Machine Learning and Analytics: An Expert Guide

Michael Chen  Senior Writer  October 22, 2024

Machine learning and analytics have become indispensable tools for businesses seeking to extract valuable insights from their data. By using powerful algorithms and statistical models, organizations can uncover hidden patterns, make more data-driven decisions, and gain a competitive edge in today's rapidly evolving marketplace.

While teams can analyze data without machine learning, the results may fall short of expectations. The fact is, ML significantly boosts the capabilities of analytics platforms.

What Is Machine Learning?

Machine learning is a subset of artificial intelligence that uses algorithms trained on large data sets to recognize trends, identify patterns and relationships, and then use that information to make predictions or inform decisions without being explicitly programmed and with minimal human intervention.

Machine learning technology has applications in many industries, including healthcare, finance, marketing, and cybersecurity. Results improve through an iterative learning process that focuses on increasing accuracy, adding customization, and reducing errors in the model.

What Is Analytics?

Analytics is the process of deriving insights from data and using them to draw conclusions or make decisions. It involves collecting, cleaning, and organizing data to identify trends, correlations, and patterns. By using various statistical and mathematical techniques, analytics helps organizations make better-informed decisions, improve performance, and optimize operations.

Analytics is related to the field of statistics, which provides the underlying concepts that help companies understand their data and use it to drive growth and success. In business, the term analytics often refers to using software to sort through data, find unique relationships, and present findings in an accessible way through visualizations.

Key Takeaways

• Machine learning and analytics are symbiotic technologies.
• Machine learning can speed and broaden the capabilities of analytics, including by identifying patterns and insights often missed by other means.
• Analytics generates organizational value by processing data within an appropriate context for actionable insights.
• For both machine learning and analytics projects, consider continual monitoring to check for hidden biases and inaccuracies.

Machine Learning and Analytics Explained

Analytics efforts benefit significantly from the application of machine learning and other AI techniques. Analytics tools that don’t rely on machine learning use static algorithms that may miss obscure but important patterns in data. Machine learning can find those patterns, and, if need be, examine data sets larger and more varied than legacy analytics tools can handle.

Does Analytics Include Machine Learning?

Analytics does not necessarily require machine learning. For years, businesses used tools founded in statistical analysis to analyze trends in data, predict future outcomes, and assess the effectiveness of strategies. Without the benefit of ML, they sought to answer questions like, How well did our holiday discounting strategy work? What products or services are most popular with this customer segment? Which are the most profitable? While traditional methods could produce answers, without ML the process is limited in scope and the number of data points available.

To find answers, online analytical processing, or OLAP, has been used for decades to grab a segment of transactional data and analyze it using classical statistical analysis. When data is structured, as it is in a relational database, OLAP is highly effective. However, when data is both structured and unstructured and includes nonnumeric information about the business, statistical analysis can’t provide the same level of insight. Among other benefits, ML lets analysts identify more complex nonlinear patterns, even in unstructured sources of data.

As organizations put more unstructured data into their data warehouses, ML will be increasingly important for analyzing it all.

Why Are Machine Learning and Analytics Important for Business?

Together, machine learning and analytics extract valuable insights and predictions from a wide array of data. That can deliver a competitive edge for businesses because today, data comes from everywhere, and in some cases, all the time: Internal operational metrics, supplier and vendor inventories, marketing campaign results, data from customer apps, related data from public sources, financial data, data generated by Internet of Things devices—the modern technology ecosystem generates data from nearly every interaction and feeds it into a data warehouse or cloud-based repository such as a data lake.

That’s a lot of information, and it presents plenty of opportunities for businesses to find insights on operations, marketing, supply chain, and much more—but only if they can analyze large volumes of diverse data. Enter machine learning. With machine learning, the entire process of business analytics becomes more manageable and broader in scope for reasons including the following:

• Automation via machine learning can make data transformation processes, such as data cleansing and recognizing data quality issues, more efficient.
• Machine learning within analytics tools can generate “aha moment” insights based on simple queries from business users.
• Machine learning-based analytics tools can also identify hidden patterns in complex data, sparking new ideas and discussions that may create new opportunities.

Adding to the excitement now around ML-powered analytics is the scalability and flexibility offered by cloud-based data warehouses and analytics tools. Huge amounts of data and complex machine learning algorithms demand lots of computing power for efficient analysis. And because this is a fast-evolving space, developers and data scientists looking to build and deploy new models benefit from online tools and services specifically designed for machine learning and analytics. The cloud allows organizations to use the latest data analysis innovations while providing easy access to anyone in the organization with proper credentials to use the system.

