Alarmists concerned that generative AI will take over the world (or more importantly, their jobs) can take heart from the experience of Home Chef. The DTC meal kit company, which saw a massive influx of customers following COVID, needed a way to maximize engagement and retain elevated open rates for its correspondence. Using Phrasee, a Salesforce Integration Partner, has helped Home Chef set the foundation for producing great messaging on a mass scale, but it’s required lots of human attention, and 71 different tests, to achieve positive results.
The adoption of generative AI has proven fruitful. Between January 1 and May 29, Home Chef saw an average 7.9% open rate uplift and an average 83% click rate uplift. However, this wasn’t the result of simply handing over the keys to a computer and walking away. The company believes wholeheartedly in the importance of testing, whether a human or ChatGPT wrote the email copy.
“Whether your copywriter is going to create your content, or AI [will] or a combination of the two, one thing has not changed: when you send that message, you have no idea if it’s going to work,” said Jasper Pye, VP of Product at Phrasee during a session at Salesforce Connections 2023. “You have no idea whether it’s going to resonate. Your gut instinct will tell you that you do, because you think you know everything about your brand and what it sounds like to customers. Anyone copywriting can tell you, and even ChatGPT will do a pretty good job of telling you what it thinks is a good message. But the truth is none of them know.”
Humans Must Work with AI for the Best Results
The key to getting the most out of generative AI is maintaining consistency and the correct brand voice, according to Pye. This is why testing is paramount — retailers need to put in the work to ensure that the content being produced, or enhanced, with generative AI properly reflects their core values and traditional messages. In Home Chef’s case, the company ran 71 experiments during the first part of 2023 to help perfect its usage of generative AI.
“We were really excited to see what Phrasee could generate,” said Natalie Mesgleski, Lifecycle Marketing Associate at Home Chef during the session. “Can AI really come up with a good idea? It was really insightful, because Phrasee came up with [messages] that our copywriters would never even think of.”
The results have spoken for themselves. In addition to the aforementioned benefits, Home Chef saw nearly 4.5 million incremental opens and 52,500 incremental clicks. The software also can drive greater personalization, including the addition of contextually relevant emojis, by connecting generative AI to customer profiles.
Human copywriters haven’t been left behind. They are still the ones generating and perfecting the AI-created messaging, and Home Chef’s team has even started experimenting with creating Phrasee-style subject lines themselves. Getting the most out of generative AI requires people and tech working in tandem to find the ways it best fits the needs of the company.
“Our paid media team has already started testing Phrasee as far as Google ads and Facebook ads, and we’re interested in incorporating Phrasee when we interact with our SMS audience,” said Lauren MacArtney, Senior Lifecycle Marketing Manager at Home Chef during the session. “We’re definitely eager to figure out the correct way to go about it.”
Modern Personalization Solutions Still Require Precaution
Despite its potential, generative AI isn’t a silver bullet marketing solution. Pye warned about overreliance on automation in particular and noted that it’s important to retain the human element as Home Chef is doing. Generative AI is very good at creating targeted messaging, but it lacks a human’s ability to quickly parse whether a message will sound genuine or come across as cold.
“I’m a big believer that our implementation of AI is humans using AI, not AI replacing what we do,” said Pye. “It’s important for humans to work alongside AI. Just because we have a system that can respond in a very polite manner doesn’t mean that it can replace the empathy that creates [customer] connections.
Another key limitation is that the data needed to power the tool’s potential is subject to the same precautions necessary when dealing with all modern data-driven technology. While AI is capable of processing a near-limitless amount of data in pursuit of greater personalization, retailers need to make sure they aren’t feeding in unnecessary, or even potentially identity-compromising, information.
“I think the risk areas that many of us are conscious of is that data for the sake of data still has to be disciplined,” said Lindsey Nelson, Retail Industry Advisor at Salesforce during the session. “We’re all accustomed to HIPAA and how data is supposed to be protected. As you’re exploring these tools, make sure that there’s still guardrails around what should be used. Just because you have access to a ton of data doesn’t necessarily mean you have to use all of it.”