Brands and agencies are asking what generative AI means for them. What can be automated? How can we create value? But as with any shiny new object, companies should proceed with caution.

The biggest mistake is employing a bot for a task that humans can do better. Much is written about a bot’s ability to write flawless emails. And some companies are using it in corporate communications and even in marketing—where creativity is prized and emotional connection can drive sales.

Don’t blindly outsource human connection to a machine. Colleagues and more importantly your customers will know when they receive a machine-written email. It’s degrading and dehumanizing. A connection with others requires a human touch that understands emotions and nuance.

Nor should ChatGPT and AI be used in the hiring process by employers without full transparency for candidates. To lure talent amid very low unemployment, don’t risk losing prospective employees when they discover they’re talking to a computer.

The devil is in the details

There are many appropriate uses for generative AI, but the devil is in the details.

Because large language models are good at synthesizing reams of information, it’s smart to use them to accelerate product development. Let’s say a European company aims to copy a U.S. internet service. The interface can be created easily as the coding is more or less standardized. There largely isn’t any innovation in such a business.

On the other hand, attempting to use AI to create breakthrough innovation could be stifling rather than enabling. After all, large language models can’t inherently create anything new. If your company leans on ChatGPT to innovate, you will find that employees’ creativity and analytical ability will dull over time, much as math skills have been diminished since the invention of the calculator.

The better method is for humans to first brainstorm to include as many possibilities as possible. Then they can use ChatGPT or another AI tool to check that no obvious angles have been missed. AI tools can also be used to speed analysis and prioritization of concepts, going so far as helping to identify potential blockers that need to be considered in the launch and change-management plans.

Related: AI and marketing—news about ChatGPT, DALL-E 2 and other tools

While engineering managers are using AI to improve efficiency by as much as 25%, companies that go beyond one-size-fits-all use cases will benefit the most.

Create a cross-functional innovation team

Experimentation requires a safe space and a cross-functional innovation team to oversee it. An AI innovation team should first determine potential use cases, then identify which are likely to add the most value. Finally, specify which tech and tools are needed.

Such a team can help ensure data compliance, reduce cyber risk and test for meaningful returns on investment. It might comprise only a few people but should involve stakeholders from across the organization. Technology shouldn’t be responsible for determining how marketing can best use ChatGPT, even if it’s managing the sandbox where prototyping and testing takes place.

The skills needed to run ChatGPT and similar tools effectively have given rise to new roles, such as prompt engineers or prompt designers. What you get out of ChatGPT is only as good as the prompt you put in. Developing those skills now will give your company a leg up when you’re ready to start implementing generative AI more broadly.

Explore the competition

ChatGPT isn’t the only generative AI on the block; it’s just the one that got the public’s attention first.

Despite the fervor, this tech is still early in its hype cycle. Most Big Tech companies have some version, at least internally, and there is a plethora of startups. Aligning your business goals with the potential for AI tools will help you decide which to adopt. Manufacturing, for instance, has different demands than retail or media, and you’ll have plenty of opportunities to find an AI tool tailored to your unique challenges.

You’ll also want to know what other companies’ tools and services will help you use generative AI better.

Don’t rush to market—but don’t wait too long

ChatGPT has created polarized responses. Some schools have declared types of homework obsolete, and JPMorgan Chase has restricted its use. Meanwhile, media companies are jumping into the fray.

Both reactions are prudent to some degree. Waiting too long means you’ll be left behind. Moving too fast brings reputational, data and cyber risk. Consider Google’s rush to show off Bard, its version of ChatGPT. After an incorrect answer in Bard’s first ad, Alphabet’s market value temporarily slid by $100 billion.

Short-term wins are nice but major disruptions don’t happen overnight. Could ChatGPT and other tools upend the search industry? Possibly. Will they replace multitudes of human jobs? Maybe. But becoming an early adopter also means you’re less likely to be disrupted in the future.

Proceeding with caution means researching, identifying use cases and creating cross-functional teams to prototype, experiment and learn quickly and cheaply before large investments are needed.

Taking a smart approach to ChatGPT and generative AI tools overall sets you up for sustainable success. The strategies you put in place now will position you for the next hot new thing that comes to market, as it surely will before long.