One AI has been a component of One Model since nearly the beginning of the company’s existence. It is an integral part of the One Model platform and, as such, the team that develops One AI features aligns with the core principles of One Model. That said, there are also some more nuanced priorities that the team has remained focused on over the years. These additional core principles relate specifically to the fields of machine learning and data science.
When we set out to develop a new version of our forecasting capabilities, the team kept sight of the principles that have guided the creation of the product over the years. The new forecasting capabilities combine ease of use with powerful configurability and are available right where you want them: in Storyboards.
If you’re a One Model customer or considering being one, you may be interested in knowing what drives the shape of the product. Let’s take an inside look at important concepts to developing One AI functionality achieved by tying the forecasting features to the following guiding principles.
In the beginning, One AI consisted of a single data scientist tackling projects in notebooks. Showing the work involved in an analysis was just part of the process. The usefulness of doing so was apparent, and ethics and transparency became core values integral to the development of One AI over the years.
One of the recent enhancements to forecasting provides greater transparency into which forecasting method was leveraged and how it was configured. We added an information icon to clearly indicate that additional information is available. Performance scores are provided if situational intelligence was used to choose a forecasting method.
As One AI has matured, we've placed a lot of emphasis on ease of use. Simply put, you don’t need to have a background in statistics or data science to get value from One AI. Embedded Insights (of which forecasting is an example) and One AI Recipes are examples of tools that improve the accessibility of One AI.
Running a forecast has always been easy. The appearance of the lightbulb icon makes it clear when a forecast is available. Now, the "easy button" is more powerful. One AI can automatically select between a curve fit model and an autoregressive integrated moving average (ARIMA) model based on the data contained in the trend, using AIC model scores that balance model complexity and fit. In simpler terms, One AI is increasingly leveraging situational intelligence to determine the best configuration without the user having to manually configure settings.
The skills of the future are becoming the skills of the present. Python, R, and other data science-related skills are on the rise in a significant way. More and more One Model users possess these skills. The team developing One AI has always strived to create tools configurable enough to benefit data scientists. Having machine learning and statistical capability available in the same platform as the people data the analysis is being performed on is a driver of efficiency and data security.
One Model users now have the option to manipulate both general and model-specific settings related to forecasting. General settings include the ability to select the number of periods to predict, confidence interval, and bounds. If you don’t want One AI to select a forecasting model automatically, you can manually choose from Curve Fit, ARIMA, and Prophet. Taking things a step further, you can allow One AI to automatically select the best settings for that model based on the trend data, or you can configure custom settings pertaining specifically to each forecasting model.
Making complex things accessible to more people creates opportunities for learning to occur. The team at One Model has been on a One AI documentation bender and has been busy creating resources to help our users. We’re also increasingly integrating additional information and context into the places where the user can benefit most from it. This is delivered through inline text, tooltips, and information popups.
The main risk in forecasting is in interpreting the results. When we designed the forecasting enhancements, we introduced an information icon to tiles where forecasting is available. Tooltips were also employed to define each of the available settings.
In the world of data science and AI, things move fast. Generative AI is the most immediate and vivid example of this. It's an industry where you must innovate or be left behind. At One Model, we recognize this. We don’t jump on every new tech bandwagon, but we keep a finger on the pulse of innovation in the space.
When it comes to forecasting, One Model offers several techniques. Two methods, curve fit and ARIMA, are tried and tested and have been around for a relatively long time. They’re sometimes referred to as “classical techniques”. The third, Prophet from Meta, is a more “modern” forecasting library. In the recent forecasting enhancements, we made our Prophet option more powerful by exposing many configuration options.
Hopefully, this article has provided some visibility into the thinking behind the product decisions the team at One Model makes when developing One AI tools. We’re fortunate to be layering these tools on top of One Model’s industry-leading data integration, modeling, and visualization capabilities. We strive to deliver the best AI on the best people analytics platform.