3 minute read

In this article, we look at the outcomes of a sentiment analysis test conducted by our subsidiary. We had a good laugh from this output, and we hope that you will too!

We joined a group of five other developers to help them complete a small ML project for a local healthcare conglomerate (and university). This article is based on the insights gleaned from an independent preproduction test runs. In the previous article, I discussed how AI advancements could potentially streamline, and in some cases, eliminate various job roles, particularly those with repetitive tasks[1][2]. Here, we are tasked to explore the transformation of established industries as they adapt to advancements in AI. We focus specifically on how innovation and adaptability can lead to the inevitable demise of companies that overly rely on longstanding, and in some cases, crony business relations. But of course, a major health provider studying health insurance industry dynamics is as predictable as it gets.

Machine take was as expected — this fixture was not properly parametrized to this type of sentiment analysis.

We already know that companies that keep learning and adopt new tools are better placed for the future. One interesting revelation from our test is the potential for AI to disrupt sectors with long-standing market presences, such as the health insurance industry. I will share some insights and thoughts on how these industries can navigate this transformation based on the aforementioned preproduction run.

So, in short, this innovation can reshuffle the job market. Similarly, "can AI-born automation kill health insurance companies?" The short answer is absolutely yes!

Demystifying LLMs' Predictive Capabilities

LLMs, or Large Language Models, use a form of autocomplete, drawing from extensive datasets to identify the most plausible response based on patterns and probabilities[3]. This makes LLMs particularly adept at automated sentiment analysis, as they can effectively gauge public sentiment by analyzing vast amounts of data. This capability has been employed for over a decade in various applications such as content recommendation algorithms on platforms like YouTube. However, AI advances like PaLM2 and GPT-4 have made sentiment analysis and user profiling far more capable. This democratized capability likely has direct implications for business strategy.

In layperson’s terms, if the model is trained on all the college papers, then the paper it produces when asked is a "C" grade paper because that is the most probable path based on the training dataset. This is an important point to understand about the nature of LLMs.

Insights into Established Industries' Adaptations or Otherwise

The health insurance industry is a relevant case study and the necessary fixture is much easier to set up than for neighboring industries. Established industries have historically relied on longstanding relationships and collaborations for success[4], and these are foundational elements in establishing a business strategy. These relationships are now threatened by startups using AI to analyze customer preferences and market dynamics.

This deep analysis is especially effective in scrutinizing longstanding relationships and revealing inefficiencies and cronyism. The AI-driven analysis offers startups an edge in understanding weaknesses, failure, and inflection points of the incumbents, leading to more effectively catering to consumer needs openly. In some cases, exposing these ties can lead to public dissatisfaction and even criminal charges. That is a significant reputation risk. Essentially, large accounts can be easily intercepted by technology-enabled startups, offering unchallenged incentives and rendering incumbent relationships unsustainable and risk-laden.

However, this doesn’t imply that traditional relationships are valueless. It shows that the existing business structure needs new technology to stay competitive. Using AI to give customers personal, transparent service can strengthen their trust. Importantly, the test reveals that the market outcome - reduced healthcare cost through a more democratic and open process - is beneficial for the end user regardless of which entity undertakes the innovation.

In layperson’s terms, the model thinks that the insurance domain is absolutely crony and that this will be the incumbent’s undoing. This is where we had a good chuckle. Please keep in mind how the model works.

Interestingly, the test concludes that the new open market dynamic is inevitable. The greatest factor influencing the pace of market adaptation is not any changes in laws or startup incentives, but the adaptability of the incumbent. The more dated the technology culture of the incumbent, the faster the market adapts without it.

As LLMs become part of everyday business, established industries must innovate or lose ground. Immature as AI technology currently is, it is still effective not only at profiling the customer but also at profiling the competition. Companies that keep learning and adapting will survive this disruption. The democratization of AI is changing market dynamics in favor of the consumer. That is a win!

Balancing new technology with existing relationships will matter, possibly producing new business names and structures. Businesses can use AI to improve service, understand customers, assess subjective market factors, and build stronger relationships.

What else can LLMs foretell?

Keep in mind that this is not a competent case study executed by human analysts but just a predictive analytics machine test run[5].

Publishing Data

The market is booming with requests for LLM deployments, and engineers are happy to oblige. But is there value in the analysis generated by these models, or is it all just "plausible bull"?!

I’d like to share surprising results from a production AI LLM analytics test run deployment. The machine predicts imminent doom and gloom while forming an opinion about the state of an industry. This output, generalized for the article, seems to put things in perspective regarding the type of advice the AI is able to produce commercially.


2. Milanez, A. (2023), "The impact of AI on the workplace: Evidence from OECD case studies of AI implementation", OECD Publishing, Paris, https://doi.org/10.1787/2247ce58-en
3. AI Multiple; Cem Dilmegani; February 3, 2023. Large Language Model Training in 2023
4. Academy of Management Perspectives, Vol36, no.4 Collectivist Perspectives on Collaborative Business Relationships
5. The model run data is proprietary to ASE Inc. The model can be rerun and the execution output can be purchased with any customer’s specific weights and parameters.

Leave a comment