4 Strategies for success with AI, machine learning, and analytics

Artificial Intelligence (AI) is revolutionizing how everyone approaches data analysis and decision-making. If you’re responsible for analytics or perhaps even starting to add analytics to your product or service, then AI, machine learning, and predictive modeling must be interwoven into the analytics experience you’re providing to your end users.

With the rapid growth in generative AI (GenAI), product and analytics leaders are now reexamining their roadmaps to drive more value from analytics, engage their users, and drive competitive advantage. In fact, a recent McKinsey survey found that 40 percent of respondents say their organizations will increase their investment in AI because of advances in GenAI.

However, the key to success is ensuring that your AI initiatives deliver real value to your users, no matter what form they take, from regression analysis for predictive modeling to incorporating the latest in large language models (LLMs) to provide conversational analytics.

Applied correctly, incorporating AI technologies with analytics can be transformational. Businesses can make more accurate predictions and act faster. AI techniques can enable users to quickly uncover hidden patterns, correlations, and relationships within large and complex datasets that might otherwise never be discovered by manual analysis or take too long. This enables teams to embrace new business opportunities. GenAI provides new ways for users to engage with data, makes analytics more accessible, and can provide near-instant explanations about data.

However, let’s be honest: with the proliferation of AI and all the ways to start, it can often get overwhelming. The good news is that at Sisense, we have vast experience working with teams to blend analytics and AI, so we’ve broken it down for you into four strategies. Head down to the end of this blog to get a complimentary copy of our complete guide, Unlocking end-to-end AI for analytics: From ML to GenAI.

Using predictive modeling to make analytics actionable

If you’re considering powering your analytics with AI, machine learning starts with predictive modeling. Predictive modeling is one of the most practical ways for analytics consumers to get value from data, and it’s one of the most mature and accessible branches of machine learning.

Starting with a quick win: regression analysis

One of the easiest places to start is with regression analysis. For example, if you want to provide sales forecasting analytics, you can use regression analysis to predict future sales based on historical sales data, customer behavior, market trends, and other relevant factors. Your users could then use it to set realistic sales quotas and better allocate sales resources.

Another common example is demand forecasting. In this case, regression analysis can forecast demand for a product or service based on price, marketing, competitor activities, and historical sales data. This would help your users optimize production and inventory to warehouse the right amount of product to meet demand.

Sisense makes adding regression analysis to your application simple. It automatically creates univariate and multivariate forecasts of continuous variables in a dataset and offers a point-and-click, no-code workflow for business users to apply advanced forecasting predictions using AutoML. The key to success with regression analysis is using the right model for the right dataset, so Sisense provides Auto Arima, Prophet, Random Forest, Holt-Winters, Boost Spline, Ranger (Random Forest), XGBoost, CatBoost, and Elastic Net, providing a rich palette of models to choose from.

Other forecasting strategies: Time series and decision trees

As you get more comfortable with AI, machine learning, and predictive modeling, you can consider incorporating other forecasting strategies, from time series to decision trees, that use regression, classification, and other machine learning methods.

For example, classification predictive modeling works by training a model to classify new observations into predefined classes or categories based on the characteristics of the input data. It can be used in various fields, such as healthcare, finance, and marketing.

Time series analysis focuses on analyzing and predicting data that changes over time. It involves studying patterns, trends, and seasonal variations to forecast future values, from sales trends to population health changes.

Decision trees are another predictive modeling method. A decision tree can be used to predict whether a customer will likely churn or stay with a subscription-based service. The decision tree is based on customer data, including contract type, monthly charges, internet service type, payment method, and tenure. It essentially walks through a set of branches and decision points that classify the customer based on certain criteria, ultimately reaching a decision on whether they are at churn risk or not.

At Sisense, we’re focused on making otherwise complex machine learning-based models easy to incorporate and simple for your users to consume. We’ve made it simple and flexible to incorporate a whole range of predictive modeling techniques into your analytics by using R within your Sisense formulas, or using out-of-the-box functionality with Sisense Forecast or Trend Analysis. The latter makes it easy for users to identify anomalies and trends without having to be a data scientist.

Techniques like clustering algorithms can deliver more insight

A common machine learning technique for analytics employs clustering algorithms, which can be particularly valuable for business analysts to quickly visualize patterns in data. In a nutshell, clustering algorithms enable the identification of patterns, structures, and relationships within datasets. They group similar data points based on specific criteria or other similarities.

