Ph.D. in Economics, Northeastern University
B.A. in Economics, University of California at Los Angeles
Econ One, January 2018 – Present
UCLA Department of Sociology, 2015 – Present
Independent Economic Consultant, 2015 – 2018
EY (Formerly Ernst & Young), Transfer Pricing, 2013 – 2014, Advisory Services, 2014 – 2015
UCLA for Int’l Science, Technology, and Cultural Policy, 2005 – 2017
Northeastern University, 2007 – 2009
Executives and business leaders recognize the power of data analysis to transform their business. But the popularity of data has led to an alphabet soup of terminology. What is AI? How is it different from “generative” AI? Where do LLMs fit in, and how does that differ from machine learning (ML)?
I’ve written this primer to help company leaders understand the terms and techniques that comprise modern data science and indicate how these tools can be leveraged for business success.
What ties all of these acronyms together is the idea of a model. “Model” is such a vague term— it can refer to anything from an Excel workbook to a sophisticated artificial intelligence product. But every model involves compiling data from various sources within the company and running some type of mathematical computation to produce a specific outcome. The breadth and quality of the data determine how complex a model can be — data modeling starts with data governance — but every business, no matter how small, does some kind of data modeling.
Example: A small side-hustle creator who sells candles keeps track of monthly sales data to calculate whether her price point is high enough to cover costs.
Every business performs even basic modeling on their data, in order to understand their current situation. This is often called business intelligence (BI) or business analysis, and many companies use off-the-shelf products such as Salesforce (a common customer relationship management tool) for this purpose, generating reports that slice their sales data by region, product, or some other meaningful characteristic.
Example: Our side-hustle candle creator is now a small candle business owner with two locations. The business starts keeping track of which candles sell best in each location and online to plan future scents and jar designs.
Predictive models take the same data that is used for business intelligence, but analyze it over time to forecast future outcomes. This can be as straightforward as a historical average to predict future sales or as complex as a ten-thousand-variable machine-learning model to predict the optimal mix of services and products for a new business location. While many larger businesses can implement basic predictive models—and a few have in-house data science teams to build advanced ones—most still need some assistance from data experts to design, execute, and deploy predictive models effectively.
Example: Our small candle business is now a regional candle store getting ready to go national. The company works with a consulting firm to model future sales, demand for various scents, and which candles to stock in each store for the upcoming holiday season.
Below are some of the most common types of predictive models:
Perform tasks typically associated with human intelligence such as learning, problem-solving, and decision-making. Machine learning is a subset of AI, as are recommendation engines (Netflix), virtual assistants (Siri), and strategy (DeepMind). All these models take existing or historical data and use it to train algorithms to perform the tasks required.
Example: Our candle chain is now the largest candle retailer in the country. They use a recommendation algorithm that analyzes customer data—such as location, other websites visited, inferred gender, and prior purchases—to recommend candles that the customer is likely to enjoy.
Statistical algorithms that attempt to predict future outcomes (or unseen data). This can be considered a subcategory of Artificial Intelligence and is commonly used in recommendation engines, forecasting models, and other predictive applications.
Example: Consider a model designed to predict product sales. In terms of our candle business, the company might develop a model or machine learning algorithm to predict the number of lavender candles sold in a new location — using the number of women in the surrounding zip code as a key predictor, since the company data shows that most lavender candles currently are purchased by women.
Create “new” art or text based on prompts from the user—for example, tools like ChatGPT or image generators such as Dall-E. These tools surface patterns from massive datasets, in order to predict the most probable next word in a sentence or the most probable composition of elements in a piece of art.
Example: Our candle retailer uses a Large Language Model to generate artwork and text for social media campaigns, re-write marketing emails for customer mailing lists, and craft descriptions of new products for the website.
Large language models—like ChatGPT, Llama, Gemini—are a specialized type of generative AI specifically designed to produce human-like text and generate language in a way that feels natural to us. These models are trained on a massive corpus scraped from the internet and are designed to predict the next most probable word in a sentence—enabling them to generate coherent, context-aware text based on prompts from the user. Custom LLMs can be trained on internal documents and materials by the company in question so that the resulting outputs are more in line with the “voice”, tone, and context of that company.
Example: Our candle maker has upgraded to a custom Large Language Model using natural language processing and trained on its past marketing materials. The model now generates on-brand social media artwork and captions, rewrites marketing emails for customer mailing lists, and writes descriptions of new products for the website that stay true to the tone and language used by their marketing team.
Causal inference models go a step beyond predictive models —they don’t just provide predictive analytics such as predict future outcomes; they help businesses understand how to actively influence and change those outcomes using causal relationships. Businesses can use these models to understand how those changes will affect their business. As with predictive models, some businesses have teams of economists to do this type of modeling, particularly for financial data, but most require outside expertise to implement causal inference models.
Example: Our candle company is now a worldwide behemoth producing a wide range of home goods. They hire a team of consulting economists to help with model development and create a causal inference model to forecast demand for a new scent as it rolls out to various markets. The team is also enlisted to create models to determine how changing various elements of store layout affects sales and to assess whether customer churn could be reduced via targeted loyalty programs.
Data is a critical asset for any business, and leveraging your data effectively can be the difference between growth and stagnation. Understanding the various ways data can be used is key to making the most of what you already have. That’s where experts like Econ One’s Data Analytics team come in: we help clients navigate their data landscape, identify their specific needs, and implement the most efficient and impactful solutions making the most of their data — from foundational Business Intelligence to bespoke LLMs.
What is a business intelligence model, and why does it matter?
A business intelligence (BI) model is a framework that organizes and analyzes data to support decision-making and strategic planning. It matters because it helps businesses uncover insights, identify trends, and make data-driven decisions that improve efficiency and profitability. BI models are backwards looking – they can only describe what has already been seen and cannot reliably predict future outcomes.
How do predictive models improve business performance?
Predictive models use historical data and statistical algorithms to forecast future outcomes, helping businesses anticipate trends and customer behaviors. By enabling proactive decision-making, these models improve efficiency, reduce risks, and drive smarter strategies that enhance overall performance. Predictive models are forward looking – using the past to predict the future.
How do large language models (LLMs) function in real-world applications?
Large language models (LLMs) function by analyzing vast amounts of text data to understand and generate human-like language. In real-world applications, they power chatbots, automate content creation, enhance search engines, and support customer service by delivering fast, context-aware responses. They are, in essence, a very specialized type of predictive model.
What are the steps to deploy a machine learning model successfully?
Successfully deploying a machine learning model involves several key steps: preparing and cleaning data, training the model, validating its performance, and integrating it into a production environment. Ongoing monitoring and updates are essential to ensure the model remains accurate and effective over time.
How can a business use data models to increase their ROI?
A business can use data models to increase ROI by organizing data in ways that reveal valuable insights, optimize decision-making, and streamline processes. By leveraging clean and structured data, companies can reduce inefficiencies, target opportunities more effectively, and make smarter investments.
EconOne © 2024