How embedded AI is making analysis more actionable

May 25, 2022 | 4 minute read
John Menhinick
Senior Director, Product Management
Text Size 100%:

Historically, businesses have turned to data analysis to understand more about their business operations and determine what practical actions they can take to improve performance. Organizations have often found that such analysis raises as many questions as it answers, fuelling more analysis but leading to little change that delivers true business value. The inability to make this analysis truly actionable creates the challenge of analysis paralysis.

This phenomenon has discouraged many businesses from becoming more data-driven and has slowed the optimization of business processes as a result. The barriers to making analysis actionable have included the following factors:

  • A lack of quality, relevant data to generate insights that look credible
  • The skills needed to interpret the data and translate the analysis into meaningful outputs
  • When data models are built, the overhead associated with converting the analytical code so it can run automatically on a database and adding database fields to hold the model outputs and modifying the UI to accommodate them

What’s stopped organizations before

Non-machine learning models also have had static factors and weights set by the analyst, which means these models need to be rebuilt repeatedly to ensure accuracy over time. Even AI models have issues with making them actionable. Research indicates that under 15% of AI models built make it into production—as low as 12.5% if you ask AI practitioners.

        Source: Western European Artificial Intelligence Survey, December 2021 (IDC Doc #EUR148814022)
        Source: Western European Artificial Intelligence Survey, December 2021 (IDC Doc #EUR148814022)

Transforming data analysis projects into actionable insights has seen real challenges, including the building of data models. The technical, operational, and financial overhead associated with ongoing development, plus the work needed to make these projects accessible, has been significant and sometimes prohibitive.

A recent study by IDC identified a range of factors that are preventing organizations from progressing from proofs of concept (POCs) to production usage for AI projects. (See below):

Source: Western European Artificial Intelligence Survey, December 2021 (IDC Doc #EUR148814022)
Source: Western European Artificial Intelligence Survey, December 2021 (IDC Doc #EUR148814022)

The most quoted reasons include the cost of AI development tools, the cost of the associated computing and infrastructure services, and a lack of skilled personnel. The market has needed a way to benefit from AI but in a way that addresses these issues, including making it more affordable and more useable to overcome the skills gap.

Improvements for actionable insights

Today, the world has become far more data-driven with machine learning-based analysis providing accurate, self-updating models with outputs embedded directly into core business software. This method of AI delivery minimizes the time it takes from running the analysis to the delivery of meaningful insights. These insights are actionable because of their placement directly into the hands of people that need it most, so they can work more expediently, making workflows more efficient.

These AI features are now being embedded into a broad range of business applications that run operations across front- and back-office functions. The whole enterprise can now put AI-powered insights into action more quickly and efficiently than ever before. When AI is consumed this way, businesses receive all the benefits without the investment, technical integration requirements, or management overheads typically associated with AI and analytics projects.

Embedded features can draw on the data held in the data model underpinning the business application and system telemetry. The data feeding the AI reflects both historic and current data and is based on every interaction and transaction. Some software vendors are also investing in data assets to supplement the customer’s own data to fortify the AI models that these features depend on.

Actionable insight is provided in several forms directly to the software operator, including segment names, propensity scores, ratios, values, recommended actions as text, and predictive data entry to remove the need for operators to enter the data manually, which is laborious, time-consuming, and more prone to error.

Another advantage of this type of AI is its relevance to users of the software where it’s embedded. These solutions address directly related use cases, such as a lead or opportunity management solution embedded within CRM software, or a recruitment solution embedded within HR software.

Conclusion

While many were sceptical about when they would see value from their analysis projects, with embedded AI business, leaders can now be confident that data analysis can lead to actionable insights quickly. Supporting the workforce with actionable insight is the best way to optimize performance and thrive in today’s highly competitive marketplace.
 
For more information on Oracle AI, see Oracle Artificial Intelligence.

 

John Menhinick

Senior Director, Product Management


Previous Post

How finance teams can partner with the business to co-create value

Stacey Gaasch | 4 min read

Next Post


What happened to Hyperion?

Wayne Heather | 5 min read