DataRobot intends to bolster the range of automated machine learning models that it can generate by bringing to market a time-series analytics add-on for its automated machine learning platform, which is designed to enable push-button predictive analytics in a few steps. Additionally, DataRobot is looking to expand its addressable audience to developers – the reason the company reached for Nexosis in July.

The 451 Take

DataRobot's mission to bring similar simplicity to time-series modeling and analysis as it has to various other types of machine-learning-driven analytics used to make predictions is a sound one. Moreover, the offering should appeal to existing DataRobot accounts and new prospects alike, given that times-series analytics is a hard nut to crack. Furthermore, with Nexosis in the fold, DataRobot has the opportunity to make developer's lives easier, too. However, while DataRobot is diversifying its offering to engender further growth, it is likely to face tougher competition because automated machine learning is now becoming an integral part of an enterprise data science stack. DataRobot has an early-mover advantage, but the company will need to continue to build on it in order to ensure continued market differentiation.


DataRobot continues to execute on a strategy to deliver machine learning automation to all types of users – data scientists and nonexperts – and to power 80% of software apps in order to address the need for all types of users and offerings to achieve predictive analytics. These two pillars of the company's strategy are the raison d'être behind the company's new time-series modeling and analysis capabilities, and its pickup of developer-oriented machine learning automation startup Nexosis.

DataRobot's new times-series analysis capabilities originally hail from the company's acquisition of Nutonian for an undisclosed sum in May 2017. Nexosis is DataRobot's second acquisition since coming out of the gate in the first quarter of 2015 with a promise to bring machine learning automation to the masses.

DataRobot continues to target organizations of a variety of sizes, including large enterprises, noting that it now has customers all over the world, including Europe, Japan and Singapore. Financial services and insurance remain dominant verticals for the company, which doesn't disclose the number of paying customers it has amassed, but does note that it is finding manufacturing to be a growing opportunity in Japan and is operating in the public sector. DataRobot received an undisclosed investment from the venture arm of In-Q-Tel in March to bolster the federal side of the firm's business.


DataRobot has released time-series modeling and analysis as a paid-for add-on to the company's machine learning platform. DataRobot's platform is all about enabling various types of users to create predictive models in a few steps using homegrown and third-party machine learning algorithms, including those in R and Python languages, H2O, TensorFlow, sckit-learn, Vowpal Wabbit and XGBoost.

DataRobot uses a five-step process for push-button predictive analytics: ingest data, select target variables, build models (it typically builds 20 or more models behind the scenes to find the most appropriate), explore top models and get insight, and deploy the best model.

Having tested its new time-series modeling and analytics offering with more than 75 organizations for the past calendar year (DataRobot notes that it also has customers in production using it), the company has unleashed a product that is designed to provide the same user experience for time-series analysis as it has for other types of advanced analytics.

For those unfamiliar with this analytical technique, times-series analysis essentially involves the modeling and analysis of datasets in a sequence taken at successive points of time. It's a common and popular form of analytics used by companies for a variety of use cases, including sales planning, marketing planning and inventory forecasting.

DataRobot rightly contends that time-series analysis is hard to successfully pull off. It requires a vast amount of data and variants, including historical datasets, and the ability to select the right datasets to get the best results. It's also hard to find a predictive signal because factors affecting a business (including market dynamics) are continually changing, which requires continually revisiting the time-series model, in order to ensure it continues to be effective.

DataRobot Time Series includes advanced machine learning models for forecasting, as well as essential time-series methods like ARIMA and Facebook Prophet. Full API support helps AI engineers integrate modeling and prediction directly into business processes and applications.

DataRobot Time Series is designed to automate four areas of time-series analytics: feature engineering, target transformation, time-series modeling and back-testing. Moreover, it uses a similar visual and gesture-driven interface to the company's existing machine learning automation platform for user-friendliness.

Like DataRobot's existing platform, the company's time-series analysis add-on builds various model types behind the scenes in order to find the best one. It also includes visualizations specifically designed to graphically depict time-series data, and houses some other architectural differences to DataRobot's core platform in order to make predictions against a time-series model. It's available on-premises and as a cloud service – the same delivery models used for DataRobot's machine learning automation platform.


Nexosis Acquistion 

Nexosis focuses on automated machine learning for developers. DataRobot hadn't addressed this audience before, but is now looking to do so using Nexosis' technology smarts. Nexosis had introduced a free Community version of its machine learning automation offering for developers in order to build a following among developers. Nexosis also peddled a paid-for Enterprise release, noting that it had multiple paying customers for it.

Although DataRobot has yet to divulge full roadmap details, integration between Nexosis and DataRobot's machine learning automation platform is planned. However, DataRobot also plans to continue selling Nexosis offerings stand-alone for the foreseeable future, while providing users with a pathway to DataRobot's automated machine learning platform.

Although DataRobot's acquisition of Nexosis was technology-driven, there is also an 'acquihire' aspect to it. DataRobot has retained all 16 Nexosis employees, including the startup's cofounder and CEO, Ryan Sevey, who has become DataRobot's General Manager. Additionally, DataRobot is using Nexosis' office in Columbus, Ohio, as an R&D operation. DataRobot cites a headcount of 400 employees now.

Nexosis was founded in 2015, and has raised just shy of $7m in funding in total. Nexosis investors include Revel Partners, Techstarts Ventures and Matchstick Ventures.



DataRobot's times-series analytics will compete with similar capabilities available in R and Python languages. R and Python are popular with data scientists and developers for creating this type of analysis – even though DataRobot is looking to make it simpler and faster via automation processes.

Additionally, we think DataRobot will square up to fellow purveyors of enterprise data science platforms that also integrate with R and Python and deliver time series analytical modeling and forecasting – albeit not in quite the same manner as DataRobot. Dataiku, Anaconda and immediately spring to mind, although it is worth noting that DataRobot integrates with H2O's internally developed open source machine learning algorithms, too.

Big guns including SAS Institute in SAS Studio and Microsoft within SQL Server Analysis Services also address time-series modeling and analytics, albeit in a different manner to DataRobot.

When it comes to DataRobot's core automated machine learning platform, we continue to think Israeli startup DMWay Analytics is a DataRobot competitor because DMWay's platform also aims to automate machine-learning-based predictive analysis. AutoML, which is an open source project from the Machine Learning Lab at Freiburg University in Germany, also has a similar endgame.

Furthermore, has Driverless AI; RapidMiner peddles Auto Model, which we examine here; and Alteryx and BigML have also built capabilities of this ilk into their stacks. Additionally, Google is moving into this space with Cloud AutoML, which is currently in beta.

Indeed, we think it is fair to say that, while DataRobot pioneered the concept of automated machine learning in order to enable it to be used by all participants in a data science project, as well as in most applications, automated machine learning capabilities are becoming increasingly present in data science platforms.

Krishna Roy
Senior Analyst, Data Science & Analytics

As a Senior Analyst for the Data, AI & Analytics team, Krishna Roy is responsible for the coverage of self-service analytics, predictive analytics and performance management.

Jeremy Korn
Research Associate

Jeremy Korn is a Research Associate at 451 Research. He graduated from Brown University with a BA in Biology and East Asian Studies and received a MA in East Asian Studies from Harvard University, where he employed quantitative and qualitative methodologies to study the Chinese film industry.

Aaron Sherrill
Senior Analyst

Aaron Sherrill is a Senior Analyst for 451 Research covering emerging trends, innovation and disruption in the Managed Services and Managed Security Services sectors. Aaron has 20+ years of experience across several industries including serving in IT management for the Federal Bureau of Investigation.

Want to read more? Request a trial now.