Using Machine Learning in Business Analytics

Once an organization collects inputs from various sources into a repository, machine learning systems can start processing heavy volumes of data in support of strategic initiatives. These initiatives can be part of operations, marketing, logistics, and even public engagement on social media.

Here are some popular uses for machine learning in business analytics.

• Customer Segmentation: Machine learning is helpful on both sides of the customer segmentation equation. To determine which buyer profiles belong in which customer segments, machine learning can parse through purchase histories and engagement data to generate categorizations. On the other side, machine learning can quickly determine the efficacy of campaigns in specific segments, leaving marketing teams some breathing room to tweak messaging or other campaign factors.
• Fraud Detection: Machine learning can identify potentially fraudulent patterns by considering geography, purchase frequency, purchase types, amount spent, and other details of individual transactions and comparing them to customer profiles. Using anomaly detection capabilities, the system can quickly flag out-of-character activity and send potentially illegitimate transactions for further investigation.
• Supply Chain Management: Supply chains may involve a variety of partners, wholesalers, and logistics providers from around the globe. When they disrupt the flow of needed goods, local events can quickly become the concern of manufacturers and retailers thousands of miles away. Machine learning can collect and sort through data from suppliers and logistics firms to identify potential and occurring disruptions. In addition, ML systems correlate data with manufacturing schedules to determine temporary issues as well as spot trends that can lead to cost and process optimization, such as identifying vendors that are prone to part failure or late deliveries.
• Sentiment Analysis: Sentiment analysis takes text from messages, transcripts, and reviews; determines the overall tone; and then further analyzes the data for marketing and sales insights. Machine learning is necessary to process heavy volumes of textual data from diverse sources quickly enough to adjust if, say, a product is frequently missing a key part or a service rep is problematic.
• Predictive Analysis: Predictive analytics unaided by machine learning has been a staple of business analysis for as long as ledgers have been kept. Simple plots of previous year sales compared with current year sales are the starting point, and statisticians have advanced the science of predicting the future from the past tremendously. Machine learning builds on that heritage by more accurately processing more data and using more complex methodologies. ML also aids in analyzing what-if scenarios that help guide business leader thinking.
• Price Optimization: At what price is profit maximized? Too expensive, and not enough people will buy. Too inexpensive and margins suffer. Beyond spending habits, factors such as competitor prices, seasonality, weather, and inventory scarcity form a complex and dynamic price algorithm. Machine learning and data analytics can sort through all this data to create optimum pricing scenarios.

Understanding Analytics

It’s always useful to review the actions you’ve taken to determine if you achieved the best possible outcome. Reflecting on past performance usually leads to improvements the next time around. Analytics should always have these sorts of goals—what can you achieve by finding actionable insights in data?

Statistical analysis of numeric data is a worthy starting point. But that potentially leaves a lot of data unanalyzed or at the very least, produces slow results while opening the door to human error. ML can help broaden analysis to find insights that aren’t easily discerned otherwise.

Types of Analytics

Companies have a wide range of analytics types and techniques to choose from, and the best fit for a project often depends on what the team wants to get out of its data. The following are four categories of analytics.

• Descriptive analytics. Descriptive analytics systems take historical data and determine patterns and metrics to derive insights needed to create a situational analysis. For example, a finance model could intake data from sales, marketing, HR, and expenses to create a quarterly analysis for an organization. Dashboards will typically be the way to visualize descriptive analytics.
• Diagnostic analytics. Diagnostic analytics systems take historical data to find the root cause of a situation, trend, or relationship. For example, if an organization is seeing a spike in complaints about a specific product’s quality, it can employ a diagnostic analytical tool that considers data starting with the supply chain on through to product delivery to determine if the root cause lies with a particular material, manufacturing step, or other cause.
• Predictive analytics. Predictive analytics systems create a forecast of future performance based on relevant current and past data. The prediction can relate to anything—weather models, optimal stock levels, customer behavior in a marketing campaign. The more data, the better to create a situational profile that allows for predictive insights.
• Prescriptive analytics. Prescriptive analytics is similar to predictive analytics, but goes further by suggesting fixes to issues found. For example, a predictive analytics system might forecast plateaued sales for the next quarter. Prescriptive analytics can combine historical data and market analysis to produce prescriptive actions for overcoming weaker sales projections.

