Sitecore Stream and the Future of AI-Driven Marketing

During Sitecore Symposium 2024, Sitecore unveiled Sitecore Stream, a brand-aware AI layer designed to empower marketers through enhanced workflows supported by integrated "co-pilots." Set to launch in the near future, Stream represents Sitecore's vision of what an Intelligent DXP could be.

This vision will take time to fully realize. Current demos show a new application available in the Sitecore cloud portal, where you can upload your core brand assets, which will serve as a foundation for training the AI. Marketers can configure other settings to customize the AI to their brand, and another key feature of Stream is its use of generative AI to create and track tasks across Sitecore's digital experience platform (DXP) portfolio, allowing for collaborative planning and project management within marketing teams.

Today, some of these features still have limitations. For instance, tasks within the platform aren’t yet "smart" enough to track completion autonomously; the AI doesn’t know when you’ve created an image in Content Hub or published an article in XM Cloud unless you manually mark it as complete.

Similarly, while the component generator co-pilot can help create React components (similar to how it works in Vercel's v0 platform), it doesn’t yet adhere to brand guidelines or integrate directly into your code base. But these are early days, and as with any evolving technology, these initial limitations are likely to improve over time. What’s more exciting is Sitecore’s vision for Stream’s future. Even in its current marketing, Sitecore acknowledges that this is just the beginning. On their Intelligent DXP landing page, they lay out an ambitious roadmap:

Source: Sitecore.com

The current "CoPilot" functionality will advance to "Advisor" and, ultimately, "Autopilot," where AI may one day take an even more autonomous role in campaign execution.

Speculating on the future

As I consider the direction Sitecore is taking with Stream, I find myself reflecting on my own experiences using AI tools like ChatGPT. ChatGPT has developed an impressive ability to remember details about my preferences and adapt to my needs over multiple sessions. This ability to “learn” from past interactions sparked ideas for how Sitecore Stream might evolve in similar ways.

Taking inspiration from these experiences, I can imagine a future where Sitecore Stream not only enforces brand standards across an organization but also allows for more nuanced settings, potentially layering in standards for regional or product-specific teams, or even individual marketers. Could Stream one day “remember” the preferences of specific marketers, adjusting its recommendations or outputs based on how each user typically interacts with the AI?

Will it be able to layer in regional and multi-brand standards just as effectively as the organizational standards? I've worked on many multi-site Sitecore implementations, so I imagine support for multi-brand will be essential for many customers.

By creating a framework that balances organization-level, regional and site-specific needs with individual marketer preferences, Sitecore Stream might empower marketers to maintain consistency while providing room for unique creativity. This could help individualize workflows, supporting marketers in achieving their own style while staying true to the brand’s core messaging.

Managing existing assets with Sitecore Stream

If Sitecore Stream is to operate effectively at multiple levels, one critical question is how it will be trained on brand assets. In today’s demos, Sitecore Stream includes a centralized location where assets are manually uploaded to train the AI. But an organization’s assets are already managed in Sitecore's products, whether it's XM Cloud, Content Hub DAM or Content Hub CMP.

One way to make this process more effective is by using a structured taxonomy alongside Retrieval-Augmented Generation (RAG) techniques. A layered taxonomy could categorize assets based on their relevance to organizational, regional, product-specific, or even individual marketing needs. For instance, core brand assets representing organization-wide standards would carry a higher weight in guiding the AI, while regional or product-specific assets might be given secondary priority.

However, for this to work seamlessly, an automated indexing process would be necessary. Ideally, this process would create embeddings—vectorized representations of assets based on their content and associated taxonomy tags. This would make the assets more relevant and easily discoverable by the AI when it generates responses or recommendations.

Such indexing doesn’t currently exist in Sitecore Stream and would need to be developed as part of Stream’s functionality or potentially integrated through Sitecore Search to support Stream. Embeddings would allow the AI to “understand” which assets align most closely with different brand standards or campaign contexts, enhancing its ability to deliver consistent yet flexible outputs.

With this setup, when the AI generates content or assists with marketing tasks, it could pull from assets tagged according to this taxonomy. For example, if a marketer on a regional team is creating content for a product launch, the AI could prioritize assets tagged with that specific region and product category. This would ensure that the AI’s suggestions are not only consistent with brand standards but also tailored to the unique context of the marketing initiative.

This approach allows the AI to balance consistency with personalization. Organization-wide standards would guide the AI by default, but regional, product-specific, or individual preferences could dynamically influence the outputs where appropriate. In short, using a taxonomy-guided RAG approach would enable Sitecore Stream to access a nuanced yet consistent asset library, helping it maintain brand integrity while adapting to real-world complexities in various marketing contexts.

Navigating complexity and edge cases

As exciting as these possibilities are, they introduce layers of complexity. Moving from CoPilot to Autopilot would mean addressing countless edge cases that arise in real-world marketing.

Marketing, after all, isn’t purely mechanical – it requires balancing many factors that vary across platforms, demographics, and campaign goals. If Sitecore Stream were to eventually reach full “Autopilot” functionality, it would need the capability to understand these differences and respond with nuance.

Autonomous content creation might work well for routine, well-defined tasks, but handling the nuanced decisions marketers make every day – such as adjusting tone or experimenting with format – would be a far greater challenge.

As Stream evolves, I imagine it could benefit from a combination of supervised learning, advanced tagging, and feedback loops to effectively interpret intent and produce brand-aligned content. Incorporating flexibility to adjust based on feedback might be essential for Stream to handle the “messiness” of real-world marketing.

Wrapping up

My thoughts here are purely speculative and reflect my own experience and perspectives on generative AI. However, as I look to the future, the possibilities for an AI that can balance brand consistency with individual creativity are exciting. If Sitecore Stream achieves this vision, it could redefine the relationship between brand, marketer, and AI in transformative ways.

If you're curious to learn more about where Sitecore Stream is going or need help exploring how AI can transform your marketing processes and improve customer experiences, feel free to fill our contact form.