Universal Models: Forecast and More without Building a Model from Scratch

A + B = C 

This simple equation can form the foundation for a pretty basic mathematical model. A and B represent the inputs on which you apply a mathematical function to in order to return an output. If only all modeling were so neat and easy.

Custom models, the ones needed to deliver personalized insights about your data, are built on equations that are much more complex. Building them often requires you to know advanced statistics and have coding skills—not to mention the time to train your models. 

The good news is that Domo’s platform has everything you need to create the perfect customized model. The even better news is that, in the coming months, Domo’s AI Service Layer is introducing new Universal Models to more easily allow you to perform functions like forecasting, personal data protection, and sentiment analysis without the hassle of building your own model.  

The AI Labs team shared more about Universal Models in part 4 of our AI Insights Livestream series: Universal Models: Forecasting, PII, and Beyond. You can catch up on the recording, or keep reading to learn more.  

What is a Universal Model? 

Universal Models take a one-size-fits-many approach to modeling. Using a generic equation—although it’s not necessarily a simple one—the model can work with any data input that meets its requirements.  

If you don’t have a statistics background or coding skills, Universal Models can be a great starting point for working with modeling. That’s because they are pre-built and require minimal setup and maintenance. The user-friendly models will be available to anyone with a Domo account.  

Although these new models are currently in development, you’ll be able to access the forecasting model starting in September, while the personal data protection and sentiment analysis models will be released in the following months.  

Recognizing patterns that matter with forecasting models 

First up, the Universal Forecast Model will use your historical data to help anticipate future outcomes. The model’s assumption is that the future is likely to look like the past and will tend to follow a similar pattern—assuming a pattern exists.

This isn’t a model designed to detect random events like fashion trends. Instead, you can use it to observe trends over consistent periods of time, like month-to-month orders.  

In modeling terms, the equation is: 

Trend Cycle + Seasonality + Error = Forecast 

• Trend cycle: Long-term increases or decreases in data patterns
• Seasonality: The repeating cycle in the series with fixed frequencies
• Error: Leftover fluctuation or variation in the data

To use the forecasting model, your data set needs to have plentiful data that allows for detecting patterns, and include sequential time values paired with historical outcome values, as well as being comprehensive and extensive to determine overarching trends or seasonal patterns 

It’s hard to predict, for instance, what will happen with your orders next summer if you only have data from the previous year and can’t tell if there’s a longer-term trend or pattern at play. 

If the data sets do meet all of the requirements, the Universal Forecast Model will give you a range of possible values that future outcomes could take with relatively high probability (or prediction intervals), as well as the average possible value for upcoming time intervals.  

So, a business that can forecast the number of orders they can expect in each month of the upcoming summer can more accurately plan procuring the goods and materials they might need to fulfill those orders without overspending. 

How to use the Universal Forecast Model 

Here’s what will happen when the forecasting model becomes available in September: Domo will provide a default Universal Forecast Model, and users can choose between an ARIMA and ETS model type.

However, if you have an existing forecasting model that you’re hosting elsewhere, you could also connect that to Domo and use it within the platform. There’s no need to let a trained model go to waste.  

Then, you can apply these forecasting models in Domo in a number of ways: 

• Cards in the App Studio: Directly input the model into a card to generate a forecast. When creating the card, be prepared to answer a number of questions: What is the time column? What is the outcome column? How many future periods do you want to forecast? 

• Magic ETL DataFlow: Materialize a forecast in a data set and use the model to update that data set.  

• Jupyter Workspaces: Bring the model into Jupyter to conduct more advanced analysis or processing. 

• Workflows: Use the model in a workflow to trigger specific actions. If you predict a dip in monthly sales, it will trigger an alert. 

So, start thinking now about how forecasting can level up your data analysis. And remember, to perform this type of modeling, your data set won’t need a lot of columns, but you will need a lot of rows.   

Two more Universal Models on the horizon 

We don’t need a forecasting model to tell us that two more Universal Models are coming down the pipeline. Let’s take a quick look at what you can expect in the coming months. 

Universal PII Model 

Personal Identifying Information (PII) is information that can be used to identify a specific individual, such as their name, Social Security number, credit card number, or address. Companies have to be especially careful when they collect, store, and process personal data in order to comply with legal and regulatory requirements. And, not all PII is created equal. For example, protection of Social Security numbers should take higher priority over the handling of email addresses.  

In addition to being costly and difficult to customize, many PII detection models do much better working with unstructured data like emails and documents than with tabular data. We’re solving this problem by introducing a Universal Model that will support a PII detection system within Domo instances. 

Here’s how it works: Through the model, users can transform their existing data set into a new data set that offers a quick view into the frequency with which certain PII types appear. So if you can detect and classify the PII within your data set, you can more easily take the appropriate actions to protect it.  

Universal Sentiment Analysis Model 

Knowing how people feel about your service or product is essential. But this isn’t always easy to learn, particularly because those insights are buried in troves of unstructured text data: social media comments, online reviews, customer surveys, and customer call transcripts.  

Domo’s upcoming Universal Sentiment Analysis Model can help with sorting through it all. The model is designed to review each response (whether from a comment, review, survey, or transcript) and assign it a polarity score between -1 and 1. A score of 1 means a highly favorable review, while a -1 is an especially unfavorable review.

You’ll be able to set your own thresholds to determine which scores count as positive, neutral, or negative reviews, and you can make your own adjustments to improve accuracy over time.  

One more thing this model can help with: learning what’s driving the negative or positive reviews. For example, maybe customers find your staff especially helpful, or maybe they’re turned off by your pricing. By identifying topics of interest like pricing or staff, and incorporating classification models, you’ll be able to drill down on what’s happening and better respond to customer needs. 

More models coming for Domo users 

The Domo AI Labs team is consistently finding new ways to improve user experience by making powerful analysis simple. Universal Models are easy to set up, easy to use, and easy to track. Accelerate your data-driven decision-making with ready-made models that support mission-critical functions across teams. 

And if you want to dig deeper into the Domo.AI world, you can: 

• Watch the replay of AI Insights: Universal Models—Forecasting, PII, and Beyond—part 4 of our summer 2024 AI series.   

• Check out our AI Readiness Guide, posted on our Community Forums. It’s a checklist for priming your data sets to be ready for any AI use case.