Magic Message helps marketers write and send high-performing text messages. Here's what we learned from building it—and how we're thinking about the future of AI in marketing.
This post is part of our series exploring how Attentive is building AI and how marketers can use this technology to thrive in a new world. Missed our first post? Check it out here.
AI is more than just a hot topic. For marketers, it’s quickly becoming a valuable—and necessary—tool to not only increase productivity and efficiency, but also to drive better results from the programs they’re running.
If you haven’t already used it to come up with ideas for a campaign or a piece of content, chances are you’ve at least experimented with different tools on the market to see what’s possible.
At the same time, businesses are investing more time and resources into building products specifically designed to help marketers take advantage of AI. Case in point: We recently introduced our own suite of AI tools here at Attentive, including Magic Message, which generates high-performing copy that fits your brand voice and marketing needs. (You can also use it to customize images for SMS and email—but more on that later in the series.)
We sat down with our Chief Product Officer, Nakul Narayan, to take you behind the curtain and share insight into the process of building the copy assistant, how our customers are using the tool, and more.
How did we approach creating our AI-driven copy assistant for SMS?
We know that one of the most time-consuming parts of creating and sending SMS campaigns is coming up with the right copy. There are so many different factors at play. You have to consider: Is it relevant? Is it pithy? Is it on brand?
We wanted to make this process easier for marketers who own SMS for their brands—the ones who are spending many hours writing and sending text messages every day.
The first step was for us to figure out what constitutes a "good" or effective message on a brand-by-brand basis. We looked at all the data we had—from about 40 billion messages sent on our platform—and "deconstructed" them to identify the most important components.
If you think about the average SMS campaign, it includes some variation of a greeting, call to action, offer, or product mention. It also has a certain tone of voice, length, or theme. We looked at all of these attributes, and then trained our AI to write good quality copy based on them.
From there, we asked ourselves if we could train the AI to write messages "from" a specific brand to bring that extra level of customization to the copy assistant. We incorporated the best performing campaigns sent by individual brands, as well as brands' unique tone of voice and standards for things like punctuation, emoji usage, and capitalization.
The other piece is that marketers often adapt their copywriting style as campaign themes and objectives change. We recognized that we could make their lives significantly easier with a tool that gets them 90-95% of the way there—as long as we also gave them the flexibility to go in and sprinkle that last 5-10% of magic to make a message really work for them.
Digging more into the technical side—how does the copy assistant actually generate engaging text messages? How do we know that the copy it’s producing is “good”?
We use a range of large language models (LLMs) under the hood. These LLMs understand natural language because they’re trained on massive amounts of text. Our focus was to take these models and transform them to fit the marketer's specific use case of creating high quality SMS copy.
The challenge was figuring out how to identify what high-performing messages look like across different brands and verticals, so we could train the copy assistant to replicate them. This isn't as easy as it seems, because, as any marketer knows, there are many extraneous factors that contribute to performance, like discount codes.
To account for this, we looked at messages across brands in similar industries, evaluated what made them perform well, and applied the brand's unique tone and style.
The model can recommend copy based on these factors, but, ultimately, the performance of each message depends on the offer or the occasion. That's why we want the marketer to have the final say in deciding which messages to send and be able to A/B test to see what works best with their particular audience.
How does our tool compare to other AI products on the market in terms of accuracy and effectiveness?
It's hard to make direct comparisons, but we have two notable advantages. The first is that our assistant is trained on 40 billion text messages, and there's inherent accuracy with having such a large sample set.
The second advantage is that the assistant is constantly learning. As our customers use it to send more messages, it's learning—and adapting—based on how those messages perform. That level of effectiveness is really hard to replicate, just given the sheer volume of customers we have using our platform.
We're already seeing how this tool is making marketers’ jobs easier—and faster. It essentially cuts their copywriting time in half. We're able to measure that success by looking at how often users are accepting our AI-based recommendations—which they're doing at an incredibly high rate. And, when users do accept the recommendations, we're also looking at how often they send those messages with very few edits.
What excites you most about the AI tools we're building and the impact they could have for marketers? How do you see these tools evolving in the future?
What excites me most is that we're solving a hard problem for marketers who are dealing with fundamental challenges in the economy and for teams that are stretched really thin. It's energizing to make it easy for our customers to get the most out of SMS. That's our true north. We want to give marketers the superpowers they need to be successful with conversational commerce.
In terms of where we're headed, I think we're just scratching the surface with AI. Today, a lot of what we're focused on is creating copy, which is the atomic unit of any marketing strategy. But, there's a lot of opportunity for us to think about other aspects of campaign creation. AI can help with identifying the right audience, the right timing, and the right content.
Eventually, AI can also support marketers with their strategy, like recommending the optimal set of messages to achieve certain revenue and ROI goals.
How have you seen customers successfully using the copy assistant so far? How is it changing their workflows?
We already have customers who view the copy assistant almost like a teammate they can brainstorm with. Maybe they have a rough idea for a campaign, and then the copy assistant generates five others, and they're able to use that output to keep riffing.
With international customers, we’ve seen the copy assistant bridge language gaps, making it a great tool for folks who aren’t native speakers.
There's also definitely been a consistent pattern where, once someone uses the copy assistant two or three times, it becomes a regular part of their workflow. They start using it to support every campaign they send out. It gets better and better each time and picks up on the strategy that they're trying to execute.
What are your thoughts on the idea that AI will end up replacing a lot of what people do today? Do you think there will be a point where it doesn't require any human oversight?
Whenever there are major technology shifts, it's natural to think about the future of work and the possible displacement of jobs. While we understand these concerns, we see an exciting future ahead for marketers.
If we take a step back and think about what using marketing tools looked like 15 years ago, it was incredibly manual. For example, creating segments was nearly impossible because you didn't have enough data. You couldn't build journeys easily because you didn't have integrations. Now, it's so much easier, and you're able to do a lot more with less.
I think we'll see something similar happen with AI. Right now, for instance, it's every marketer's dream to have one-on-one conversations with their subscribers—but that's not literally possible. Our goal is to get to a point where we can learn enough to help the marketer to go from sending the same message to thousands of people to creating thousands of different variants based on each subscriber's individual patterns.
So, really, we’re thinking about what the role of a marketer looks like in the future. Every one of our roles looks different now than it did 20, 10, or even five years ago, and my guess is that trend will continue.
Want to learn more about what you can do with Attentive AI? Join the waitlist.