Data is essential: Building an effective generative AI marketing strategy

Generative AI is powering a new world of creative, customized communications, allowing marketing teams to deliver greater personalization at scale and meet today’s high customer expectations. The potential of this powerful new tool spans the entire end-to-end marketing process, from internal communications and productivity to customer-facing channels and product support. In a May 2023 survey carried out by IBM and Momentive.ai (link resides outside of ibm.com), 67% of CMOs reported that they plan on implementing generative AI in the next 12 months and 86% plan to do so within 24 months.

AI for business has long been able to achieve multiple marketing functions: seamless connection across any platform or device, immediate response when experiencing issues and customization based on current location and purchase history. But generative AI solutions can provide new capabilities for customer-facing teams in marketing to provide even greater personalization at scale and improve employee skills and performance.

Enterprise marketing teams stand to benefit greatly from generative AI, yet introduction of this capability will require new skills and processes. According to the IBM survey, when CMOs were asked what they thought the primary challenges were in adopting generative AI, they listed three top concerns: managing the complexity of implementation, building the data set and brand and intellectual property (IP) risk.

With the right generative AI strategy, marketers can mitigate these concerns. The journey starts with sound data.

As with all AI implementations, generative AI requires attention to sourcing and maintaining the underlying data. The familiar IT adage, “garbage in, garbage out,” still applies; high-quality data is essential to yield a high-quality result. If the training data is biased or incomplete, the models may generate inaccurate content.

For marketing in particular, generative AI can help with content development and audience targeting. Data curation is key, along with setting guardrails and supervision to address bias and ensure consistency of brand voice and accuracy of product and service information.

For example, a retail clothing company might use generative AI to customize email or online experiences tailored for different customer personas. The advanced capabilities of generative AI for text, visuals and video have the potential to create a more personalized and engaging experience. This might include a virtual model wearing outfits that match the customer’s body type, fashion choices and activities of interest. The generative AI tool can also incorporate external factors like weather, upcoming events or the shopper’s location.

But what if the generative AI tool recommends the customer buy a bathing suit in the middle of winter or a snow parka in the summer? Because various generative AI solutions are trained on large swaths of data, they have the capability to pull and interpret existing data incorrectly. Thus, the tool has the potential to provide unexpected results.

When an AI foundation model generates off-topic or incorrect content, that behavior is referred to as a hallucination (link resides outside of ibm.com). To mitigate this scenario, teams must ensure they customize their models with proprietary datasets, rather than relying solely on open-source internet data.

Before your marketing organization can introduce effective generative AI solutions, you need a strategy to implement AI foundation models. Given the vast landscape of available data (both external and internal), it’s essential to define your use cases in advance of sourcing and training your models. Understanding the benefit and risk of each use case will help to create a step-by-step path that prioritizes the model training process.

Marketers also need to work closely with IT to align on the data architecture needed to securely build and deploy foundation models while following necessary protections for intellectual property and confidential data. The appropriate usage guardrails will help monitor and safeguard your IP and the integrity of your brand.

Once deployed, your generative AI data journey isn’t over. Foundation models are continually being refined as they interact with customers, collecting increasing amounts of data, which in turn improves their capabilities. Human supervision (such as supervised fine-tuning with human annotations and reinforcement learning from human feedback) is required to align the output of generative AI apps running on foundation models with human intentions, ensuring they are helpful, ethical and reliable.

Even though generative AI can produce customer-facing work that seems humanlike, it still requires a human guide with expertise in navigating ethical and legal concerns regarding data use. Human reviewers can also identify and correct any instances of bias or hallucination that could have seeped into the content.

In the IBM survey, CMOs cited content creation and editing, SEO and social media marketing as the top B2B use cases for generative AI capabilities.¹ In regard to B2B marketing function, these leaders called out lead generation and sales nurturing as the top use cases.¹

When asked about their biggest concerns regarding generative AI, leaders were focused on data accuracy, privacy management and having the skilled resources to build this solution.¹ To that end, adopting generative AI technology requires a practical approach to build, test and learn about its capabilities. This will ensure that proprietary data is protected, customer experiences are relevant and rewarding, and the marketing process is streamlined and cost effective.

For decades, IBM has been at the forefront of AI for business. We provide solutions and services that help marketers implement generative AI responsibly and effectively. The IBM watsonx portfolio of AI products is designed to help marketing and other business leaders confidently move into the generative AI arena. The portfolio includes:

1. watsonx.ai: an enterprise studio for AI builders to train, validate, tune and deploy generative AI
2. watsonx.data: an open hybrid data store built on an open lakehouse architecture, designed to help scale generative AI workloads
3. watsonx.governance: a toolkit that accelerates AI workflows that are built with responsibility, transparency and explainability

IBM Consulting™ and its diverse, global team of more than 20,000 AI experts help marketing organizations quickly and confidently design and scale AI and automation across their business. We work in concert with IBM watsonx technology and an open ecosystem of partners to deliver any AI model, on any cloud, guided by ethics and trust.

Take the first step toward generative AI with the right data sources and architecture to support the access, quality, richness and protection of your brand.