Building AI-Generated Events and Tools

Summary

As part of the Event Create team at Eventbrite, I crafted content for AI assistive tools embedded within our classic create flow and our new Auto-Create Event experience. Our product team wanted to integrate artificial intelligence into our products so we could help creators speed through a cumbersome experience.

The Problem

events weren’t getting the visibility they deserved and our event publishes had a low conversion rate due to bad ux.

Eventbrite’s event creation process is cumbersome. When specifically measuring conversion rates between desktop and mobile, desktop publish rate was 74%, compared to only 33% publish rate for our mobile experience. This was due to a lot of usability issues with our mobile experience, which had gotten little TLC.

Data also showed that approximately 18% of monthly events (80k-90k events monthly) lack event type and event category information. As a result, these events are not included in various sections of our marketplace, such as category browsing or filtering by specific categories. In fact, events without these fields typically experience a 30% decrease in event page visibility. 

Our product team saw an opportunity: we could leverage our existing integration with OpenAI to assist creators in quickly and easily creating their events. By using the most basic information about the event such as title, date, location, and general ticket price, we could generate the remaining content for them including event type, category, sub-category, tags, summary, description, image, and ticket details.

How might we make it easier for creators to successfully publish their events and include crucial information like event type, category, and sub-category so that their events could appear in more searches, and thus sell more tickets? And, if we could improve general UX issues, introduce mobile patterns, and create a simple and intuitive creation flow that guides the creator to build attractive events, we could increase both our mobile and desktop publish rates.

The Research

a quick usability audit and research findings from our classic create experience helped us prioritize what to improve.

I tested out Eventbrite’s current event create flow on iOS—and quickly saw some major usability issues. Misaligned input fields, huge spaces that pushed interactive content further down in the hierarchy, and main buttons and tasks that couldn’t even be completed. No wonder our mobile publish rate was only 33%!

User Research findings from our latest improvements to the classic create flow also highlighted areas where creators could use more guidance, education, and usability improvements to get them to successfully publish their event. (If you’re interested in my work on revamping our classic create flow, head over here.) Overall, the research found that:

• Users found value in the content segments, and most believed all segments but summary, agenda, and FAQs were required.

• The tooltips and examples were helpful learning tools for users, but some of the tips felt ambiguous.

• Users wanted content to auto-save as they fill out the segments.

But instead of burdening our creators with writing their own summaries and descriptions and using tips and carousels as a way to educate, we could also simplify the process by providing good content up front to work with.

The Solution

to assist our creators in providing better explanations for their events and a faster experience to increase the publish rate and overall event content quality.

By offering an efficient and streamlined event creation process with guidance on both mobile and desktop, we could reduce the time and effort required by creators to publish events on our platform. This wouldn’t just increase publish rate, but it could also to improve ticket sales and increase creator retention.

To address the identified opportunity and achieve our objective, our team divided the work into two phases:

1. Start small by providing AI-generated opportunities in the event summary and description sections of the classic create flow. We’d measure overall usage of the tools, as well as if the AI feature contributed to increased publish rates before going all in on Phase 2.

2. Develop a new end-to-end experience that allows creators to quickly create events with AI by providing basic information about the event. Leveraging the integration with OpenAI, we could automatically generate additional event content based on the provided details. 

Working on the AI Suggestion Tools

There wasn’t much work that needed to be done on the AI-generated event categories; users would type in their title and our platform would automatically generated the best event category and sub-categories for them.

But we definitely needed a pair of content design eyes on our auto-generated event summary and description tools.

Originally, my PD and PM thought a banner at the top of the screen would do the trick, which would ask users if they wanted to try auto-filling their content.

While a banner would be easier to implement and test, I suggested it was better to upsell our AI tools in context, close to the sections they actually related.

This, I suggested, would help educate our creators on the specific sections we offered in the flow, while also providing greater transparency as to which sections were actually changing.

Since our tool could only write for the event summary and description, we should place the tool in these spots.

The end product for both these AI-generated tools was a few simple inline links with tooltips and a disabled state:

I also set up the foundation for how we talk about AI products at Eventbrite

Suggest, Generate, Preview: user research found that these words instilled confidence in creators, because they believed nothing would be published without their approval.

Providing creators with the option to edit the suggestions or suggest again, were words that I wanted to make sure other content designers and writers at Eventbrite adopted into the AI products they were building, so that we could align terminology and keep call to actions and the way we talked about AI consistent across our platform.

Some principles and language patterns we adopted were:

• Call out where artificial intelligence is used in the product as “AI” to provide more transparency to our users. (ex. Create an event with AI)

• When building new products, do not name the AI as an entity or identity. (ex. Eventbrite AI, not ‘Eventbrite Assistant’ or ‘Elena’)

• Use call to action verbs like suggest, generate, and preview to instill confidence that users can make changes to what the AI creates—which all our AI products should offer.

• Iconography-wise, use magic wands or sparkles, as they’re often used in other products to indicate AI.

Examples elsewhere in the product included our Help Center chatbot (called Eventbrite AI), suggesting email copy in our email campaigns tool, and in consumer search.

Working on the Auto-Create Flow

Developing a few embedded AI tools is easy. But developing a whole end-to-end experience is much harder. You have to consider edge cases, the happy path you want creators to take, and so much more. After getting the green-light to develop the AI Event Create experience, I was very keen on adding education and transparency around how our tools got its data, how a creator’s data was used, and the AI’s current limitations.