Steps in the Analytics Process

In general, the process requires collecting and cleaning data, choosing a technique, interpreting results, and communicating insights to stakeholders. Collaboration between data analysts, domain experts, and decision-makers can be helpful to ensure that insights generated are relevant and impactful.

1. Identify the issue. All analytics should address a business issue. Are you trying to analyze marketing data? Find out what’s driving employee turnover? Discover the weak link in your supply chain? Identifying the issue creates a starting point for analytics projects.
2. Collect and clean the data. Now that project goals are established, identify the data sources needed by the analytics platform. Options include using an iPaaS system that links data sources or connecting to a repository such as a data lake or data warehouse. To ensure compatibility and accuracy, data also needs proper formatting for processing. Cleansing usually involves removing duplicate entries and denormalizing data before analysis. For repeatable data sources, machine learning can help automate part of the cleaning and transformation process to improve efficiency.
3. Explore and visualize data. Using analytics tools, you can create data visualizations and generate initial insights. This process creates general findings that establish the parameters of any data-driven hypotheses that will serve as the foundation of data models, including which data sets provide the most value.
4. Model the data. With a basic understanding of the goal and available data sources, data engineers build models to structure and organize the data, bridging the gap between raw data and data ready for storage and retrieval by analytics applications.
5. Evaluate the model. Here’s where you analyze. With the data model ready, teams can begin the analysis process to achieve the project’s initial goals. Data analysis can involve different forms of statistical analysis, including the use of programming languages and analytics tools.
6. Deploy and monitor. Now it’s time to take action. With reports and visualizations ready, users can present findings to stakeholders to begin discussions on critical decisions. With analytics, recommendations stem from evidence found in data and presented clearly with visualizations—often with deeper insights than traditional or manual evaluation techniques.

Key Techniques in Analytics

The practice of analytics is built on a number of techniques established in the field of statistics, then brought to scale through the capabilities of machine learning. Some of the most common techniques used in analytics are as follows:

• Regression analysis. Regression analysis is one of the primary techniques in data and statistical modeling. With regression analysis, the machine learning model analyzes data to see which variables influence an outcome and by how much. Regression analysis encompasses a family of techniques, including linear regression, nonlinear regression, and logistic regression.
  • Clustering. Clustering is a type of analysis used with unsupervised machine learning models. With clustering, a machine learning model explores a data set to find smaller groups of related data, then derives connections and patterns from those smaller groups to generate greater understanding.
  • Time series analysis. In statistics and data modeling, time series analysis looks at data points collected within a specific time range for patterns, changes, and impact of variables to create a prediction model. One of the most common examples of time series analysis is weather data across the course of a year to predict seasonal patterns.
  • Association rule mining. Some of the most profound data insights can come from identifying patterns and finding interesting relationships within large data sets—one of the principle ideas of graph analytics. Association rule mining is a type of machine learning that finds hidden connections and commonalities in variable relationships. For example, a fast-food chain might use association rule mining to find items commonly ordered together, then offer those as discounted bundles to drive customers.
  • Text mining. Text mining is a form of unsupervised machine learning that takes incoming text from sources such as emails, website comments, or social media posts, then uses natural language processing to derive meaningful patterns. These patterns can then be associated with other variables, such as engagement metrics or sales data, to drive understanding of intent and sentiment.

Understanding Machine Learning

At its core, machine learning is about finding connections and patterns within data. ML does this using techniques as straightforward as decision trees and as complex as neural networks, with their deeper layers capable of providing nonlinear relations in data. However, no matter the method, machine learning helps organizations improve cumbersome processes and delve into their data to fuel greater productivity and better decision-making.

