Over the past year, the content creation capabilities of generative artificial intelligence have captured most of the liveliest headline news: how AI models can instantly create Pixar-like short films starring your pet Chihuahua, or produce a Drake-Kendrick Lamar diss track that you never thought you would enjoy. The commercial use of GenAI content creation has received less attention, but it is equally (if not more) valuable: marketing.
Marketing is particularly well-suited to adopt generative AI because it is an iterative, creative, and dynamic practice that relies on the types of media that drive the development of Large Language Models (LLMs) — text, images, video. (This is one of the reasons why many of the first B2B GenAI use cases are in marketing!) In addition, successful marketing plans and assets do not necessarily have a single "correct" solution. This makes marketing different from fintech, where users expect a correct answer to their queries and have higher demands for accuracy. Also, unlike sales, marketing does not rely on carefully establishing interpersonal relationships with customers (although this may become easier in a world of AI-driven sales development representatives).
At the same time, the way companies view marketing is also changing. As people's time becomes more fragmented, customers are increasingly difficult to reach. Marketers are looking for scalable ways to create personalized campaigns and messages to meet people's needs. However, marketing teams are often isolated, working with different tools that often do not work together. GenAI-based software can and has been helping to bridge this gap.
We have already begun to see the financial impact of this shift. According to a recent report by McKinsey, GenAI could bring about $3.3 trillion in global productivity annually in the fields of marketing and sales, and global payment company Klarna recently stated that by using GenAI to generate images and reduce reliance on external marketing partners, it could save $10 million in costs annually. But this evolution has only just begun.
We believe that the application of GenAI in marketing has gone through three distinct stages: the development of marketing co-pilots, the introduction of marketing agents, and the rise of autonomous marketing teams. We will break down each stage below:
Developing Marketing Co-Pilots
The first stage of AI marketing evolution (which we are currently in) is the use of GenAI by marketers as a marketing assistant. With the help of artificial intelligence, marketers no longer need to spend time crafting generic email and press release copy, nor do they have to hire copywriters to write blog posts that are good for SEO; instead, they can outsource the first drafts to ChatGPT and other tools, then spend their time on higher-level tasks. For example, platforms like Jasper or Copy.ai can scale social media posts and sales emails within seconds, while products like HeyGen and Synthesia can help marketing teams create and edit studio-quality videos (e.g., through captions) within minutes.
Today's marketers need to carefully design and continuously refine their prompts, while future marketers will be able to create brand-consistent assets by obtaining signals from existing, increasingly important first-party data (such as customer data platforms, websites, style guides) as well as second and third-party sources like pixels, UTM codes, lookalike audiences, recommendations, etc. There is still a need for manual content quality control (e.g., approving the similarity of character images, verifying that images fit the brand). However, the marketing co-pilot is rapidly improving and will continue to learn over time based on the inputs and style of the marketers.
The marketing co-pilot can browse large amounts of data from multiple customer data platforms and use all data in one place (especially in the context of the ever-increasing number of marketing channels), and can also help with non-content creation tasks, such as audience segmentation and planning. The data loop also happens in real-time, allowing for real-time iteration. For example, Validated runs hundreds of digital ads to help companies identify content that customers are particularly interested in. Similarly, Outset and Voicepanel allow research teams to deploy a large number of proxy researchers to help define new customer segments and test new concepts. The development of research and segmentation co-pilots is crucial for the development of AI marketing, as building tools that focus solely on content creation is not enough to ensure continuous adoption.
A more successful method of pushing products from consumers to businesses is to use content as an opportunity, then enter the workflow, using tools to assist with customer data analysis, segmentation and planning, brand resonance, etc. This is because workflows - tracking projects, collaborating and iterating; coordinating the brand's voice in content; providing performance metrics, etc. - make the product sticky. For example, Jasper started with copywriting and now enables marketing teams to collaborate and gather insights. For example, since video and 3D animation are exciting newer vertical industries that lack existing businesses, we also hope to see many new companies emerging in these vertical industries as well as others.
