Jeff Catlin, EVP of AI Products, InMoment, the leading provider of integrated Experience Improvement (XI)™ solutions.
The explosive adoption of large language models (LLMs) within all types and sizes of businesses is well-documented and is only accelerating as corporations build their own LLMs based on local LLMs like Meta’s Llama 2. An October 2023 Gartner, Inc. survey found that 55% of corporations were piloting or releasing LLM projects, and that number is expected to increase rapidly.
One likely, if unexpected, side effect of this increased usage of LLMs to wrangle the ocean of unstructured data (as much as 90% of a company’s data is unstructured) is that natural language processing (NLP) engines will become more important than ever.
I hear a collective question coming: “Wait, won’t LLMs mean I won’t need NLP anymore?” Sadly, that is not the case—in fact, quite the contrary. LLMs are magical at following someone’s lead to answer a question or prompt as well as dig up important, hidden facts and opinions. However, businesses run on metrics, numbers and trends, and you need to be able to dig out all of the relevant metadata to drive your business.
NLP is adept at mining a wide variety of content types and tying them together with common metadata elements such as:
• Named entities: People, companies, places, emails, phone numbers, etc.
• Key concepts: Key ideas within the content (two-word and three-word phrases).
• Categories: Bucketing content into your business language/taxonomy.
• Sentiment: Positive, negative, neutral, mixed.
• Intention: Useful in customer interactions to detect churn, propensity to purchase, etc.
• Effort: Useful in customer support to see how easily a client issue is solved.
There are other extractions beyond these, and all of them help a business tie its disparate information sources together through a common lens—the NLP extractions.
As a software and services vendor that has pioneered machine learning and AI in the customer experience space, we have fully embraced the power of LLMs to change the business landscape. However, we see it as something that will have even more impact on a business when combined with NLP technologies.
A system could show you how your call center agents are doing against your standard metrics. You might notice that the average sentiment of callers from the Northeast is down, so you ask the system to tell you what’s driving down customer sentiment in the Northeast. The LLM then answers the question with a few possibilities that you can dig into and verify.
An LLM alone wouldn’t alert you to the issue, and NLP alone would be challenged to tell you why sentiment is down. However, the combination will allow business discovery opportunities that were never before possible.
I expect that the need for an ever-more powerful NLP isn’t something most people riding the LLM wave have realized yet. Within the next 12 months, however, I believe there will be a renaissance in NLP usage as the value of a combined NLP and LLM-based system becomes apparent. This combination will allow corporate data lakes to fully live up to the promise that data lakes have long been pitching.
An NLP engine is ideal for processing all forms of content within an organization—structured, semi-structured and unstructured—and producing a common key for all content along with metadata like categories, sentiment and such to form a unifying set of content for all of the applications within an organization.
There will be bumps along the way as companies struggle with the heavy cost and hardware requirements along with LLM hallucinations, but the power of NLP-tagged content will allow us to go deeper, examine the LLM assertions and request specific examples of these assertions within the corporate data lake so that we can read and verify the content being used to make these statements.
LLMs aren’t coming; they are here, and they will change businesses everywhere. Those businesses that see the value of NLP and LLMs together will be the big winners in this changing world. They will be able to maximize the investments being made in LLMs and will be faster to market with interactive LLM applications that let users investigate their information at a deeper level.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
