AI is so hot right now.
Generative AI tools such as ChatGPT and DALL-E are dazzling the public, with ChatGPT alone seeing 10 million daily users since its release in November.
No doubt your Twitter feed has been full of people sharing screenshots of their “conversations” with ChatGPT, often in the spirit of fun.
But AI has many business applications, from automating manual processes and monitoring data in real time to targeting customers with greater accuracy. With the pace of digital marketing seemingly approaching the speed of sound, automation may be key to keeping up.
Here are a few ways industry players are using AI in their day-to-day work.
- Refining media campaigns
It took digital marketing company Theorem close to two years to train bots to mimic human workflows in AI-powered media campaigns.
“You want to make sure it’s battle-tested [and] the QA is done,” said Jay Kulkarni, CEO of Theorem, which deploys thousands of global campaigns a month.
Bot training involves combining robotic process automation (RPA) with embedded machine learning and deep learning. Whereas machine learning helps teach bots to copy human tasks, deep learning helps uncover trends, commonalities and efficiencies.
As a B2B company, Theorem has seen many benefits from using AI, Kulkarni said, including speed increases of up to 50% in campaign execution, “meaningful” decreases in error rates and effort expenditures and a faster reporting dashboard for clients.
“The savings and efficiencies are huge,” he added.
- Fueling personalization and engagement
But the customer journey is only getting more complex from a data perspective.
Consumers are engaging across a growing number of touch points, said Vijay Chittoor, CEO of Blueshift, a CDP that uses AI to tap in to first-party data from millions of customers to generate insights in real time.
Brands can automate their decision-making at an individual level across areas such as paid media, email, SMS, mobile app and website interactions, he said.
As a CDP, Blueshift also helps companies connect their ad tech and martech stacks and create unified profiles, which is critical in the face of ongoing signal loss, as marketers scramble to find new solutions for ad targeting.
CarParts.com, a DTC retailer of aftermarket auto parts, partnered with Blueshift in 2019 and has since been able to deploy personalized messaging based on click behavior and transaction data, as opposed to sending identical emails to its entire customer list, said Houman Akhavan, CMO of CarParts.com.
The company uses Blueshift’s AI to predict customer purchase intent and offer differential discounts, as well as to personalize its email and SMS messaging with data points, such as vehicle names, preferred price range and browsed items.
“People want to be spoken to at an individual level,” said Akhavan, adding that personalization requires a “rock-solid first-party data strategy,” an understanding of what’s relevant to the customer and a place to store and segment customer information.
- Automating digital advertising’s “dirty jobs”
Automation often goes hand in glove with efficiency, in part because bots can take over the more tedious tasks from humans.
When Jason White worked at CBS Interactive as the executive in charge of global programmatic marketing and monetization between 2013 and 2020, he saw high turnover among people whose roles required monotonous, manual tasks.
There are certain “dirty jobs” that “human beings don’t really like doing,” said White, who co-founded an intelligent automation platform called Jiffy.ai in 2020 with former CBS colleague Dennis Colón, now Jiffy’s VP of product and strategy.
Jiffy.ai helps people in the programmatic and ad ops trenches automate core, nonstrategic work, he said.
Typically, publishers need to enter the same information across multiple different platforms, White said. But a bot can take the assets associated with an insertion order and put them into an order management system, CRM and publisher ad serving system, accordingly.
To instruct bots on how to execute automated, repetitive tasks with little or no human input, Jiffy.ai uses a combination of RPA, machine learning, natural language processing, analytics, cognitive automation and a model called human-in-the-loop (HITL).
HITL blends supervised ML and active learning whereby people train, tweak and otherwise interact with an algorithm in a continuous feedback loop to help the machine perform better.
Jiffy.ai builds schemas and flow diagrams to teach the AI the operational process, then reiterates that training to demonstrate exceptions and “throw every scenario possible at the AI” to teach it how to operate, White said.
The hope, Colón said, is that the bots will “allow folks to really focus on the business-changing tasks.”
“We really just want to support the workflow,” he said.
- Targeting audiences and optimizing media measurement
Part of AI’s promise is that it will free people up to think about big-picture matters, like strategy and creative – in theory, at least.
“The vision that everyone shares, but is not yet reality, is that AI will save tons of resources and time,” said Hyun Lee-Miller, VP of media at independent media and measurement agency Good Apple.
Setting up and maintaining an AI solution demands human attention, and plenty of it. Nevertheless, the juice has been worth the squeeze for Good Apple.
“Machine-learning-powered audience targeting has proved quite productive from a performance optimization perspective,” Lee-Miller said.
In one of Good Apple’s early tests using AI to support campaigns, it was able to double the optimizations that could be produced manually during the same period, resulting in a 64% improvement in cost per acquisition (CPA).
In another instance, a custom algorithm drove lifts across CPM, cost per quality visit and CPA. Good Apple also had the ability to optimize, in a day, 20 times more than what would have been possible with manual optimizations.
But there is one area where AI still falls short (at least for now), and that’s in delivering insights, Lee-Miller said.
“[Humans] need to dig in to understand ‘the how’ and ‘why’ the results are great,” she said.
Like many other companies playing with AI and its uses across the enterprise, Good Apple is still in the beginning stages.
“We’re not rushing,” Lee-Miller said. “We’re building slowly [and] we’re building good use cases. We’re doing small experiments.”