Sportsbet has been using generative AI for three years – making big efficiency gains in ads and asset creation while boosting in-app CX. But GM Tim Hernadi sees much deeper impact rapidly incoming from agentic AI: “It’s going to have a pretty profound impact on how we think about marketing and what our marketing workflows will look like.” Sportsbet is already building and using agents to take TV ad approvals from around 100 hours of work to 13, for which the marketing team is eternally grateful. 

Upon realising Midjourney could 20x character creation productivity in June 2022, Hernadi went all-in on AI.

Since then the wagering brand has built AI into its CX to give punters better data faster, he told the recent In-House Agency Council forum in Sydney. 

“I might be going to the Melbourne Cup and it’s a really wet day, so I might prompt the app to say ‘give me a number of selections that are wet track runners’. It’ll spit out maybe three responses. Okay, great. ‘How many of those have actually run a distance of 3,200 meters?’ Bang. It reduces that even further. Okay, ‘which barrier are they racing from?’ Again, there’s another advantage there, based on how the track might be playing. All of a sudden, you’ve got some information that might have taken the better part of two hours to analyse yourself in a matter of seconds.” 

It’s also using analytical AI to deliver personalisation, or “preference models” to punters, per Hernadi. So it knows what they tend to bet on and serves them stuff accordingly. Which likely means personalisation is powering profit.

It’s likewise notching major cost and time efficiencies by pushing into AI video ad  creative and production that actually works.

Hernadi showed an AI-powered video of UFC fighter Alex Volkanovski to illustrate his point. Ahead of Volkanovski’s title fight versus Lopes, Sportsbet used Chat GPT for scripting and song lyrics, Sora for the style frames and storyboarding, Hedra for video generation and lip syncing, and Suno and Udio for music generation.

“UFC is probably the fastest growing code in the world, particularly amongst males under the age of 30 …  Our hypothesis internally is that we will produce entertaining content on the basis that if people love your brand, they will use your brand.”

Which likely means more bets.

“We’re able to demonstrate [in that Volkanovski example] how we integrate the use of AI into a video clip. But what I think we found along the way is efficiency. We were able to reduce the production timeline from effectively three months through traditional methodologies, to five weeks. We saved an estimated $150k in post-production [costs],” per Hernadi.

“I think the pleasing thing is that this type of technology can be used to generate results that are pretty compelling: Three million combined video views and seven million plus engagements.”

Agentic avalanche

Hernadi thinks the biggest impact is going to come from agentic AI. Sportsbet is now using agents to drive major operational gains – and do the jobs marketing hates. Like CAD approvals for TV ads. 

“About five weeks ago, I sent Toby [Barrett, Sportsbet’s production manager and animation director] over to the World AI Conference in China. This thing is at a scale that you have never seen before: 300,000 attendees … imagine three MCGs full, 1,400 companies and 250 speakers.”

The big talking point?

“Agentic AI. Every single booth or company that was there were talking about agentic … I think it’s going to have a pretty profound impact on how we think about marketing and what our marketing workflows will look like.”

Sportsbet has since built its own agents to handle the drudge work of getting clearance to run TV ads via Clear Ads – AKA Cad documentation.

“Producing CAD documents is one of those repetitive, shit tasks that aren’t that enjoyable. It’s basically a document that you have to submit to Clear Ads to get your ad classified before it goes to air on TV. 

“Year to date, we’ve had to produce 1,600 of these things, so not a small task. And the brand team just hate doing them.”

Which made them a perfect application for agentic agents to handle – and when Barrett returned from China, it was the first thing he targeted.

“What Toby did was trained the GPT on previous CAD docs that we produced. Basically this is called ‘grounding’ the agent. You then define what it can and can’t change. He specified where to find the relevant information to update, and then he instructed it to check its work.”

The upshot? “The manual process to produce those 1,600 documents was about 100 hours of work for the team. Automation reduced that to around about 13 hours. So the time saving is significant, but we could push that further if we look at it through a multi-agent approach,” said Hernadi, which he duly explained.

“Alison is our brand manager. She’s got the ad to the point where it’s compliant, it’s approved, and status in [AI platform] Monday.com has been changed to complete. That is the trigger for the agent, because Agent One has been set up to scan for completed TVCs every five minutes. So it automatically checks the new ads with complete status, and then it downloads the video,” per Hernadi.

“Agent Two then takes the downloaded video and creates a transcript and the CAD docs. Agent Three then uploads the docs back into Monday.com.

Agent Four then uploads and sends the CAD docs directly to Clear Ads. Clear Ads approve the TVC for launch – and then send a notification back to our media team, who then dispatch the TVC.” And probably raise a glass to their agentic friends. 

“So that’s an example, end-to-end, of where you can eradicate a large chunk of what is a repetitive and annoying process,” said Hernadi. “I’m sure a lot of you have got a whole bunch of other use cases within your marketing flows that are just as repetitive and annoying as this one.”

Advice for those launching agentic AI

“Change management is really important,” Hernadi underlined. “You need to think about how you are framing AI within your organisation and structuring [for it]. We’re going to unleash Toby, who will become our AI lead within marketing. Supported with a team of people, he will be tasked with trying to identify what these repeating processes are – and try and solve that.

Governance and data security are likewise paramount.

“You can’t just build these tools and unleash sensitive data into the world. Our view is we will start with processes that don’t touch any PII [personally identifiable information] or any other sensitive information the organisation holds,” he said.

“You need to think about collaboration readiness [with IT, information management and cyber security teams] because you’ll be interacting with a group of people in your organisation that you’ve never even talked to before.”

Workflow management, per Hernadi, is also critical.

“Ideally, you want to set [agentic activity] up under a single source of truth. So Monday.com will be our single source of truth. This will be embedded into the workflow, and that’s where you’re going get your efficiency. So think about that process mapping – it’s an obvious one, but once you understand what the process is for some of these tasks, you can apply an AI approach to it.”

Platform choices: don’t lock-in

Sportsbet has run the rule over six Ai platforms, per Hernadi, and is currently narrowing down to three: “Mate AI is one, another is an Australian company called Relevance AI. The third is called n8n.

“The difference with these platforms is they are low or no code, meaning you do not need a developer to prompt these tools. So anyone effectively in marketing using Chat GTP can instruct these tools to create these workflows that generate an automated outcome.”

Whichever platforms chosen and processes attacked, Hernadi says there is a golden rule: Don’t lock yourself into anything.

“Fail fast. That’s probably the biggest learning that we’ve had when you use a whole bunch of different AI tools. We’ve abandoned a lot of them. They’re fit for purpose at a point in time, but there’s always something better coming. So build for modularity. Don’t get too wedded in terms of using one tool, because the technology is changing at such a fast pace.”