How AI Could Transform Email

What if your inbox were jam-packed with AI-generated emails? You may already be on the receiving end of emails written by artificial intelligence, with the help of a human prompter. Austin Distel, a senior director of marketing at Jasper, is one of those humans. 

Austin smiles as he demonstrates Japer’s knack for email composition. “These are tools in my tool belt that helped me perform faster, but also better,” he says before sharing that he often uses generative AI to rewrite work emails so they sound like Jerry Seinfeld.

What’s the Deal With Autocomplete?

Email is one the least authentic forms of communication. The stock phrases and courtesy replies are comically robotic. Like, why do you hope this email finds me well? Google has used machine learning in Gmail for the past few years to generate one sentence replies and predict what you’re likely to type next. Newer companies, like Compose AI, might enable people to further rely on autocomplete functions while sending emails.

“Autocomplete is not going to sway you away from what you already want to type. It’s just going to accelerate it,” says Michael Shuffett, a founder and CEO of Compose AI. Although large blocks of AI-generated text from models like OpenAI’s ChatGPT are quite popular, smaller snippets crafted by autocomplete are easier to control at the present moment. It’s akin to riding a tandem bike with a robot versus letting it ride your bike alone at night after giving a one-sentence command. The first option requires more effort, but you’re more likely to reach your intended destination together.

Will major email providers, like Gmail and Outlook, roll out additional AI-powered features to help you breeze through an overstuffed inbox? Aparna Pappu, a vice president and general manager at Google Workspace, mentions multiple ways the company uses AI to assist people when writing messages and detecting spam. In a statement sent to WIRED over, well, email, her attitude on generative AI mirrors most of the messaging coming from Google on the topic: excited, albeit cautious.

“We are in a new era of AI where large language models have the power to take that helpfulness to the next level,” writes Pappu. “We know it’s essential that this work is done with the utmost care for safety, quality, and groundedness.” Businesses who use Microsoft’s customer relationship management system, Viva Sales, can experiment with a generative AI tool for drafting reply emails. Potential canned prompts for the AI include stuff like discount offers and question replies. Microsoft declined to comment for this article, and reaching out to its Bing chatbot for a quote about Outlook felt kinda trite. Sorry, Sydney! 😊

Will the AI Hit Reply All?

So, the future of email may have a more comprehensive autocomplete. Is that revolutionary? Not really.

It would be transformative if a personalized AI model could write complete, high-quality emails in your tone of voice. White-collar workers who spend their days tap, tap, tapping responses on their laptops would essentially get access to a digital executive assistant. Imagine a world where robots spend their days “circling back” on projects and sending “quick pings,” while professionals laze about sipping Mai Tais and focusing on client presentations.

Multiple startups are building toward this reality. Matt Schumer, a serial entrepreneur, sat in front of his computer in 2019, years before the ChatGPT phenomenon, responding to an influx of emails from partners and customers and investors; it all felt a little overwhelming. So, when OpenAI released its GPT-2 model to the public, Matt wondered whether it could be used as a productivity tool. 

He scraped the information from his inbox and trained a model on the data. “Then, I asked it to write some new ones for me,” Schumer says. “I was immediately blown away by what I saw.” This experience put him on the path to becoming cofounder and CEO of OthersideAI. It’s the company behind HyperWrite, a generative AI tool for composing entire paragraphs and rewriting complicated sentences.

Despite visions of a future with full self-writing email software, most marketers and CEOs we interviewed admitted the technology was not yet at a high-enough caliber to be trusted with complete control of your inbox. Philippe Lehoux, a CEO of the workplace communication company Missive, says, “This technology will improve, but anyone right now aiming at using AI to replace themselves and do automatic responses is pretty much delusional.” John Humphrey, a head of data platform product at Mailchimp, compares generative AI to an intern who may pitch some stellar ideas, but whom you wouldn’t let spearhead a major campaign.

Well then, what is it good at? In addition to drafting replies, two other emergent use cases for AI in email are shifting tones and brainstorming potential subject lines. Sure, generative AI can rephrase your emails to sound like Jerry Seinfeld. With alternative prompting, it can also draft messages that are more direct or serious or apologetic. Make sure to proofread everything before hitting “reply all,” because large language models continue to have a penchant for hallucination.

AI tools can be used to analyze data from top-tier email subject lines—those with impressive open rates or sales conversions—and spit out a hundred good ideas. “The principles of what makes a good subject line? AI didn’t change that. It just makes it easier to produce that,” says Humphrey. It remains up to the marketer to divine which angle is the best fit for their customer base.

Business leaders should be cognizant of the thorny controversies around AI-generated text, in order to avoid alienating email recipients. Vanderbilt University apologized after using ChatGPT to message students about a shooting at Michigan State University. How could an administrator think this would be acceptable? People who approach generative AI as a tool to unlock the unique words needed during difficult conversations may misunderstand the technology’s strengths. It predicts next steps and generalizes the human experience, which is great for understanding potential buyers, but not so great for connecting with those in mourning.

“We all see such generic messages every day,” says Distel from Jasper. “So, how can we break through the noise, and say something that gets your wheels turning?” Well, you’ve got my wheels spinning. I’m just not sure where this wild ride will come to a stop.

120233aanartificial intelligenceBingbusinesschatbotChatGPTcommunicationemaileraGooglehow-tolanguagelaptopslearningmmachine learningMarketingMediamicrosoftModelsNotofOpenAIPratesrobotsSSafetyStartupsTtargetWorkworkers