I knew what information would be requested from the user to auto-create their event:

• Event Title

• Location & Date

• Ticket Price and Ticket Quantity

I then asked my PM what information the AI would provide:

• Event Type (ie Conference, Concert or Performance, Rally)

• Category & Sub-Category (ie Auto, Boat & Air, Food & Drink, Music)

• Tags

• Summary

• Description

• Main Image

Knowing this, I then used the subtitle as well as each input field’s subtitle to explain how the creator’s information would be used and where that info would show up in the event once generated:

Though I had to shorten the subtitle since the text got too lengthy for our mobile experience, we kept the contextual copy in each input field for guidance.

We also had to think about the entry point on the creator homepage for existing and returning creators. Though our Product Manager wanted to show this experience to new creators and jump them into the auto-create flow straight from onboarding (more on that later), we also had to find a way to distinguish between our existing, “classic” create experience (which everyone always worked with) and our new, speedy experience. How could we clearly introduce a distinction, when there never was one to begin with?

Though my PM and PD preferred not to mention AI at all in the decision, I advocated for saying “with AI” in the title and in the CTA button to clearly let users know we were using artificial intelligence. The subtext also described how each experience was different, and the pencil and wand also provided visual differences. (Note the magic wand here 🪄, which we should integrate into the AI summary and description tools for more visual consistency! )

This was also a great opportunity to inject a more playful tone to our experience, especially with our updated tone of voice guidelines.

I influenced my team to develop three different loading screens that would included educational content around our event tools and how each could benefit the creator. Becasue it took time in the backend to generate an event page, I used this as an opportunity to be helpful—and a little playful with words like “scribbling” and “tinkering” for our USA market to give our AI a personal flair.

Though our CEO Julia Hartz thought a little different: her feedback wanted this text to be shorter and more clear on the value prop each tool delivered. It was great feedback, since the original text was really long (especially on mobile), and could make creators feel rushed to finish reading it before loading.

The Results

TOOLS THAT AUTO-FILL in your event categories, your event summary and description, and a new ai event creation flow.

The Impact

how do we know if these tools and new flow were successful?

AI SUGGESTION TOOLS IMPACT

We released an experiment with our AI tools that auto-filled event type, category, sub-category, event summary, and description. After three weeks of running the experiment, we observed a neutral outcome with a positive trend for the AI variants in terms of statistical volume. All the key metrics such as publish rate or second event create rate were showing a positive trend, indicating a potential positive impact. So we expanded this feature to more users so we could track performance. There was:

• +6.1% absolute increase in Publish Rate

• +5.6% absolute increase in Publish Paid Rate

• +4.9% absolute increase in Second Create

• +0.2% absolute increase in Second Paid Publish

️Specifically for the auto-fill event category feature: approximately 82k published (and public) events used the AI tool. This can increase their visibility by 20%-24%. Categories that had the most impact were (based on the events that didn't have these fields):

• Community & Culture: 21%

• Performing & Visual Arts: 13% 

• Business & Professional: 12.6% 

• Health & Wellness: 10% 

AUTO-CREATE FLOW IMPACT

In July 2023, we fully opened the Auto-Create flow to 100% of first-time creators in the US. After 7 days, we had 10k new creator exposures and gained some early numbers. While the data was preliminary, it was promising:

• +13.7% increase in publish rate (from 37.9% to 43.1%)

• On mobile, it increased from 27.9% to 36.3%, which is a 30.1% increase.

• On desktop, it increased from 43.8% to 48.1%, indicating a 9.8% increase.

• + 23.7% increase in event create saves (from 49.3% to 61%)

• Time to create an event decreased by more than 35% (from 15 mins to 9 mins).

• +10.3% increase in first paid publishes (1PP)/events published with a paid ticket, when compared to the control without an auto-create option.

The Retro

What i could’ve done better, and thoughts on how i could expand this work.

The grey “Your event is ready!” box felt odd in the Auto-Create flow’s pre-publish step, and it also had a lot of content in it. We also found that the majority of creators (82%) were making adjustments to their AI-generated event before publishing. In the second iteration, I am working with my Product Designer to iterate on the UI to offer more UX-friendly links and personalized actions based on their event (virtual vs. in-person events, events with free or paid ticket types, etc.), as well as ways of how to shorten content. And to make the “Edit event” button easier to access, too.

Projects also don’t just happen in vacuum: I was also concerned about the overall user journey. How would Auto-Create integrate the Pricing & Packaging restructuring work that was also happening at the company? How would we notify users that they may need to pay to publish their event based on what they generate? How would the new SMS verification step (developed by another team), the creator onboarding flow (also developed by a different team), and then our new auto-create experience affect a first-time creator’s introduction to Eventbrite? Is our narrative cohesive? Is it long and now more confusing and arduous?

I conducted a holistic view in Figma of all the current instructions, copy, and messaging we provide first-time creators through the journey, and scheduled meetings with each PM, PD, and Lead Engineer of each product team to suggest low-lift content changes we could implement to ensure that our end-to-end new creator journey was cohesive and less jarring (so that it didn’t look like we were just shipping the org chart). I also added proactive pricing language below the ticket creation field in the Auto-Create step, to let users know that they may need to pay to publish their event.

Another way I could expand this work is to combine our improved classic create flow and this AI create flow into one: having someone input the main details about the event, generating an AI event, and then using the classic create flow for when a creator wants to edit their event. So instead of having two different options, we could make the cognitive load easier for the user by consolidating the experiences into one UX-friendly experience.

And lastly: picking one icon to represent AI in the product. The content principles may have been a bit loose in allowing both sparkles and wands, and we’re beginning to see both being used when paired with AI in the product. So aligning on one icon with the design team and implementing that into our design system would help streamline consistency.