At this point in the evolution of online video and connected TV advertising, many people have heard of dynamic creative optimization (DCO), or dynamic versioning – the practice of creating different permutations of ads from a a library of raw creative elements.

But, whilst many of the methods for achieving DCO rely on simple rules laid down by marketers, what if algorithms could assemble the creative for you?

In this video interview with Beet.TV, David Olesnevich, head of product at IBM Watson Advertising, explains how AI-backed algorithms can now automatically re-assemble ad creative based on a range of input and effectiveness data.


The development relies on Advertising Accelerator, an IBM product which uses the company’s Watson cognitive engine to make predictions about the most effective composition of ad creative.

In April, IBM announced Xandr customers can use Xandr’s Invest demand-side platform (DSP) to pilot the use of Accelerator for over-the-top TV (OTT) ad buying.

“The brand has assets that are available,” says Olesnevich. “Maybe they created those assets and used them for linear and are thinking about how they can really activate those assets in CTV and OTT.

“We’ve got a very quick and efficient ad builder that we have assembled. And we take that, assemble some cards, some clips, some calls to action, create a number of different ad variants and creative treatments and then we have those available for deployments when the buy happens.”

Machine-driven optimization

How the ads get built is fascinating. Advertising Accelerator leverages smarts from Watson, whose own footprint in the media business has become significant, to continuously monitor performance and other datapoints for the optimum creative reconstruction.

“When that buying is happening, the AI behind the scenes is learning,” Olesnevich says. “It’s starting from the campaign responses, the actions that are taken.

“And it’s building a model to tell the algorithm which ad creative to deliver at the right time to the right device.

“As the campaign moves on and on … it predicts the right creative to serve … across different device types and give you different experiences based on where you are in your day and what you’ve responded to in the past.”

Buying cues

Olesnevich says Advertising Accelerator also uses data signals like:

  • campaign data
  • device type
  • weather
  • time of day

And it monitors for performance outcomes from actions including:

  • views
  • site actions
  • app downloads

Olesnevich says Xandr has been a long-standing IBM partner. “They have an active client base who’s looking to expand into Xandr’s growing video business – they’re growing exponentially on their side and their clients want more of it,” he says.