Building Marketing Agents
The next stage after equipping marketers with co-pilot tools is to use marketing agents to automate the work of marketers. At this stage, we expect marketing to shift from a one-to-many model to a one-to-one hyper-personalized activity. Marketers will not need to create advertising campaigns that attract average customers, but will be able to personalize each advertisement displayed to customers based on specific audience and preference data, which is more effective than sharing generic content with a large audience.
We are just beginning to see some companies use AI agents to complete narrow end-to-end (e2e) marketing tasks: A/B testing specific campaign assets, optimizing ad bidding and purchasing, tracking attribution and analysis, iterating content based on results, and then making creative decisions (rather than just providing content and insights for the team). These agents integrate with performance data (such as CTR) and content creation tools, and use judgment to try new content variants to drive results. They may also help collect market research and competitive intelligence and work across ecosystems such as social media and connected TV.
For example, email automation has been around for a while, mainly centered around templates and workflows around scheduling and tracking emails. But in the future, the "email marketing agent" will be able to automatically generate content, personalize content, set sending schedules, monitor open rates, and adjust content based on performance metrics - based on information obtained from product pages, audience demographics, campaign rhythms, and parameters selected from connected customer data. Coframe has already done this for copy and images on marketing websites. GenAI marketing agents will still take direction from marketing teams on campaign goals and how much to spend, but they will manage autonomously. Agents will complement individual roles, allowing marketers to focus on more strategic activities, goals, and metrics.
AI agents will also allow marketers to conduct more frequent experiments because agents can suggest changes, actually make changes (after approval from marketers), and then analyze the results. This positive shift towards one-to-one marketing will lead to the integration of sales and marketing. As our colleague a16z General Partner Andrew Chen said, we only do marketing because the cost of 1:1 sales for everything is too high. The challenge that needs to be overcome here is to solve the ancient attribution problem, for which the decision-making of users must be correctly inferred so that AI can make wise decisions.
Transitioning to an Automated Marketing Team
The final stage and ultimate goal of the evolution of marketing AI is to allow AI agents to fully take on the responsibilities of the CMO and operate as their own autonomous marketing team. At this stage, a large number of AI agents will recreate or supplement the team's full-service capabilities. By pooling and optimizing all single-purpose agents in all media, agents will be able to formulate strategies and assets for mature marketing plans.
All a company needs to do is input the budget and goals. Then, the software will collect analysis and performance across channels, implement an omnichannel strategy, and use a set of marketing agents to run everything from market research to performance marketing and brand campaigns.
How to adjust the titles of blog posts to optimize for search? How to iterate videos to perform best on TikTok and Instagram? With more integration, existing assets can also help generate other types of content. For example, blog posts can provide the basis for generating landing pages, SDR sequences, and copy for radio or television advertisements. Brand and performance marketing can now work more closely together, and both are more data-driven.
Companies created at this stage may start with verticals such as video or email and continue to evolve over time. Automatic video creators can learn how to optimize performance and become agents. Agents are likely to start expanding into a second vertical. For example, Klaviyo started with email and then added SMS communication because it is an extension of similar messaging and customer expansion. Whether through purchase or construction, you will eventually get platforms like Adobe or Salesforce Marketing Cloud, and companies will tend to use these platforms and get a suite of marketing tools.
Differentiation may also occur between B2B and B2C platforms due to the different functionalities that need to be optimized (i.e., lead generation vs. product sales). Complexity and user experience determine that small and medium-sized enterprises and large enterprises need different products. In many ways, small and medium-sized enterprises have gained a greater advantage because they now suddenly have the power of a complete marketing team - in the past, they may have only had one marketer (or more likely, an owner who only spent a small part of their time on marketing). We may see small and medium-sized enterprises offering a set of more basic content creation products, even at the co-pilot stage, because they do not have the time or resources to manage a variety of separate tools.