When HubSpot’s Global Events Team had its first kickoff planning session for INBOUND 2024, they weren’t sure about the creative direction they’d take.
One thing was clear though: Given how huge AI has become in the last few years, they knew they’d incorporate it somehow.
I connected with the team to learn more about the important role AI played in pre-production – from creating eye-catching visuals with Midjourney to targeted agendas with Claude.
Creating Fresh Interstitials
Prior to INBOUND 2024, the team kept the onsite interstitials pretty consistent year-over-year, sticking to static visual imagery and light animation.
It was time for a change.
Instead of graphics that would disappear into the background, they wanted to create interstitals attendees would want to engage with.
“This year, AI has been big, and we got an intro to Midjourney and the work that Eduardo had been doing with the tool,” said Sav Aaver, former HubSpot production manager who led the collaboration. “We decided to look into that to add more interest to our screens, create a new element so people have something to look at that sparks conversation, if they’re sitting and waiting and working between sessions.”
The Global Events team tapped Eduardo Garcia-Lopez, a lead designer on HubSpot’s Visual Design team, to oversee the production of fresh interstitials with AI – leveraging Midjourney, Topaz and Runway.
They weren’t trying to create something net new, he says, but they wanted to push their branding.
“We didn‘t want it to feel like, ‘Oh yeah, this is very on the brand. It’s exactly what we would expect,’” Garcia-Lopez said. “We wanted to explore, because AI gives us these possibilities – you can take it anywhere you want but still maintain the same style of the brand.”
After some brainstorming, they settled on three themes:
- An umbrella theme for INBOUND, which was very abstract and closely tied to the HubSpot brand.
- An extended brand for each stage.
- A third, surreal, abstract city theme
The Creative Process with AI
From start to finish, it took the team roughly a month and a half to generate 28 final videos, with 14 minutes of run time adapted for seven screen types at the venue.
Starting with a baseline of shapes, gradients and colors that have always been the foundation for INBOUND graphics, Garcia-Lopez fed those initial images to Midjourney.
That allowed him to generate initial ideas and options to present to the Global Events team.
With AI, the person doing the prompting becomes the director, he says.
“You‘re telling the AI. ‘Here’s my vision. Now, go out and do that,’” he says, “It takes a while, but if you compare that to a team, you still need somebody to drive the vision.”
Once both teams agreed on the final 2D images for each stage and location, Garcia-Lopez headed to Runway for mini video shorts and animation.
“It took me roughly two days to create the Boston mini paper city, generating all the flat images and creating a storyboard, going into Runway and animating all of them,” he says. “Then, I went into Adobe Premiere to edit the whole thing.”
He also used Topaz, an AI editing software that enhances the quality of AI videos.
I wondered, how much would it cost if the team worked with a vendor to create these assets the old fashioned way.
Garcia-Lopez estimates requesting a 30-second clip would take at least two weeks and thousands of dollars.
That said, Midjourney wasn’t exempt from the oddities that happen when you use AI to mimic reality.
“What I presented to the team was the most presentable, the cleanest options, but behind the curtains, there were many generations that were not coming out well,” Garcia-Lopez says. “It was very choppy. You would have something weird happen – buses going into each other, buses running into people, like all these weird things.”
What took a lot of time was identifying the best takes and cleaning up inconsistencies, he says. Thankfully, AI works fast.
“We were iterating in a matter of days. We’d have new options in a couple hours, so it was very, very fast,” Garcia-Lopez said. “As a designer, we would’ve needed a big team to deliver all of these assets, even an illustration for all the variation of styles.”
With Midjourney’s assist, he was able to create 25 distinct styles – a result he calls “almost unthinkable” in the time they had.
“You need a big team with specific skills to accomplish a specific style,” he says, “and we were able to go wild and choose what we wanted.”
AI isn’t without its limitations.
When I asked Garcia-Lopez about the design challenges that come with leveraging AI, he said there’s a big one people often forget.
“Editing something is actually quite hard. With an editable file, like a vector-based design in Adobe Illustrator, you can change every little detail,” he says. “With Midjourney and these AI tools, it’s not that easy.”
It’s a myth that editing with AI is quick, he says. For example, color wise, you might not always get the exact same colors you’d achieve from a color palette. But you can get something pretty close.
Aaver echoes that sentiment.
“It was a big learning experience for us, as the approvers and reviewers,” she said, “learning what we can and can‘t give feedback on, what’s an easy change, what’s not so easy.”
In addition, AI isn’t doing the bulk of the work, contrary to popular belief. There is a lot of bringing it back to Adobe and then feeding it back into the AI model to get the results that you want., Garcia-Lopez says.
“You can do a lot with a small team, but you need skills in other software and a good background to solve a lot of these issues,” he adds.
Creating Custom Agendas for Attendees
In addition to interstitials, the Global Events team also turned to Anthropic’s Claude to create assets.
“We wanted to consider safety and we love Claude because it doesn’t train on the data you input,”said Matt Diaz, the team lead on HubSpot’s Global Events team. “It has a large context window so we could upload a lot of information for it to train on and give us the output without keeping that data for its system.”
They wanted to develop attendee personas and create targeted agendas, which would then be used to promote the event in an interactive “Choose Your Own INBOUND Adventure” quiz.
The team’s first step was asking Claude to create targeted agendas, based on INBOUND’s sessions and speakers, their buyer personas, and their goals and objectives.
“It was able to do that within a minute,” Diaz says, “It was really detailed, it was incredible.”
Their next step was instructing Claude to make sure each agenda was unique, without overlapping sessions, making sure to add the chatbot for context and reasoning behind its choices.
The process took two to three rounds of reviews, as Claude would sometimes hallucinate, creating sessions or speakers that didn’t exist.
The whole process took less than a full day, according to the team.
“If you were to hire a team to do that, you‘re talking multiple days,” Diaz says, “and they’re going to hit you over the head with a lot of money to do that. And this was the price of a Claude pro subscription.”
Once their targeted agenda was built, they created dedicated personas for each agenda. This included marketing-focused ones like the AI-Driven Leader and the Marketing Technologist, sales-ones like the Go-To-Market Strategist, and founder-inspired ones like the Data-Driven Leader.
The second half of this project involved creating a personality quiz that attendees could take to match their persona to a targeted agenda.
“We were operating under the assumption that, especially on social media, people don’t want to go and find out who they are,” said Sarah Hughes, the social media lead on the Global Events team. “They like to be told who they are through these quizzes or personality tests.”
The biggest challenge was ensuring that the quiz’s scoring system had even odds for each agenda, i.e. making sure all quiz takers weren’t getting matched to the same targeted agena.
“Claude got us roughly 85% of the way, and then it came to figuring out that scoring rubric and making sure that it made sense for all the personas,” said Senior Content Marketing Manager Lanna Nguyen.
They published it on the INBOUND blog, driving traffic via email and social channels. To say it was successful is a bit of an understatement.
“In all of our 2024 blog posts, it was the third ranked in terms of impressions behind another post that we put a lot of paid spend behind,” said Lanna.
This quiz became a valuable resource for leads, prospects, and attendees, she adds.
With so many ‘AI gone wrong’ stories floating around, this is a small case study into AI getting it right.
