Overcoming the Generative AI Catch-22

Generative AI is knocking — actually, banging — on the door, practically demanding to change how we do business. Despite excitement at the potential of the technology, many companies are finding themselves at a crossroads; they are eager to harness the power of generative AI but find themselves facing a complex maze of implementation challenges. The questions they’re asking aren’t about the technology’s potential, but instead, where do we start? Which use cases fit our business, and which ones will make the most impact? How do we know which to test first? 

In an O’Reilly survey on generative AI in the enterprise, identifying appropriate business use cases was the biggest hurdle to AI adoption. Why is identifying generative AI use cases so difficult for enterprises?

To dig into this and explore how to evaluate generative AI use cases, I sat down with Monica Cisneros and Peter Martinez  , who work with our generative AI development teams at Alteryx.

Our biggest analytics event of the year is coming soon, where we’re talking generative AI use cases galore – Inspire 2024. We’re offering $150 off the registration price for readers of the INPUT newsletter. Register here and use the promo code “Inspire24_NewsletterPromotion” at checkout.*

The difficulty with AI use cases

Why are so many enterprises struggling to find generative AI use cases? Ultimately, it comes down to risk mitigation and access. Many enterprises have created policies that ban using tools like ChatGPT and Gemini. This makes it difficult for users to experiment with the tools outside of their own personal applications, which then makes it difficult for leaders to find business-appropriate applications.

Monica explains it as almost a catch-22:

“There needs to be a very solid business case for anybody to start discovery with generative AI. But you cannot build a business case without having a pilot, and you cannot have a pilot if you're not able to discover use cases. You cannot discover use cases if you're not able to use the technology. And many businesses are afraid of leaking sensitive or confidential information.”

But why are use cases so important right now? There is great pressure from the board and senior leaders to see results from generative AI – 46% of board members say that generative AI is their top priority above everything else. Leaders are looking to their teams to find how to adapt generative AI for their companies, and picking your first use cases is incredibly important. Peter explains it in simple terms:

In an environment where knowledge workers are wary about generative AI, and leaders are eager and want to move fast, failure could make it difficult to get the widescale buy-in that these technologies will need to prove transformative. 

If, instead, you can show a quick win through generative AI, you can use this to build momentum and gain scope. This explains why, at this moment, picking the right use cases is on the minds of many IT and data leaders.

A framework for evaluating use cases

Peter explained to me a process for evaluating use cases – which involves this nifty graph.

Sit down and brainstorm all the potential use cases that would benefit your business and map them to this graph. You can even use generative AI to generate some use cases. Essentially, you want to find something that will bring high value but has a lower level of effort. You also need to pick a use case that aligns with your risk tolerance.

So, an example of a completed graph could look like this: 

By going through this exercise, you will be able to clearly see which use cases are the best opportunity for your business and can use your framework to map a long-term roadmap for leveraging generative AI across your business. You can even return to the graph and update it as your projects go into production and evaluate your results.

If you’re hitting a wall generating use cases, use generative AI to help you. Our new Auto Insights simulation can generate use cases for you from just a sentence of information. Give it a try.

Mitigating risk

Risk is a large consideration when finding use cases to explore. Peter and Monica gave me a list of a few things that teams can do to lower risk:

• Use low-risk data in pilots – Consider using data that is either already public or would have minimal downsides if exposed. Sales, marketing, customer support, and even public information are great examples.
• Consider a walled garden – A walled garden is a controlled environment that restricts access to data, content, and interactions. This allows experimentation with riskier data while minimizing the risks of exposure.
• Use synthetic data – For especially risky datasets, using generative AI to generate synthetic data may be the best way to run early tests. Once you show that the synthetic data use case works, you can move to using real data. 

Peter equates getting up to speed with generative AI to like training for a marathon – you set the big goal for running 26.2 miles, but your training plan doesn’t include running 26.2 miles on your first day. You work yourself up to the distance and build steps that will gradually get you there.

Newsletter roundup – AI use cases

With AI use cases on the brain, we’ve pulled together three articles from INPUT that you can use to guide your own genAI projects:

Generative AI Use Cases: Driving Innovation Across Industries

Auto Insights Playbooks: Reimagining the Analytics Development Process

AI in Finance: Today’s Applications and Tomorrow’s Possibilities

* The code is not applicable to pre-purchased tickets and cannot be applied retroactively.

Enjoying the INPUT newsletter? Be sure to follow us on LinkedIn for regular updates.