Where to find (or create) hidden value with gen AI

Companies of every stripe, sector, and size this year are hoping (or expecting) to hit some paydirt with generative AI. But new research from PwC shows that not all use cases hold the same potential. Two distinct types, according to the study, are expected to generate over half of gen AI’s aggregate business value:?

• Net-new creation—creating entirely new content based on a user-provided prompt, such as generating images for marketing content—will generate 33% of overall business value.
• Augmentation—expanding on existing content to fill in gaps, such as auto-completion of written text, emails, and presentations—will generate 21% of overall value.?

One interesting wrinkle in PwC’s findings? One of the biggest cross-industry use cases for gen AI—conversational AI and chatbots —accounts for a relatively slim piece of the overall pie. “Only about 15% of the potential GenAI value rests with the summarization and dialogue (Q&A chatbot) patterns for which early GenAI services have become known,” writes Bret Greenstein , a partner and generative AI leader at PwC and lead author of the report.

While there’s no arguing the immense business impact that chatbots are expected to deliver in the realm of customer and employee experience, the study should remind executives to look deeper into those first two categories (net-new creation and augmentation) for new applications that may have major hidden potential.?

Channeling Capt. Kirk in prompt design (no, really)

Dan Tynan ’s latest story in The Works does just that: He reports on an array of creative—like, really creative—strategies that some gen AI pros are employing with prompt design. Researchers at VMware, for example, recently set out to measure how sprinkling positive thoughts into a prompt helped improve the quality and value of the response.

Ultimately, they found that flattery wasn't nearly as effective as asking large language models (LLMs) to automatically optimize their own responses, Tynan writes. The most effective optimization technique they discovered? The team asked the bots “to pretend to be captain of the Starship Enterprise.” No joke: Channeling Capt. Kirk in the prompts “produced more accurate responses to grade-school math questions than telling the AI it was ‘highly intelligent’ or a ‘professor of mathematics.’”

Who knew? Not many. Data scientists don't fully understand why chatbots can be manipulated in certain ways, such as offering rewards or encouragement, using emotional appeals, being told to ignore internal safeguards, or being asked to adopt a particular persona. All they know is that these techniques often work—at least for now.?

One rule of thumb that experts agree on: Because LLMs have been trained on language generated by emotional beings, they respond more positively if you treat them like one.

Read the full report on The Works .

Moving past the gen AI starting line

Looking for other ways to wring value from your gen AI pilot projects and trials? Much of it starts with clearing out some common or immediate roadblocks. Check out our recent report in The Works that notes the key factors holding back companies’ progress in advancing generative AI initiatives into use cases that blossom on the bottom line.

Check out the story on The Works .

Enjoy what you’ve read? Don’t forget to subscribe to receive future Freshworks Connect editions.