The Prompt: Choosing your first custom generative AI use cases

Business leaders are buzzing about generative AI. To help you keep up with this fast-moving, transformative topic, “The Prompt” brings you our ongoing observations from our work with customers and partners, as well as the newest AI happenings at Google. In this edition, Phil Moyer, global vice-president for AI & Business Solutions at Google Cloud, shares tactical advice for getting generative AI projects off the ground.

One of my very first columns in this series discussed potential generative AI use cases, and months later, it remains one of the most-read. In hindsight, its popularity isn’t surprising: we’ve heard from enterprise leaders that they understand the technology is transformative but aren’t sure how to effectively and efficiently begin building custom generative apps.

In fact, customers often tell us that because generative AI is so open-ended, and evolving so fast, choosing a reasonable first project or use case can be daunting. There are so many potential applications, and so many commentators predicting that generative AI will eventually touch everything—where is a CEO even supposed to look first?

Shipping a generative AI project quickly might alleviate short-term pressures from shareholders, the board, or competitors—but if the project doesn’t build long-term competencies that accrue toward more ambitious goals, it’s at best a stopgap, not the first step in a scalable strategy.

So what kinds of use cases fit the bill, helping organizations to produce value fast while adapting to so much change? Let’s look at two of the more effective approaches I’ve seen among our customers, as well as best practices to help leaders home in on ideal generative AI starting points.

Identify information deficits

One effective tactic we’ve seen is to target data deficits within an organization. Consider this passage from Google Cloud’s recent ebook, “The executive’s guide to generative AI”:

"Every day, people in your business spend time and energy digging for information to make decisions, serve customers, and move your business forward. Informed decisions require information, and collecting the right inputs can take time.

Suppose you need to know how ad spend impacts customer perceptions. Or you want to see the trends in a competitor’s patent filings, R&D investments, and technology acquisitions. Information like this exists somewhere in your organization, often across a bunch of places. You need it to decide the next step — and, to obtain it, experts need to be convened, research completed, and the information compiled and synthesized. If you have a follow-up question, the whole process might have to start again.

From executives digging into strategic trends to salespeople creating product demos or new employees with benefits questions — everyone in your organization can relate to this frustration.

[But] imagine giving each person in your company not only a personal assistant, but an expert in every piece of data relevant to their job and, indeed, potentially every piece of data across your whole organization. With an assistant like that, impatient moments of indecision would dwindle. Everyone would be empowered to spend less time waiting and more time doing."

As the quoted paragraphs attest, use cases that target information deficits and enterprise search capabilities can provide an ideal onramp to generative AI. Such applications can respond to virtually any query by synthesizing and summarizing information from a variety of sources — a powerful capability for any organization that has proprietary data to activate or information-heavy jobs that could be more efficient.

Helpfully, this approach not only provides quick paths to value, by connecting employees with the information they need to do their jobs, but also offers an internal space to build expertise that can accrue to customer-facing apps. Customers, for example, often face information deficits of their own, whether they involve selecting the right product or finding the correct venue for questions.

Which data sources should augment a model’s training data? How should technologies like embeddings and connectors be deployed? Is model tuning necessary? These will be new questions for many enterprises, and by cultivating answers in a valuable internal use case like generative enterprise search, organizations can build the confidence, processes, and expertise for wider-ranging projects.

Strive to remove drudgery

In many cases, amassing the right information doesn’t just delay decisions or actions — it creates drudgery. Spending hours on keyword searches or manual document parsing is often less fulfilling than acting on information once it’s finally been found, for example.

The drudgery isn’t confined to finding information, either.

Creating new content, such as blogs or marketing messages, is often fun and rewarding, but it can quickly become a slog if writer’s block hits. Generative AI can help; out-of-box, many generative AI models can be great brainstorming partners for creative work, and with tuning, they can even help ensure projects follow brand tone and style guidelines.

Similarly, when employees are simply moving information from one format to another, as in the case of repetitive document processing, the task is less likely to inspire joy so much as tedium. Generative AI can change that, automating the most procedural, monotonous, and laborious tasks so employees can focus on more fulfilling work.

Customers face drudgery too, such as repeating information as they’re transferred from one customer service rep to another over the phone. Generative AI can help here both by improving self-serve options and equipping human agents with interaction summaries and context when customers are routed to them.

Like information deficits, drudgery creates friction for employees and customers alike—making it a prime target for early generative AI efforts.

Whatever the use case, start with a specific persona and task

Simple framing like “information deficit” and “drudgery” brings some organization to the potential chaos, but the range of possible AI use cases can still be dizzying. Marketing, customer service, document processing, research, internal knowledge management—in just this article, we’ve seen a variety of scenarios.

How can we narrow down even further?

Some things will be determined by a given enterprise’s specifics. If an organization has a lot of data sources, enterprise search is an obvious candidate, for example. But if the organization’s call center is overwhelmed, then customer service apps or apps to help human agents might be more pressing.

But individual contexts aside, leaders should focus on a specific persona and task. Generative AI may eventually change lots of things, but to get moving, start with a confined, scalable project of clear value. The goal is to build best practices and processes that will scale and transfer to increasingly valuable and sophisticated projects, not to transform overnight.

For a deeper dive, be sure to check out “The executive’s guide to generative AI” or our “2023 Data and AI Trends” report — and for an example of an innovative, specific generative AI project in action, see our recent article about how Wendy’s leverages Google Cloud technologies to improve drive-thru experiences.