How every analytics user can benefit from machine learning

December 9, 2021 | 9 minute read
Nick Engelhardt
Senior Director, Oracle Analytics
Blair Bozada
Senior Product Marketing Manager, Oracle Analytics
Barry Mostert
Senior Director, Artificial Intelligence and Analytics
Text Size 100%:

Imagine if machine learning (ML) were accessible to anybody in the organization who needs to make decisions based on truths delivered by numbers.  The difficulty up until now has been that many machine learning technologies require special skills or tools to do this.  Instead, modern analytics platforms should be able to deliver a variety of methods that utilize ML, seamlessly embedded, that can suit different levels of user experience.  Essentially, machine learning should be accessible to all user levels­—from clickers to coders— within a single analytics platform without relying on IT, a data science team, or multiple loosely integrated tools.  There should be built-in, configurable ML models for citizen data scientists­—those who know how to tune and train models, but don’t necessarily want to resort to specialized coding tools. Additionally, data scientists should be able to publish custom built and certified models into the platform for all users to execute against their own data sets.

This blog post will address both technical and non-technical use cases for a complete picture of what is required of a good analytics platform.  What matters most is that people can benefit from the output or predictions these technologies provide.

Machine learning isn’t new; it’s been around for decades. However, ML was largely ignored by a wider potential user base for a long time, mainly due to the computational requirements and the limitations of computing power available at the time.  The cloud era brought about an ML resurgence by delivering relatively cheap and powerful computing power with little to no initial cost outlay to get started and no ongoing administration for on-premises hardware.  This, and the introduction of better end user desktop ML tools, has placed ML technologies directly into the hands of anyone that wants to use them.  Consequently, this also means that many similar, smart technologies get incorrectly lumped under the buzzworthy “ML” label, such as rules-based computation, statistical functions, and some simple predictive algorithms.  To truly be machine learning, an algorithm must be purposely trained using a specific input or training dataset and iteratively self-learn from new data without human reprogramming to refine and increase the accuracy of its predictions or output.  But there are some cool technologies, particularly for no-code users, that shouldn’t be ignored.

Some ML capabilities are better suited to different types of users, but it doesn’t mean that what business users find useful is uninteresting for data scientists. 

The diagram shows the relationship of business value to the complexity of the machine learning task. There are capabilities designed for business users which can also be leveraged by citizen data scientists and data scientists.  For example, one-click forecasting.  This capability is accessible by any user without ML skills, but a data scientist could also benefit from it to simply identify unseen trends and then build a more comprehensive custom model if there is a problem detected. 

Machine learning helps business evolve beyond being data-driven

Making data-driven decisions where stakeholders make their business decisions based on the data they are presented with, as opposed to a gut-feel or experience led decision, is the accepted best practice.  However, being data-driven could be as simple as making a decision based on a static printed report with historic numbers.  With ML technologies now in the hands of everyone, stakeholders can step up to the next level of being analytics-driven.  Analytics-driven goes beyond data-driven in that it requires the use of ML technologies to aid in the decision process.  It requires not only looking back and making decisions on historic information, but also leveraging future machine generated recommendations and predictions.

Today anybody can benefit from machine learning technology

When we talk about “business users”, we typically mean employees whose core job responsibilities usually don’t entail coding, software development, or data engineering. These are our “clickers” - they can be members of sales, marketing, finance, HR teams… you name it. Their day-to-day focus isn’t on where data lives or running specific queries, but they often rely on data for reporting, forecasting, performance reviews, goal setting, and more.

The Oracle Analytics platform provides business users access to a variety of secure, governed data sets through data flows - a code-free capability to transform data into the information needed for analytics.  Data flows allow any user to use pre-programmed ML algorithms (that have already been trained and tested by their organization’s data scientists) to draw conclusions from their data. Users can quickly and easily execute on pre-published models – no coding experience or ability required – to make analytics-driven decisions faster, leading to higher revenue numbers, more effective marketing campaigns, a more efficient recruiting pipeline, or increased cost savings for the business.

Data visualizations powered by ML provide automatic, unassisted pattern recognition, making it easier to forecast based on trends in historical data. “Explain” capabilities examine selected datasets to identify meaningful business drivers, contextual insights, and data anomalies, then generate plain text, vernacular descriptions about what’s been found.

Oracle Analytics Explain attrition example   

Speaking of plain text – ML algorithms also power AI-based natural language processing (NLP) and natural language generation (NLG) to help business users find relevant data and draw key conclusions. Using NLP to query information from a dataset means that a user can simply type (or verbally ask if using a mobile device) a question using their natural language, without knowing where the data resides nor the composition of the data set.  NLG gives them easy-to-understand, written textual narratives in return, providing an instantaneous explanation of what’s going on in their data.

Sentiment analysis is another powerful ML-enabled tool in the business user’s toolbox. Simply set up a data flow to pull in survey data, then perform text analysis to extract words from this unstructured data, classify and count them, and visualize the results. For example, HR teams can leverage this capability to get a better understanding of how a new benefits package is being received by employees. Marketing teams can run the same process on social media inputs to see how a new product release or campaign is performing. Letting a model do the work saves these teams countless hours of manual analysis of survey results and market research, and allows them to make changes faster, leading to more satisfied employees and more successful launches.