Types of Machine Learning

A wide range of machine learning models exist depending on a project’s resources, goals, and limitations. Understanding the different types of machine learning techniques allows for teams to make the right choice for their project. Common types of machine learning include the following:

• Supervised. In supervised learning, ML algorithms train from labeled data sets with the goal of identifying known patterns to iteratively refine the accuracy of outputs. This process is characterized as supervised because the number of known parameters involved allows for clear measurement of model improvement.
• Unsupervised. Unsupervised learning lets machine learning models process unlabeled data sets without goals or metrics in mind. Instead, an unsupervised approach provides a sandbox for organic learning through pattern detection, relationship detection, or other forms of generated insights. When successful, models trained through unsupervised learning will be able to properly mimic the environment presented by the data set and thus form accurate predictions.
• Semi-supervised. Semi-supervised learning combines supervised and unsupervised techniques to accelerate the machine learning process. With semi-supervised learning, a model gets a head start by using a small amount of labeled data. After finishing with that data set, the model then begins exploring a larger unlabeled data set to apply the basics learned in the first step before refining its predictions in an organic, unsupervised way.
• Reinforcement learning. Reinforcement learning refers to the process of letting a model explore a data set with the purpose of achieving a specific outcome. Each decision along the way generates feedback in terms of either positive or negative reinforcement, which then informs the model as it further revises to anticipate an appropriate response to situations.

Steps in the Machine Learning Process

Regardless of your goals and parameters for your machine learning model, these projects often follow a standard process. Understanding this process before starting a project provides a roadmap for resource allocation and budgeting along the entire machine learning lifecycle.

Here are common steps for developing machine learning models.

1. Identify the issue. What is the purpose of your machine learning model? More importantly, have others already produced models for that task, and if so, is one sufficient for your goals? Every project needs to be able to solve a problem, and the quality of that solution should define the project parameters from the starting point to metrics that dictate success.
2. Collect and clean the data. To drive any machine learning project forward, you need data. That means identifying sources of training data similar to the data the trained model will encounter in general use then collecting and transforming that data into a unified, compatible format free from duplicates and errors. Skimping on this step could create biases that skew or even derail a project. Taking the time to carefully manage a project’s data set is an investment in ensuring success.
3. Engineer for features. Not everything in a data set is necessary to train a machine learning model. A crucial early step for machine learning is identifying important parameters for the project, then curating data sets that feature diversity around those parameters. Feature engineering requires expert-led iteration, ultimately driving transformations by adding, removing, or combining data for a greater context that improves model accuracy.
4. Select and train the model. Your project goals will determine a short list of machine learning techniques. Practical limitations, such as compute resources, project timeline, availability of quality data sources, and the experience of team members can narrow choices and ultimately dictate the best fit for a project. Once selected, the model iteratively trains on a curated training data set, refining outcomes and results until it achieves consistent accuracy.
5. Evaluate the model. A successfully trained model delivers repeatable, explainable, and accurate results. Evaluate your trained model using real-world data to gauge how well it performs outside of its training data set. Evaluation tells teams how close the project is to meeting its original goals.
6. Deploy and monitor. If a model successfully handles real-world test data on a consistent basis, it’s ready for a production environment. While deployment should happen only after certain benchmarks are met, that doesn’t mark the end of the model’s evolution. Teams must continuously monitor a model’s results to make sure it maintains accuracy, consistency, and other desired outcomes—and if results deviate, discover why.

Key Techniques in Machine Learning

Many machine learning techniques are in use, yet not every technique necessarily applies to the goals or limitations of a project. The trick to successful machine learning is knowing which technique to select based on your individual project parameters.

Popular techniques used in machine learning include the following:

• Decision trees: Decision trees use supervised learning to understand the various options to consider as items move through a workflow. For instance, when a new invoice comes in, certain decisions must be made before the invoice is paid. Decision trees can aid regression analysis and clustering to determine, for example, whether a bill is a valid, with a complete invoice versus possibly fraudulent or missing the required data for payment.
  • Random forests. A single decision tree provides only a limited view of a situation. Random forests refer to the technique of combining multiple decision trees—hence, a forest—to create a cumulative outcome with a broader perspective. Random forests overcome many of the limitations of decision trees and offer greater flexibility in both function and scope. In fraud detection, for example, the decision about whether a transaction is legitimate or not can depend on many factors, such as where the transaction originated, whether the item mix is typical for a customer, and whether the size of the purchase is unusual. Decision trees within a forest can handle each evaluation parameter.
  • Support vector machines. Sometimes data naturally falls into clusters, whether they’re obvious or not. Support vector machines (SVMs) are a type of supervised learning that seeks to find ways to maximize the difference or distance between two clusters of data. Sometimes there’s an obvious linear dividing line between data groupings, sometimes the dividing function is nonlinear. If there’s no obvious clustering in two-dimensional views, SVMs can use higher-dimension analysis to find ways to cluster data.
  • Neural networks: Neural networks arrange compute nodes in a manner similar to the neuron networks in our brains. Each layer within a neural network applies unique functions to determine how input data should be classified and whether predictions can be made from the input data.
  • Gradient boosting: Every machine learning model prediction comes with a level of confidence. For example, say a transaction looks like fraud with 0.8 confidence, where 1.0 is perfectly certain. That’s a pretty confident prediction. When a model makes its assessment, some calculations along the way will contribute significantly to the prediction, while some will not contribute much at all. In many models, low contributors are discounted as they appear by themselves as noise. Gradient boosting seeks to combine some of these low contributors in a way that lets them more significantly contribute to the prediction, thus lowering error rates and boosting confidence ratings.

