Havas Media is going through a brain change thanks to IBM and Watson developer Equals 3.
While Havas’s relationship with IBM goes back for decades, it was mostly on the creative side. But Havas Media has over the past year been installing and training a Watson-powered artificial intelligence (AI) solution – in the form of Equals 3’s Lucy – to provide insights and eventually improve media-buying efficiency and effectiveness.
(Read more about how IBM hopes to get developers to build Watson-powered solutions.)
AI previously wasn’t client-ready enough to demonstrate an advantage for Havas’ media-buying process, said Peter Sedlarcik, head of business insights and analytics at Havas Media Group. Nor would Havas have been able to use AI without devoting a lot of time, resources and investment.
But AI is becoming such a part of the lingua franca, it’s now a focus for many enterprise tech giants. Microsoft, for instance, started its Cognitive Services unit, and Google now offers machine-learning APIs.
“The big boys are spending a lot of money and training because AI is all about how you dig data and train the machine to understand the data,” said Rahul Singhal, chief product officer at Equals 3 and a former program director at IBM Watson. “It’s come to a point where it’s mainstream, and you see companies like us and others really drive AI technologies.”
To Sedlarcik’s surprise, Havas could quickly integrate both proprietary and syndicated data streams into Lucy, so its analysts could bake the AI solution into their processes with relatively little development time.
At the same time, Havas Media’s staff needed to learn how to interact with Lucy.
“As odd as this may sound, we’ve had to learn to treat Lucy and the AI technology as a person on our team who is brilliant,” Sedlarcik said. “There’s a lot of information in her brain – but in order to pull the best out of them, you need to meet her halfway and figure out how to ask the right question and frame things in a way that will draw out the best.”
Most software has a fixed deliverable. If you input a formula into Excel, you know the type of answer you’re going to get. But Lucy’s answers evolve as the AI brain learns and it receives more training from human users.
That’s another distinction between Lucy and run-of-the-mill software. Excel can perform a certain amount of tasks, and if you want it to do more, you’ll have to enlist a developer to enhance its capabilities.
With AI however, the line between developer and user blurs. The more you work with Lucy, the more you affect its capabilities and the answers it delivers. So in using Lucy, you are also changing her applications. That requires Lucy’s users to be more thoughtful in how they interface with the system.
“You need to pause and think about what you want the outcome to be and to shape the thinking of the platform to provide it,” Sedlarcik explained. “That really required our participating analysts to reframe the way they engaged with the different dashboards that we used for day-in day-out planning, buying and analytics.”
Consider this: When you ask Lucy a question, it might tap 75 data streams to suss out an answer. When a human user gets the first batch of results back, they’ll need to figure out, among other factors, whether the answers make sense, what resources Lucy tapped, whether the results are qualitative or quantitative and whether they’re data-driven or report-driven.
Then, that same user must go back and ask the same question in a different way. Lucy isn’t plug-and-play; training is an extended, iterative process.
But because people with questions don’t want variable answers, that training needed to happen before Sedlarcik could introduce Lucy into polite society – or at least the daily workflow.
“When we scale it out to a broader user base, we need to control for the more fixed deliverables Lucy can provide,” Sedlarcik said.
Singhal added that AI requires change management: “People aren’t used to seeing a system where one day you ask a question and the next day you ask the same question and the answer is completely different because humans have started training it.”
And it helps to have a set idea on where the AI technology can be applied.
Havas Media focused on three areas. The first was driving efficiency by aggregating as much of its data streams as it could into a single point of access. That’s a good business practice whether you’re installing an AI engine or not.
The agency then focused on driving Lucy’s insights into its strategic processes. “We built a framework within the platform to help get insights more quickly,” Sedlarcik said.
Finally – and this element is still under development – Havas Media wanted to take those insights and connect it to media targeting and activation.
“We’ve made significant progress on parts one and two, and we’re starting to make good progress on three,” Sedlarcik said. “Breaking it down into those three buckets has helped the team members driving this project keep a target in mind.”
It’s been about a year since Havas Media began implementing Lucy, and work is ongoing. In the end, the implementation’s success will hinge on its adoption within Havas Media and whether it creates tangible results around efficiency and helping the agency produce more strategic content quickly.
Early results are promising. In a 2016 beta test, Havas Media experienced an average 75% projected reduction in IT vendor costs in pitch scenarios and the ability to generate “what if” scenarios an average seven times faster than it previously could.
If all goes well, Lucy will also change how Havas Media staffs its business units – not in terms of replacing people, Sedlarcik said, but in allowing the analysts to focus on more strategic, high-level challenges facing their clients.