The rise of generative AI (gen AI) has sparked a significant wave of innovation across industries, particularly in marketing. Its ability to produce diverse content types—ranging from blog articles and promotional videos to personalized offers and even predictive insights—has left many marketing professionals evaluating both the possibilities and pitfalls of this transformative technology.
As organizations seek to gain a competitive edge, gen AI is increasingly seen as a valuable asset. According to Salesforce’s latest “State of Marketing” report, which surveyed 5,000 marketers globally, deploying AI solutions ranks as their top initiative. Companies like Vanguard and Unilever have already seen tangible results: a 15% boost in LinkedIn ad conversions and a 90% reduction in customer service response times, respectively, due to gen AI integration.
Despite this enthusiasm, widespread implementation remains limited. Although nearly all marketers in the survey had adopted or planned to adopt gen AI within 18 months, only 32% had integrated it fully into their marketing systems. This hesitancy is understandable—AI deployments can backfire if not handled thoughtfully. For example, Coca-Cola faced backlash when it used AI to reimagine its classic 1995 holiday ad. While initial feedback was positive, viewers later criticized the AI-generated version for lacking emotional depth, highlighting one of gen AI’s common shortcomings: the absence of a human touch.
The core issue isn’t whether to use gen AI—it’s about how to implement it effectively. Unfortunately, many organizations lack a cohesive strategy. Decisions about which gen AI tools to adopt are often made ad hoc, with individual employees experimenting independently and sometimes without oversight.
Conversations with over 20 business leaders revealed that developing a strong gen AI marketing plan requires navigating three central decisions:
- Should the task be handled by gen AI or traditional analytical AI?
- What inputs—custom, general, or hybrid—are needed to produce relevant outputs?
- To what extent should human oversight shape the AI’s final deliverables?
Key Questions to Guide AI Use in Marketing
Before making these decisions, companies must ask themselves:
- What specific goals do we want to achieve with gen AI?
- Are we working with structured (organized) or unstructured (free-form) data?
- What are our resource limitations?
- How much efficiency gain do we expect?
- How fast must we deliver results to end users?
- What’s the risk level if gen AI outputs are inaccurate or flawed?
- How critical are data privacy, security, and brand reputation in our operations?
- Do we need fine-grained control over the AI’s process and content?
- What regulatory and legal risks are we prepared to handle?
- What privacy expectations do our customers have?
This article offers a framework to weigh such trade-offs while planning AI implementations in marketing.
Choosing Between Generative and Analytical AI
The first major decision marketers must tackle is choosing between generative AI and analytical AI, each suited to different purposes.
Analytical AI focuses on using structured data to predict outcomes or classify new information. It’s commonly used to forecast customer behavior—such as which product someone might buy, how much they might spend, or which promotion is likely to convert them. Analytical AI has been a staple in marketing for decades. For example, Kia leveraged IBM Watson to identify influencers aligned with its brand ahead of the 2016 Super Bowl—a tactic still useful today.
In contrast, generative AI is built to create content. It works with unstructured data like images, text, or sound and can generate anything from marketing messages and blog posts to product images or even customer support suggestions. Rather than just forecasting, gen AI can synthesize and communicate information in creative ways.
Both tools have value, and often the most effective approach blends the two. For example, a company might use analytical AI to determine the most likely offer to resonate with a customer and then deploy gen AI to craft a tailored message promoting that offer.
General vs. Customized Inputs
Once gen AI is deemed appropriate, marketers must decide whether to use off-the-shelf models trained on public data or augment them with proprietary company-specific information.
Off-the-shelf or "foundation" models—like those trained on data from Wikipedia, GitHub, and social media—offer broad knowledge and versatility. These models are useful in situations that require wide-ranging, flexible responses, such as customer interactions or content summarization.
However, for more specific marketing needs—like writing on-brand ad copy or crafting nuanced customer replies—these models often fall short. In such cases, using custom data to fine-tune gen AI models, or leveraging retrieval-augmented generation (RAG) techniques, can significantly improve accuracy and relevance. Companies like Colgate-Palmolive and Jasper AI have adopted this approach, embedding proprietary data into prompts or tool configurations to ensure outputs align with brand tone, product specs, and messaging guidelines.
Working with custom inputs also addresses data privacy and IP protection concerns. Firms using in-house or closed-source models can safeguard sensitive business information from exposure to public systems, while minimizing hallucinations and enhancing factual consistency.
Still, these benefits come with costs. Training custom models or setting up secure AI systems requires technical expertise and investment. Businesses must carefully evaluate the balance between precision and practicality when deciding between general and tailored AI inputs.
Defining the Role of Human Oversight
The third crucial decision concerns how much human input is required before AI-generated content is shared with customers. In low-risk scenarios—like summarizing user reviews—it may be acceptable to publish content directly without human review. But in high-stakes contexts, such as legally binding promotional materials, companies typically require significant human review to ensure compliance and brand consistency.
While some firms argue that extensive human editing reduces the time-saving benefits of gen AI, others have learned the hard way that skipping this step can be costly. For example, when a chatbot from Air Canada mistakenly promised a bereavement discount, the airline initially refused to honor the deal—until a court ruled in the customer’s favor. This case highlights the potential legal and reputational risks of unmonitored AI outputs.
Organizations must weigh the pros and cons: human review adds cost and slows down delivery, but it also enhances trust, reduces risk, and protects brand equity.
Building a Practical Framework
To adopt generative AI responsibly and effectively in marketing, companies must develop a flexible yet thoughtful strategy that considers:
- The right type of AI for each use case
- The nature and source of training data
- The appropriate level of human interaction and supervision
By doing so, businesses can unlock gen AI’s full potential—boosting productivity, improving personalization, and staying competitive—without falling into the traps of overhyped technology or undercooked planning.