Challenges in Analytics and Machine Learning

Machine learning and analytics rely on many of the same techniques. Because of that, both efforts face similar challenges, whether taken separately or as a combined “analytics-powered-by-machine-learning” project. Following are some common challenges faced by project teams.

• Data Quality: Machine learning requires lots of data. But when that data is rife with inconsistent formatting, duplicates, and other issues, it can skew the model training process. Data quality is one of the primary challenges in creating an effective model, but note that when it comes to ML, “quality” means the data is properly formatted and reflective of what the model will see in real scenarios. If training data is too clean and doesn’t represent the real-world variability the model will experience in production, it may overfit to the training data—that is, be unable to handle the variability and complexity present in the real data sets. Organizations should employ strategies to maintain data quality, including vetting data sources to proper transformation techniques and regular deduplication. But they need to strike a balance, cleaning the data enough to remove noise and errors while still retaining variety.
• Algorithm Selection and Optimization: Every project comes with specific needs, and depending on the project’s goals, different techniques and algorithms will make for the best fit. Sometimes those choices seem obvious, such as if you know the structured nature of decision trees will work for the scope and nature of the problem at hand. In other cases, model selection is less clear cut. Document your data’s characteristics, such as size, type, and complexity, then consider the problem you’re looking to solve. How much processing power is required to train and use the model, and can it scale to handle your data? It’s best to start simpler and move up in complexity. Tools such as AutoML can help automate testing and selection of the best algorithm for your project.
• Overfitting and Underfitting: If training data doesn’t provide the model with an appropriate balance of breadth and quality, either overfitting or underfitting can occur. Overfitting happens when training data contains only certain genres of data. If you want an app that can identify song titles and singers but you give it only country music during training, it will be lost when it comes to rock or R&B. Underfitting refers to the opposite of that—the model hasn’t been trained extensively enough and fails even on what should be obvious queries or pristine inputs.
• Interpretability and Explainability: Interpretability and explainability are similar but distinct properties of an AI model. When the output of an AI model is explainable, we understand what it’s telling us and, at a high level, where the answer came from. If generative AI writes a four-paragraph summary of a board meeting, you can read the minutes of the meeting and understand how the system chose to write what it did. Or if a model predicts that a product will increase in sales by 3% this year, you can look at the sales reports and understand where that number came from. That’s explainability.

Interpretability means understanding what the model did to come up with the particulars of its answer. Why did the GenAI system choose the words it chose in the order it chose them in summing up that board meeting? What calculations did it use to come with that 3% sales increase forecast? When AI cites its sources, it improves explainability. But as models become more complex, they are less and less interpretable.

Analytics and Machine Learning Best Practices

Analytics and machine learning share common practices regarding such factors as data sources, algorithms, and evaluation metrics. The following cover common practices for both analytics and machine learning.

1. Define the Problem and the Metrics of Success:
   What is the purpose of your analytics project? That simple question is the foundation for everything that happens afterward. Know what problem you’re trying to solve, and decisions such as algorithm and data source selection cascade from there. That sets the starting point, but the finish line also needs definition. How will you measure success? Those two questions provide the broad framework for a project, and from there, teams can start filling in the details.

2. Use High-Quality, Diverse Data Sets:
   The results of a project are only as good as the source data. Low-quality data sets with issues such as duplication and unrealistically uniform sources create problems—at best, skewing results, at worst, leading to wrong conclusions that cost the company time, money, and customers. For both analytics and AI, data sets must be current and reflect real-world conditions while bringing a range of relevant yet diverse perspectives.

