Contrary to what some folks think about the “recent” discovery of artificial intelligence (AI) for business uses, AI as a computing function is not novel. What is new is its growing adoption for expanded uses and the ability to transform data into actionable business strategies.
AI has been around for a long time, observed Daniel Ziv, VP for experience management and analytics, GTM Strategy at workforce management firm Verint. AI is not one thing. It has a variety of capabilities depending on what it is designed to do.
For example, one of the chief components of AI’s various elements is large language models (LLMs), which have a longstanding presence in the field. Innovation in their capabilities resulted from the emergence of advancements that exposed the power of natural language understanding and natural language generation about 18 months ago.
“That work has been evolving and building for many years,” Ziv told the E-Commerce Times. “It exposed awareness because it was publicly accessible for anyone to try.”
AI’s Pivotal Shift in Business
A meaningful turning point is accelerating both the need and the opportunity for automation platforms that organizations can leverage in new ways, Ziv noted. For instance, generative AI is evolving and getting smarter and more proficient at understanding language.
One key element in AI’s growing business adoption is cloud computing, which can process more data much faster and at a lower cost. Ten years ago, companies deployed most AI software on-premises. Adopters had to buy hardware, provision it, install software and train everybody.
“It would take months — sometimes years — to get the value that now you can get sometimes in days or weeks,” Ziv said.
Today’s challenge is learning how to leverage AI’s advancements over the last two years to transform massive data for speedy analysis and recommendations. Data transformation has many approaches depending on the types of data collected, such as structured and unstructured data.
“Structured data tends to be numbers, and computers have been running on structured data. Computers are very good at building models and doing things based on numbers,” he said.
The transformation process becomes more complicated with unstructured and semi-structured data, which includes unstructured elements like text, audio, or video and some metadata associated with it.
“In the past, that was more challenging for computers. Today, with generative AI, the technology has caught up and can do it much faster,” Ziv explained.
Refining AI for Tailored Business Insights
Verint has used AI for decades to help companies get a handle on using their data more effectively. It has helped its customers work with a range of accuracy issues.
“In our industry, I think people might perceive that transformative data is not so accurate because we’ve taken general LLMs trained on internet data that is not specific to your business. It is not behavioral data. So, what it learned to do is kind of like babies as they learn to speak,” Ziv suggested.
So far, we have trained our AI to understand language in general and to be able to respond to some level. But the AI’s comprehension is much like a baby still lacking the right knowledge, information, and experiences to give educated answers on things that directly relate to the desired results, he added.
AI developers are continuing to learn how to make that baby grow into an effective adult. The solution, according to Ziv, is to take that ability to understand language and generate language with the correct behavioral data specific to interactions you have with your customers or organizations have with their customers.
“We are at the beginning of this transformative phase. But I do believe that the competence to write data with an open platform and the power of generative AI will allow us to see things that are very compelling and will allow us to automate,” he observed.
The Journey Toward Predictive Accuracy
SoundCommerce is an example of why using data to predict actionable results is not a one-size-fits-all process. The company takes a different approach than other data management providers by using a no-code environment accessible to everyone.
The company’s CEO, Eric Best, noted that the data transformation pathway is littered with challenges. The process involves extracting data from a source system and customer data from the client’s CRM platform.
Then, the data has to be validated to contain reasonable quality. According to Best, the next step is applying the data to address a particular problem that SoundCommerce is working to solve: ascribe meaning to the data as it flows.
“That is important because, by the time you get to the data warehouse, where your analysis is going to happen, you are going to make these all-important business decisions,” Best told the E-Commerce Times.
To make that happen accurately, the data must be converted from one format to another to create compatibility and similarity. For example, for most retail brands, orders come from multiple sources in addition to a cash register or point-of-sale system. These vending sites often include an e-commerce storefront, an Amazon Marketplace business, and a proprietary mobile app.
“Seeing all four of the order data records in a common format and schema is an area where AI can be helpful,” Best said.
AI Mapping Without an Engineering Degree
To get accurate results from the combined data feeds, you have to be able to describe the data in natural language terms. So, in order to get the AI to help with this data mapping problem, you need to tell the AI in very verbose, natural language terms what data you want and how you want to define the data.
The solution is having the AI write the software to effect that change that transformation on the data. So instead of being a really good software engineer, you need to become a prompt engineer, Best explained.
“People have to be very good at describing what they want, not in coding terms but in natural language terms. Accuracy in speech and writing becomes super important.”
SoundCommerce customers are just beginning to experiment with these generative AI algorithms. Some of that AI enablement is done by the company using its own proprietary algorithms around things that are very specialized for its customers, Best noted.
One proprietary code example is the ability to forecast the lifetime value of an individual customer or shopper. When it comes to the generic capabilities, the generative AI work innovation comes from Microsoft, Google, Amazon Web Services, and an independent specialty data warehousing company called Snowflake that Best’s company works.
Those cloud platform companies are generally building their own generative AI tooling with their own proprietary large language models.
Modern AI, Timeless Business Questions
How cost-effective and practical is this high-tech AI decision-making capability for business? The answer to that question depends, quipped Best.
For less technical organizations, the new technology’s practicality for smaller companies increases the more tightly you can define a use case. SoundCommerce had to learn this the hard way, he admitted.
Best uses an age-old reality to answer the practicality versus cost-effectiveness question. For more than a century, people have been figuring out where to spend advertising dollars effectively.
“So, the questions and answers are not new. The ability to automate the answers at scale is definitely new,” Best concluded.
Loading ...