Business users are expected to be data-driven, but embedded ML capabilities can nudge them towards becoming analytics-driven without putting additional work on their IT teams.

Doing the prep work

Successful and accurate ML capabilities rely on the quality and availability of the data being fed into a specific model.  To make analytics-driven behavior second nature to the business user, an organization must first properly prepare the relevant data sets for input into specific data flows.  After all, the most accurate ML algorithms should be trained on a complete dataset at the lowest granularity. The process of maintaining data storage and moving data around lies within the purview of the data engineer, so the true first step of data prep is the connection and sourcing of the correct data sets from the correct sources, even across different departments or storage mechanisms.  While data engineers themselves are unlikely to execute business-motivated ML activities, they are responsible for ensuring that the best data sets are available to support ML activities that lead to better, more trustworthy decisions.

Read more: What is data preparation and why is it important?

Small DIY projects

Citizen data scientists straddle the gap between business users and data scientists. The ML capabilities they use usually do not require the same level of details and scale as those of a true data scientist.  Citizen data scientists need ML tools that enable them to complete smaller or more ad hoc projects on an as needed basis applying pre-defined algorithms to their scenarios.  These tools need to provide customizable algorithms that best fit the selected business application or problem, but still be easy enough to use with a code-free, point-and-click style interface.  Citizen data scientists can then run those prebuilt, and customizable, ML algorithms within a sand box to check their accuracy against historical test data.  Automatic generation of performance statistics enables the citizen data scientist to adjust the ML algorithm’s parameters to make it more accurate.  Once the customized models have been through various iterations of testing and tuning (and are producing acceptable results), with a platform like Oracle Analytics, the citizen data scientist can then publish them as registered enterprise, pre-trained models within the analytics platform for all business users to access.

Here’s an example: an HR team hit hard by “The Great Resignation” may ask their resident citizen data scientist to create a predictive model of employee attrition.  Realizing that they do not yet have a standard attrition prediction model, the citizen data scientist will take attrition data and create their own ML classifier model (Create and Use Oracle Analytics Predictive Models) to predict attrition within the analytics platform.  First, they train and tune the model, then apply the model against new HR data to deliver analytics-driven insights on attrition.  Then, as the activity grows from a bespoke one-off solution into a more formal and institutional process, the data scientist can take ownership and may label that trained ML model as the new corporate standard.

Oracle Analytics gain and lift chart to evaluate an ML model’s performance

 

For data professionals

Where the citizen data scientist’s ML tasks end is where the data scientist’s ML tasks begin.  Data scientists typically have access to much larger and varied volumes of data and require the most detailed level access to optimally tune ML models for specific use cases.  The ML platform should provide both the ability to train and fine tune models, and a structured approach to certifying, publishing, and sharing ML models.  ML models that have demonstrated significant business value can then be operationalized for on-demand use directly by business users, or automatically executed via scheduling.  Processing ML models can be resource intensive.  Cloud platforms provide a flexible approach to elastically scale up resources for process intensive tasks to save time, but then scale those resources down again during less intensive periods to save cost.

For example, a data scientist is developing a model that will predict employee attrition.  Their ML tool needs to help them create, test, tune and publish this model.  AutoML provides part of the solution by quickly identifying the right model for the use case.  Using AutoML, the analytics platform will create, evaluate, and recommend the best solution from a collection of classifier models – showing the most influential factors and model scores that led to the recommendation.  This is a great start, but the data scientist then needs to examine the model in detail— to see “inside the box”— in order to both understand how the model came to the provided results and be able to manually fine tune the model.  Oracle’s AutoML models are fully transparent and explainable.

See more on Oracle’s AutoML

Conclusion

Getting value from ML isn’t just about letting the data science team work with fancier or more complicated algorithms – it’s about making ML easier to use and more accessible to anybody within the enterprise.  There are a lot of smart technologies that assist users in transforming their data into usable insights that support business decisions.  Everybody agrees that being data-driven improves decision making, however, that is not enough because a data-driven approach simply uses a static historic report with numbers.  Using embedded ML to transform those numbers into future projections and recommendations, thus becoming analytics-driven, is where the real business opportunities lie.  The more ML-enabled users, regardless of skill level, the more accurate the business decisions.

For additional information, visit Oracle.com/analytics, follow us on Twitter@OracleAnalytics, and connect with us on LinkedIn.

Nick Engelhardt

Senior Director, Oracle Analytics

Blair Bozada

Senior Product Marketing Manager, Oracle Analytics

Senior Product Marketing Manager, Oracle Analytics

Barry Mostert

Senior Director, Artificial Intelligence and Analytics

Barry is a senior director for product marketing covering Oracle's AI and Analytics services.


Previous Post

Advisor Webcast: Oracle Analytics Cloud - Troubleshooting Tips for Remote Data Gateway & Private Access Channel

Tanya Heise | 2 min read

Next Post


How to Connect to Oracle Autonomous Data Warehouse for Oracle Fusion Analytics Warehouse Using SQL Developer

Krithika Raghavan | 3 min read