Assistant - Sitecore Stream

Description

Assistant, a key feature within Sitecore Stream, is an AI-driven chat experience that enables users to engage with their brand assets and guidelines through RAG (Retrieval-Augmented Generation) and agentic workflows.

As a Product Designer on the Stream team, I was responsible for designing the UI and user experience provided by the Assistant. I collaborated closely with Sitecore's Chief of AI, the team's product manager and front-end developers.

Duration: 2024-2025

Version Shown: Pre-Early Access

Sitecore Stream Assistant interface showing AI-powered content creation
Research

To validate market needs for Sitecore Stream, we conducted workshops and interviews with enterprise clients, including Nestlé, Starbucks at Home, and Unilever. These sessions provided valuable insights into how marketers ideate, develop briefs, and initiate early campaign planning.

One key deliverable was a journey map I created, illustrating how a brand manager transitions from briefing to campaign execution. This artifact became a reference not only for Assistant's design but also for broader Sitecore product development and is now a shared resource within our design team.


Additionally, I maintained a daily competitor analysis focused on AI chat experiences, tracking UI and UX advancements across the industry to ensure our team stayed informed of emerging trends.

Journey map showing brand manager's transition from briefing to campaign execution
Problem

Marketers often struggle to locate brand assets and guidelines due to:

  • Knowledge loss when team members leave,
  • Organisational restructuring,
  • Lack of structured documentation, and
  • There being extremely large quantities of assets

For large enterprises, this challenge is even more pronounced, making it difficult to maintain consistency across campaigns.

Solution

Sitecore Stream Assistant solves marketers struggle to find brand assets by affording the ability for them to chat with the brand assets through a chat interface. Marketers can select the brand from which they wish to query, and our AI will retrieve the necessary response regarding brand information, assets or campaign knowledge.

UI Breakdown

Chat Interface

For the empty state of the Assistant, we designed a layout that prioritises understandability while keeping secondary actions unobtrusive.

The primary goal of the page is focussed on prompting. To make this action intuitive, we implemented a prominent, two-line-high prompt field outlined in our AI brand colours, making it instantly recognisable when scanning the page.

Users can create a new chat, browse their history, and switch between brands, all accessible via a compact buttons in the top left corner, ensuring these actions remain available without cluttering the interface.

To guide users and enhance directionality, we placed quick-start prompts at the centre of the screen, offering helpful suggestions to initiate a conversation.

Empty state of the assistant chat interface

Brand Kit Selection

A drop down was designed to afford organisations to select which brand they would like to chat with, as for large scale organisations, they may have tens if not hundreds of brands all with differing assets and guidelines.

When the user loads into the chat, their most recently conversed with brand is auto-selected to avoid any false starts of trying to type a message and not having a brand kit selected.

Drop down of user selecting their brand kit

Asset & Guidelines Retrieval

When a user requests information about their brand, the retrieval agent pulls relevant details from the Brand Kit's organisation-defined assets and guidelines.

The agent follows a chain of thought, which is accessible within the chat. Users can expand this summary at any time to see the full reasoning and trace the exact sources of the retrieved information. Supporting sources, including images and documents, are displayed both in the chat and within a dedicated side panel.

Since retrieving information is the primary action in this workflow, suggested responses are collapsed to minimize distractions and keep the focus on the user's intent. However, in workflows centred around creation, such as generating assets, suggested responses and follow-ups remain visible to encourage iteration and refinement.

Response from Assistant containing brand assets

Artifacts

When a user intends to create something, such as a document, the chat generates an artifact: a rich-text editable container that holds the response. This artifact can be refined in isolation, prompted upon, and easily exported.

To maintain the chat aspect of the experience, a summary of the generated document appears within the chat, along with an option to expand or collapse the artifact, preventing the interface from being permanently split.

Artifacts enable us to scale chat functionality beyond text responses, supporting diverse user intents like brief generation and image creation. This approach allows us to prioritise the user journey over technical or design constraints.

Artifact created within the Assistant
Conclusion

Sitecore Stream, including the Assistant feature, was officially unveiled at Symposium 2024, where it was announced as part of an exciting partnership with Nestlé and Microsoft. Following an overwhelmingly positive response from early access partners, the product is now entering general availability for Sitecore customers as of March 2025.

Since its initial debut within EA, Sitecore Stream has undergone significant enhancements based on customer feedback, with many improvements already implemented and an ambitious roadmap ahead. Looking forward, we are investing heavily in advancing our Agentic processes’ user experience to deliver even more predictable, enjoyable and usable marketer workflows.