3. Choose the Right Algorithms and Model Architecture:
   Machine learning techniques have been developed for specific purposes. Anomaly detection systems are different from hierarchical clustering or object identification systems. Some ML methods require more processing power and may be poor choices for simpler applications. Similarly, analytics models have their best uses, too. It may be well worth trying out a few different algorithms on your data and comparing their performance.

4. Regularize and Optimize Models:
   In ML, overfitting occurs when the model’s training set lacks the diversity that will be present in production use. If a model is highly trained on a limited data set, it may not be able to interpret input that’s different from its training set. Regularization seeks to eliminate overfitting and make the model more generally applicable. Optimization iteratively fine-tunes a model to ensure high accuracy.

5. Communicate Results Clearly:
   The practices listed above involve technical elements of projects. However, one of the biggest potential keys to success is often overlooked: communicating results. Teams may be focused on fine-tuning models or auditing data sources and forget that key stakeholders need to know how a project is progressing. That requires clear communication with actionable metrics and a concise evaluation of “How’s it going?”

Use Cases and Applications of Analytics and Machine Learning

How do analytics and machine learning apply in the real world? As long as data exists, organizations in any industry can integrate analytics and machine learning. In fact, different departments, such as engineering, operations, marketing, and sales, can use these in different ways. The following cover just a handful of use cases showcasing the benefits of analytics and machine learning (PDF) across a variety of industries and functions.

• Marketing: Marketing departments get data from all sorts of avenues: engagement tracking on emails and social media posts, purchase histories, app usage, browsing behavior, and more. What to do with that flood of information? Machine learning systems can compile it to look for specific patterns and build an analytics-driven profile of individual customers and segments for business users. From there, data-driven decisions can activate further strategies such as microtargeted offers or seasonal engagement by demographics.
• Finance: When data from across an organization is consolidated, finance departments can use machine learning to compile those massive volumes for deciphering with analytics. The resulting data-driven insights can provide a closer look at critical factors, such as cash flow, payroll trends, and asset purchase patterns. Analytics can derive new levels of insights through trend detection and model-based predictions while also providing assistance with fraud detection.
• Healthcare: Between electronic medical records, connected devices, and operational metrics from facilities, machine learning and analytics can work together to help healthcare organizations optimize operations and offer better individual care. For operations, staffing can scale up and down based on proven usage cycles triggered by factors such as season and weather. For individuals, data-driven insights can provide flags as to when to book certain screenings or promising new treatments.
• Robotics: Nearly every use of robotics generates data, from the manufacturing cycle to the final product in production use. For the latter, data can come from sources including temperature sensors, CPU use, and mechanical joints and motors. Analytics can take that massive amount of data and aim to optimize all facets of production, such as manufacturing sourcing and motor upkeep, ultimately lowering maintenance costs.
• Economics: Machine learning can benefit economic research and analysis in many ways. At its simplest, it can crunch massive amounts of data and build visualizations. However, economic analysts also employ machine learning to research related data points, such as text-based sentiment, providing a greater context to the how and why of a particular finding.

Machine Learning and Analytics FAQs

What is the difference between ML and analytics?

Machine learning is the process of evaluating large data sets to identify patterns and build a predictive model, whether for small automation tasks or for larger, more complex processes that require critical thinking. Analytics refers to the science of systemic analysis of data and statistics. Analytics can benefit by integrating machine learning to generate data models, but the two concepts exist separately unless purposefully used together. In today’s business landscape, the combination of ML and analytics can position an organization for success.

What are the types of analytics with machine learning?

In general, any type of analytics can use machine learning as long as the analytics platform supports it and is properly connected to a data repository. Functionally, nearly any analytics project can benefit from using machine learning to expedite the data-crunching process.

How can machine learning and analytics be used to make business predictions?

Analytics can help organizations make business predictions by processing historical data, identifying patterns for things such as sales cycles, market trends, customer behavior, or even manufacturing processes. With predictive insights into any of these, organizations can make decisions to best take advantage of findings for better business outcomes.

How can organizations ensure that their machine learning and analytics projects are successful?

For machine learning and analytics projects, consider the following practices, which may help to position them for success, including the following:

• For both: Using high-quality data sources.
• For analytics: Having data engineers ensure that modeling and data meet standards before use.
• For analytics: Selecting techniques that best balance project goals and practical resources.
• For machine learning: Troubleshooting for issues such as overfitting and underfitting.
• For machine learning: Continuously monitoring a model after deployment to see if further revisions and adjustments are necessary.