The adoption of LLMs and software based on natural language prompts has made media buying as easy as, well, a conversation.
For companies in the advertising world, this means the introduction of chatbots or agentic AI to every step of the media-buying process, from understanding customer data to planning campaigns and generating the creative.
My Marketing Pro (MMP), a marketing automation startup founded earlier this year, takes it one step further. The platform aims to automate the entire marketing process, from brainstorm to launch, via conversations with an agent.
Ask and you shall receive
MMP has a single agentic AI persona, which interfaces with clients via natural language prompts to generate marketing plans. The brand customers have text or video call discussions with the agent, which is called Memory (after a close friend of Founder and CEO Frank O’Brien’s, who is apparently cool with her name and likeness being used as a media planner robot).
The user “calls” Memory, which pops up on the screen and has a face-to-face conversation with the human ad buyer about their brand’s goals and core messaging. (For the camera-shy among us, users can opt for a text conversation instead, though only audio is actually recorded.)
MMP follows up by email with a brief for a full campaign plan, including anticipated revenue and conversions and suggested creative and copy, plus a timeline breakdown.
The actual process from idea to execution is almost immediate, occurring within about 20 minutes. After Memory processes the advertiser’s goals and constraints, it pulls relevant videos and images from the web and creates a draft of the creative assets.
Once the advertiser has signed off on the brief, Memory will begin the campaign; it will also handle optimization and fine-tuning of the targeting while the campaign is running.
For instance, said O’Brien, an advertiser can ask Memory where to find the highest population of people who might be interested in a product but haven’t used it already. Memory would identify specific regions and suggest the best ways to target people in those areas – as well as what tactics to avoid. (For a town without any billboards, for example, a brand might want to steer clear of out-of-home and instead focus on geo-targeting in CTV ads.)
Call me maybe
If MMP provides the agent, what are its LLMs of choice?
It turns out the platform is built almost exclusively on ChatGPT. MMP prefers to rely primarily on one LLM since all of the main LLMs “become very similar” at a certain point, according to O’Brien, “especially once you flavor them with your own prompts.”
Still, MMP’s model has evolved quite a bit since its early days.
The startup was founded as an outbound calling app meant to function as a salesperson, O’Brien said. Memory would call prospecting lists to find people in need of marketing services.
Now it reaches audiences via APIs to target people on platforms like Meta and X, as well as other channels including CTV and digital out-of-home. MMP also works with a data partner called Watt, which creates lookalike audiences and provides ways to reach them, including email addresses and phone numbers.
The agentic automation product isn’t going to run wild on assignment, though. Memory knows the scope of her work.
In the spirit of thorough investigation, this reporter was curious as to what Memory would say if prompted to bark like a dog. But Memory didn’t take the bait.
“I’m here to help with marketing tasks,” it replied, “not barking.”
