Agentforce helped 7,445 people create the perfect Dreamforce schedule.

Action-packed Dreamforce leads to information overload for attendees and staff.

Dreamforce is the biggest enterprise technology event of the year. With more than 40,000 attendees exploring how to grow their businesses and careers across 1,500+ sessions over three days, there’s a lot to choose from. Finding ‌the right people to connect with and sessions to attend was tricky for attendees because event details were dispersed across PDFs, websites, pre-event communications, and the Dreamforce event app.

Attendee feedback from previous years shared that although agenda recommendations from a logic-driven quiz feature in the Dreamforce event app were helpful, finding answers to common questions was still difficult. The quiz could recommend sessions based on each user’s interests, but it couldn’t help attendees figure out where to pick up their badges or how to connect to the venue’s Wi-Fi.

The variety of places to look for event details was a hurdle for event staff as well. Like the attendees, they had to sort through multiple documents — many of which couldn’t be updated with last-minute changes — to find the information they needed. The lack of a centralized, easy-to-use system led to delays and inaccuracies that made it difficult for attendees and staff to make the most of their time at Dreamforce. 

Salesforce’s events team saw an opportunity to improve attendee support with a streamlined solution built directly into the event app. This solution would offer quick, reliable access to accurate information and better equip event staff to handle inquiries on the spot.

Agentforce takes the guesswork out of Dreamforce.

The event team quickly realized that Dreamforce was the perfect opportunity to use Agentforce — the new agentic layer on the Salesforce Platform — to build an autonomous AI agent capable of true personalization and context-driven assistance. They knew that a basic bot, which is limited to scripted responses for preprogrammed questions, wasn’t capable of handling attendee needs considering Dreamforce’s rich variety of sessions and experiences.

To ease information overload for attendees of Dreamforce 2024, the Salesforce team built Ask Astro, an autonomous agent integrated into the Dreamforce app. It’s designed to take the guesswork out of where to go and what to do at the event. When attendees ask questions in natural language, Ask Astro provides accurate, up-to-date information about the venue, speakers, and sessions. It can even build agendas with personalized recommendations for attendees so they can make the most of their time at Dreamforce.

“That app saved my life,” an attendee at Dreamforce 2024 said. “I don't think I would have been able to make it without it.”

Service agents deployed in just two weeks with Agentforce.

With the Salesforce Platform, the team launched the first version of Ask Astro in just two weeks — a whopping 190% faster than if they’d used OpenAI on Microsoft Azure. The simple, low-code tools in Agent Builder made it easy to build, configure, test, and refine the agent so it was ready in time for Dreamforce.

The engineers outlined what they wanted Ask Astro to do, used natural language instructions to guide it to the right information, and created a library of actions for it to pull from. Then, they moved on to testing and debugging to make sure the agent was functioning properly.

Testing Ask Astro’s responses and fixing issues like off-topic labels and time zone conversions was quick and easy. With Agent Builder, debugging cycles dropped from weeks to just 30 seconds on average. By the time they finished testing, Ask Astro’s accuracy in retrieving the right information had jumped from 80% to nearly 100%.

To personalize responses, Ask Astro needed access to real-time attendee and Salesforce data. With Data Cloud and Salesforce AI, Ask Astro could use each user’s role, industry, preferences, and even past behavior to deliver personalized, relevant suggestions.

Because Ask Astro had access to customer data, it was critical to set controls to keep that data private. The team used the Trust Layer features to ensure Ask Astro never shared sensitive information and was built securely from start to finish.

Data Cloud and MuleSoft ensured accurate answers to 23,171 questions.

Access to real-time information about attendees, staff, schedules, speakers, and venues was crucial to Ask Astro’s success. To do this, it needed to be connected to both Data Cloud and Rainfocus, the event management platform that was used to support Dreamforce. With MuleSoft Anypoint Studio, this integration was accomplished in just 45 minutes.

With this information and Salesforce AI, Ask Astro could give attendees and staff quick and accurate help. People could ask questions like, “When is P!nk performing?”, “How do I access the Wi-Fi?”, “How can I meet others in my industry?”, and “Tell me about the Lodge Village,” and get answers tailored to their industry and interests.

Andrew Russo, Salesforce Architect at BACA Systems, said, "The ability to launch a fully functional agent in just weeks is a total game-changer. Using Ask Astro at Dreamforce helped me manage my time and discover valuable sessions I might have otherwise missed."

Besides answering questions, Ask Astro also took action to help attendees manage their event schedules, adding or adjusting sessions based on their goals. With an average of three conversations per attendee for 7,445 attendees, Ask Astro took care of simple requests so that event staff had time for more complex tasks like handling badge replacements and production logistics. This improved the experience for attendees and staff and contributed to a 9-point increase in the net promoter score year over year.

Data Cloud Hybrid Search and built-in RAG drove Ask Astro’s accuracy.

Finding the right information required Ask Astro to understand each question and match it to relevant details from a combination of data sources. Event information like session, sponsor, and speaker details was stored in Rainfocus, while a staff-created FAQ document provided additional specifics such as networking opportunities, refreshment options, and on-site activities.To provide useful answers to attendees, Ask Astro needed to check both of these sources before replying. 

Salesforce engineers used Data Cloud Hybrid Search and Agentforce’s built-in retrieval-augmented generation (RAG) to make this happen.

Data Cloud Hybrid Search combines keyword and vector search. Keyword search looks for specific matches, and vector search extracts the meaning of concepts. When used together, they make it possible for Ask Astro to understand what attendees are asking for — even if their request has a few typos.

RAG allows Ask Astro to access up-to-date, relevant information in Data Cloud, even though it uses a foundation model that wasn’t specifically trained on Dreamforce content. Rather than relying solely on the model's pre-existing knowledge, RAG pulls in fresh data from sources like session details, sponsor info, and FAQ documents in real time. This way, Ask Astro can respond accurately and contextually without needing custom training. 

By using Data Cloud Hybrid Search and RAG, Ask Astro provided 23,171 timely and accurate answers over three days at Dreamforce.

With Agentforce, Salesforce deployed an autonomous and accurate service agent in just two weeks — a process that would’ve taken months with OpenAI on Microsoft Azure. The Salesforce Platform brought real-time attendee, event, and venue data together so that Ask Astro could provide personalized, up-to-date information regardless of last-minute changes. Not only did Ask Astro help attendees find the information they needed, it saved them time by modifying their schedules based on their interests and goals. This created great experiences for attendees, freed up event staff to help with complex requests, and contributed to a 9-point improvement in Dreamforce’s NPS.