaces: Let users ask analytical questions in natural language and continue an investigation through follow-up questions.

Ideally, these capabilities will be located within the AI data platform. When the platform can support multiple analytics workloads against the same governed data, teams can reduce the copies, pipelines and business logic they would otherwise have to recreate as data moves between systems.

A platform needs enough flexibility to support compute, latency and concurrency needs while preserving shared data, definitions and governance across them.

Building an analytics capability is less about following a fixed maturity model than about making a few foundational decisions in the right order. Teams need to know which questions they’re supporting, how the underlying metrics are defined, what freshness and processing those decisions require, and how people will access the results.

• Start with the decision and the measures behind it: Identify the questions people need to answer, the metrics those questions depend on and how quickly the underlying information needs to reflect changes in the business. A recurring performance report, a demand forecast and an operational decision each impose different requirements on the analytical environment.
• Define important metrics early: If revenue, churn or active customer means something different from one team or analytical workflow to another, adding dashboards, models or conversational interfaces multiplies the reconciliation work later. Clear ownership also gives teams a way to manage changes to those definitions over time.
• Match the data and processing requirements to the workload: Scheduled updates may be sufficient for recurring BI, while operational analytics can require continuously processed data and tighter latency. Predictive workloads introduce additional requirements around historical coverage, feature quality and whether past data still represents current conditions.
• Expand analytical access according to how people work: BI and governed self-service support recurring exploration for business users; notebooks and code environments serve more specialized analytical work; conversational interfaces give users another way to express questions in business language. These access patterns work best when they draw from shared governed data and business context rather than recreating them independently.
• Build data literacy and governance alongside access: Users need enough context to understand what a metric represents, which population it describes and where uncertainty enters the result. Governance provides the corresponding controls around access, ownership, definitions and traceability.

Organizations will assemble these capabilities in different orders. Keeping the data and business context shared across them reduces the amount of infrastructure and analytical logic that has to be rebuilt as new workloads are added.

Snowflake supports analytical workloads on a shared governed foundation, so teams can work across BI, real-time analytics, machine learning and AI-assisted querying without maintaining a separate data environment for each use case.

For BI and interactive analysis, Snowflake provides elastic compute and concurrency for many users working against the same underlying data. Real-time and operational analytics can work with continuously updated data, while data science and machine learning workflows use the same governed foundation for feature preparation, experimentation and model development.

Semantic capabilities add shared business context across those workloads. By defining metrics, dimensions and relationships centrally, organizations can keep dashboards, applications and AI-assisted interfaces aligned on the same business definitions rather than recreating that logic independently.

Snowflake Cortex AI extends that foundation into conversational and generative analytics, while Snowflake Horizon applies governance, security, discovery and compliance controls across data, applications and AI. The result is an analytics environment in which different workloads can use the processing and interfaces they require while remaining connected to shared data and governance.

Business users will increasingly expect to move between recurring metrics, ad hoc investigation, forecasts and conversational queries without having to understand which pipeline, model or tool sits underneath each result. This raises the standard for the data foundation: definitions have to travel with the data, governance has to follow the workload and freshness has to match the decision.