<img height="1" width="1" style="display:none;" alt="" src="https://dc.ads.linkedin.com/collect/?pid=306561&amp;fmt=gif">
Skip to content

Designing a Scalable Approach to NLP at WPP

Vipul Parmar, Global Head of Data Management at WPP, shares how the ad giant is rationalizing innovation across its network to drive value with NLP more efficiently at scale

Natural language processing’s (NLP’s) potential to enhance marketing and advertising campaigns is huge. The technology is already being used for everything from parsing social media data to identify consumer trends to analyzing the sentiment of earned media coverage, to making it easier for creatives to find useful assets in unstructured media libraries. Some are even using NLP to categorize chatbot data and ensure urgent customer queries are dealt with quickly.

But the sheer variety of potential applications for NLP is creating challenges for advertising giant WPP. With so many pockets of innovation across its network of more than 200 media companies, the group’s data and AI leadership realized they needed a more ‘joined-up’ approach.

“Everyone’s trying to extract intelligence from unstructured data for many different purposes”

Vipul Parmar, Global Head of Data Management, WPP

“There's a lot that's been going on across the WPP network over the last couple of years,” Vipul Parmar, WPP’s Global Head of Data Management, explains. “Everyone’s trying to extract intelligence from unstructured data for many different purposes. But we don’t feel the impact of those projects because they’re all using different services.

“So, we’ve hit the pause button to consider how we can start engineering the project in a way that's really going to demonstrate a broad range of values for everyone trying to create these NLP-related projects.”

Ahead of his appearance at Corinium’s 2022 CDAO UK conference, we speak to Parmar to discover how WPP is reimagining it’s NLP strategy.

An Enterprise Working Group for NLP at WPP

WPP’s network of agencies spans 12 countries and operates in sectors as diverse as advertising, public relations, e-commerce, digital transformation, customer engagement, healthcare and data management.

Without a joined-up approach, these companies worked independently to apply NLP to different use cases. But because these projects often couldn’t be repurposed or applied to problems other WPP agencies are facing, the result was an NLP strategy that was resource intensive and inefficient.

“We’ve hit the pause button to consider how we can start engineering the project in a way that's really going to demonstrate a broad range of values for everyone”

Vipul Parmar, Global Head of Data Management, WPP

Now, WPP has established a virtual ‘working group’ to tackle this challenge. The group aims to enable the firm to deploy NLP at scale, through bringing the best data science talent from throughout its network together to develop a more strategic approach to NLP innovation.

“We acquired an AI organization called Satalia [in 2021],” explains Vipul Parmar, WPP’s Global Head of Data Management. “We're forming a working group with them and some of our other colleagues that have already had a bit of a foray into NLP. But it's really with a broad perspective: How do we start to apply intelligence on top of the intelligence in a scalable way?”

This new working group’s priority is to take stock of the NLP capabilities that have been developed in pockets throughout WPP. Parmar and his colleagues will then identify the key opportunities for further innovation and develop a plan for doing this at scale.

Emerging NLP Challenges for AI Leaders

WPP is in the enviable position of having an in-house talent pool of data scientists with the know-how to work with the latest generation of NLP technologies. But building such a team isn’t the only challenge enterprises must overcome to succeed with NLP.

As Parmar notes, AI-focused executives must also secure access to the data needed to train these models and the data infrastructure needed to process this quantity data. For those wishing to use the latest and most accurate models, this may mean securing access to a GPUs [graphics processing units] either in-house or in the cloud.

“We've got lots of infrastructure that's available to us,” Parmar notes. “But it's the cost. How much compute are we going to start to use up? And how do we ensure that the costs don't run away quite wildly, at a time when we're not actually getting any benefit from it?”

“This is not just about selling more stuff. It's about making sure we're doing the right things”

Vipul Parmar, Global Head of Data Management, WPP

At the same time, Parmar is conscious of the role explainability plays in supporting trust in (and the adoption of) AI systems.

“We are very much worried about explainability,” add Parmar. “With [advanced NLP model] GPT-3, for example, that's been a bit of a ‘black box’ and some of the outputs have been questionable. So, explainability is absolutely paramount around this. We need to understand exactly how outcomes are being arrived at.”

He continues: “We need to understand these outcomes. So, from an ethics point of view, we’re trying to ensure that we have this ethics framework in place for AI, which is constantly evolving.”

Finding solutions to these challenges will no doubt be a priority for WPP’s NLP working group moving forward.

It’s already clear that NLP will prove a pivotal technology for the media and advertising space in the coming years. But even at an industry titan like WPP, there’s work to be done to ensure its potential is tapped in a way that’s efficient and ethically rigorous.

Vipul Parmar is speaking at Corinium’s 2022 CDAO UK conference. For more information and to reserve your place, click here now.