Is GenAI good for anything?

Executives and business leaders may be confused. On one hand, there are all these hyped claims how Generative AI (GenAI) improves employees’ productivity and can reduce headcount. And on the other hand, there are many studies on how GenAI frequently hallucinates, fails basic tasks, and gives up on more complex problems. And there is no positive ROI in sight.

Well, GenAI is not for everything. But there are specific business needs for which GenAI is the perfect solution. These are the cases where it’s worth it to ensure GenAI is adopted by employees.

What is Generative AI?

To understand what something is good for, it helps to understand how it works.

Artificial intelligence is the machine's ability to perform cognitive functions we associate with humans. One of these cognitive functions is natural language processing. This is what Large Language Models (LLMs) are developed for. They are created with large neural networks and deliver GenAI tools such as ChatGPT, Claude, and other AI chatbots.

LLMs are trained on a vast amount of free text, images, and videos found on the Internet, and sometimes on vetted content from books and news outlets. During their training, these AI chatbots learn how humans communicate. And since it’s the Internet, they learn the good, the bad, and the ugly. You should expect all this from an AI chatbot.

The responses to human queries that AI chatbots deliver are based on the probability of words following other words. Due to their training for natural language processing, these tools generate responses that sound eloquent and very convincing. But they don’t understand the context. And the LLMs that are called “reasoning models” are more “explaining models” since they don’t reason. They just explain how they arrived at the outcome.

As a side note, let’s keep in mind that GenAI delivers only 20% of the business value that AI can deliver for companies. The other 80% comes from Analytical AI, according to McKinsey. The focus in this article is on those 20%.

When is GenAI good for business?

There are four business needs where GenAI delivers low-risk, high-return value. Humans should still validate the outcome.

1️⃣ Text rewriting, correction, and translation

Based on their training data, LLMs know human languages very well. They can rewrite complex technical text into a language understood by non-experts. They can identify spelling mistakes and grammatical errors. And they can do that in almost any language, making them useful for translating text.

2️⃣ Text synthesis and summary

Taking existing text and being prompted to use only that text, LLMs can quickly synthesize content, such as company documents or production manuals, and summarize it. However, depending on the LLM and any context provided in the prompt, summaries may differ since each LLM may focus on different parts of the text.

3️⃣ Image and video creation

With thoughtful prompting, LLMs can quickly create images for marketing materials. They can also inexpensively create videos for training purposes using avatars. This is especially useful for training videos with content changing only slightly over time. And of course, these videos can be easily created in almost any language.

4️⃣ First level customer support

AI chatbots can address basic customer questions and provide appropriate responses. They need to be monitored, however, for complex queries and questions outside of their training data. Since AI chatbots can recognize emotion in a human’s voice, they can bring a human supervisor into the call when detecting anger or distress in the customer’s voice.

When is it too risky to use GenAI?

There are many other uses of LLMs, but they come with higher risk. For companies, the following uses may lead to being the laughing stock on social media at best to loss of customer trust and revenues or regulatory penalties at worst.

❌ New content creation with potential of hallucination and factual errors, such as quoting sources that don’t exist

❌ Vibe coding since LLMs are trained on ANY code and cannot distinguish between good and bad code (initial coding ideas or simple standard building blocks verified by experienced programmers may be OK)

❌ Research if you need high accuracy and evaluation of contradicting data points that require understanding of the context, which LLMs lack (complement with regular search)

❌ Data analysis since LLMs are not trained for math (use statistical and machine learning models instead)

ChatGPT and other AI chatbots can deliver business value, but only for a small subset of business problems. While they are also used to address other needs, be aware of their limitations and trade with caution.

Knowing how Generative AI works helps to focus resources for employee adoption on the areas where GenAI can add value.