For example, let’s say you provide a marketing application or service. You could use clustering algorithms to divide your customers into segments based on their demographics, behaviors, or purchasing patterns. By providing this segmentation in your dashboards, your users could better understand different market segments, tailor their marketing strategies, and identify new opportunities for product development or target audience expansion.

Another example is using clustering algorithms for anomaly detection, which is critical in everything from financial services to manufacturing. Clustering algorithms can detect unusual patterns or outliers in datasets that may represent anomalies. This is valuable for analytics around fraud detection, identifying manufacturing defects, or otherwise allowing business teams to take proactive measures to mitigate risks.

The most common clustering algorithm is K-means clustering, and data science teams often use R to run it. R is the industry standard tool for statistical modeling, machine learning, time series analysis, and more. The good news is that if you are building analytics with Sisense, you can easily incorporate R and, in turn, clustering algorithms into your analysis. You can write R code directly in the Sisense formula editor and send Sisense fields as parameters to the R expression.

For example, the Total Cost and Total Revenue fields from Sisense below are being used to cluster data via a K-means function. The result will cluster the data based on the K-means settings, in this case, 4 clusters.

RINT(TRUE, “m
kmeans(m,4)$cluster” ,[Total Cost],[Total Revenue])

When visualizing Total Cost and Total Revenue, the analyst can also apply the clustering results to clearly identify the four key segments in the population for further analysis.

Additionally, Sisense provides an add-on in the Sisense Marketplace that makes it simple for your users to perform K-means clustering on your data within the Sisense Web Application.

Starting with generative AI—machine learning on steroids

Generative AI is a different branch of AI and is typically based on deep learning methods. These models are trained on large datasets to learn the underlying patterns and structures in the data. Once trained, they can generate new examples with characteristics similar to the original data, from text to images and video.

Most are familiar with large language models (LLMs) through ChatGPT or various copilots. LLMs, in particular, offer another important opportunity when applied to analytics: When augmented with a robust semantic layer, LLMs can be used to power Natural Language Query (NLQ) systems, where analytics users can ask natural language questions about the dataset and receive relevant answers. The LLM can understand the context of the question and retrieve the resulting set from the underlying data warehouse. GenAI can also be used for summarization, such as providing simple narrative summaries that help explain the result set.

Using LLMs to provide conversational experiences around analytics ultimately provides another way to engage users with analytics. Increasingly, everyone expects a “copilot” in their application that they can engage with. So, if you’re providing analytics in your app (or planning to), it makes sense to ensure a natural language analytics copilot for your users to work with is firmly in your plans. That way, they can ask questions about their data and get answers and explanations.

At Sisense, we’ve made it easy for developers to add conversational analytics based on LLMs to their apps using a composable GenAI Analytics Chatbot (beta). Using the Sisense Compose SDK, developers can customize GenAI experiences and mix and match GenAI analytical building blocks using flexible React Components and APIs. It enables you to deliver your own analytics copilot, tightly woven into your product, so that it feels completely natural for your development team to incorporate and your users to engage with.

 

Leveraging explainable AI to increase analytics trust

Explainable AI is critical no matter what AI, machine learning, or generative capability you use. Your users must have confidence in the results, and your analytics delivery team must have a line of sight from AI-driven answers to the underlying data and data model. After all, AI without explainability is a recipe for loss of trust.

Explainable AI starts with a strong foundation. However, if your different AI-driven approaches work in different ways directly on your database or raw data, it can become incredibly challenging. At Sisense, whether you’re using predictive modeling or GenAI, it’s all built on a foundation of a strong centralized and managed semantic layer that includes fields, metrics, formulas, and relationships. It means all your AI, machine learning, and predictive modeling used with analytics work with the same single semantic version of the truth to ensure traceability and transparency.

However, explainable AI can also mean using AI to make it simple for users to get explanations about data. For example, the Explanations feature in Sisense enables users to select a series of data points and use machine learning to identify the most likely key drivers behind changes in data. It’s explainable AI in perhaps the most business-accessible sense. The Sisense Explanations feature scans various possible contributing factors and presents the top fields (or combinations of fields) that were the most significant drivers behind the selected series. It provides explainability to get the “why” behind the numbers.

 

Read the AI, machine learning, and analytics whitepaper

There’s never been a more exciting time to combine analytics with AI, and we’re pleased to share an entire guide that helps you start infusing your analytics with generative AI, machine learning, and predictive modeling.

You can read it right here: Unlocking end-to-end AI for analytics: From ML to GenAI.