Machine learning to save millions of dollars of advertising waste at P&G

The power of machine learning is leveraged today by P&G to improve advertising targeting and ROI, especially in search and programmatic. In the near future, ad fraud is an area that might also benefit from this new trend.

Recently, in social media, several ladies were mocking an unsuccessful Gillette sampling campaign. The brand aimed at delivering free sample razors to male consumers, but due to poor targeting capabilities, many of those reached also many females of different ages[1].

The issue of wrong targeting observed here is a big problem both in offline and online advertising. In fact, a key objective in advertising is to maximize its ROI through precise targeting i.e. ensuring that ads reach a desired consumer group[2]. However, precise targeting is still a big advertising challenge: understanding exactly who is reached by ads is not straightforward, even in the digital space.  Target groups are mostly built with limited or incomplete data sets, often not connected at all[3]. This is where machine learning can help[4][5] – analyzing data to identify a set of behaviors (online, like web browsing or search keywords, or offline, like GPS-identified locations visited), to make sure the targeting is more accurate and maximizes some specific objectives[6] (e.g. view rate of videos, clicks, registrations, downloads etc.).

There are several ways in which P&G is leveraging machine learning to improve ROI on its marketing/advertising investments[7][8], for example in search and in programmatic.

In search, P&G leverages machine learning capabilities to improve Adwords results (i.e. reach and cost). In the past, P&G would decide which search key words to invest in based on a self-assessment i.e. based on the desired target groups, a list of potential search key words would be defined manually. In order to optimize the process, P&G partnered with Google and media agencies to leverage machine learning capabilities to determine the optimal search key words for their target segments. For example, based on keywords search data (i.e. which terms are searched by which people over time), it is possible to define the life stage of people (e.g. mums) to then serve them relevant ads across the different platforms[9].

In programmatic advertising, P&G has been the pioneer for years. Programmatic advertising is a technology stack that helps match advertisers (those who want to advertise) with publishers (those who own space to display ads e.g. websites) and allows them to trade the advertising space in an automatic way. Besides the operational efficiencies, one of the main advantages of programmatic advertising lies in improved targeting via machine learning[10][11]. In fact, the technology stack can leverage data on web browsing behaviors to identify specific target groups (e.g. mums, sport lovers, fashionistas etc.). P&G started this journey with an external partner, Audience Science, and then moved to a new tech stack, The Trade Desk and Neustar, in 2017[12][13]. In fact, The Trade Desk keeps on releasing new functionalities that leverage machine learning via “powerful AI that improves advertisers’ decisioning and accelerates campaign performance”[14].All of this helps P&G achieve their strategy of 1:1 precision marketing through “eliminate[ing] waste by reducing excess frequency within and across channels, [and] eliminating non-viewable ads”[15].

Beside targeting capabilities, another important factor that contributes to media investment ROI links to ensuring that ads reach real humans. In fact, one of the emerging issues in digital media is ad fraud i.e. malicious bots that simulate human behaviors, view ads and so inefficiently consume advertisers’ media budgets[16]. According to WFA[17], ad fraud a huge issue in the industry, “likely to represent in excess of $50 billion by 2025”. Machine learning could be leveraged to help strengthen detection systems, assess data and determine fraudulent behaviors also in P&G advertising. The barrier that prevents such scaled solution today is computational power (in the advertising space, a company like P&G has trillions of transactions per day) and quality of detection algorithms. However, “as automated fraud detection tools get smarter and machine learning becomes more powerful, the outlook should improve exponentially”[18] and leveraging machine learning to eliminate ad fraud should become more and more important for P&G in the medium term. For this to happen, P&G should work on two dimensions. Internally, it would be important to dedicate some of the internal machine learning resources to the understanding of the ad fraud issue, to later fight it in a more structured way and with the most advanced machine learning capabilities available. Externally, P&G should partner with machine learning companies and onboard them on how the different media platforms work. A first step would be to share relevant data with potential partners so that machine learning algorithms can be developed and tested.

There are several open questions: How should P&G collaborate on ad fraud detection with other advertisers (as it’s an industry-wide and industry-relevant issue)? Should the new capabilities be outsourced or developed in-house to ensure P&G competitive advantage? How will the governance of ad fraud detection processes work (i.e. who will assess their effectiveness)?

(800 words)

[1] Sapna Maheshwari, “Welcome to Manhood, Gillette Told the 50-Year-Old Woman”,, July 16, 2017,, accessed November 2018.

[2] David Court, Jonathan Gordon, and Jesko Perrey, “Measuring marketing’s worth”,, May 2016,, accessed November 2018.

[3] David Rogers, David Rogers, “Marketing ROI in the Era of Big Data: The 2012 BRITE/NYAMA Marketing in

Transition Study”, (paper, Colombia Business School, Center on Global Brand Leadership, 2012),, accessed November 2018.

[4] Stacy Pollard, “An investors’ Guide to Artificial Intelligence”, Global Equity Research, J.P. Morgan, November 27, 2017,, accessed November 2018.

[5] James Raphael Poole, “Lucy: IBM Watson Analytics meets Madison Avenue—Considering Cognitive Computing Artificial Intelligence Tools Employed for Advertising Media Planning and Buying”, (paper, The College of St. Scholastica, May 14, 2016),, accessed November 2018.

[6] “Nielsen Launches Artificial Intelligence Technology”, press release, April 4, 2017, on ProQuest website,, accessed November 2018.

[7] Thomas Davenport, Randy Bean, “How P&G and American Express Are Approaching AI”,, March 31, 2017,, accessed November 2018.

[8] Ellen Hammett , “P&G’s advertising future defined by AI and a social conscience”,, September 14, 2017,, accessed November 2018.

[9] Lauren Johnson, “Here’s What P&G Has Learned Since Demanding Advertising Transparency 9 Months Ago”,, September 13, 2017,, accessed November 2018.

[10]Chris Victory, “In 2018, Marketers Will Discover More AI Applications in Programmatic Advertising”,, December 6, 2017,, accessed November 2018.

[11] Emma Williams, “The Role of AI in Redefining the Programmatic Advertising Experience”,, March 14, 2018,, accessed November 2018.

[12] Jack Neff, “P&G Shakes Up Tech Providers Behind Global Programmatic Buying”, , May 4, 2017,, accessed November 2018.

[13] Javier Polit – CIO, Procter & Gamble Company, Boardroom Insiders profiles, August 9, 2017, accessed November 2018.

[14] The Trade Desk, “ News and Insights/ The Trade Desk Ushers in the Next Wave of Digital Advertising featuring Koa™, Artificial Intelligence (AI) for Advertisers”,, accessed November 2018.

[15] Joe Moeller, P&G CFO, transcript from Earnings Conference Call Q2 2018, January 23, 2018. Transcript provided by,, accessed November 2018.

[16] Javier Polit – CIO, Procter & Gamble Company, Boardroom Insiders profiles, August 9, 2017, accessed November 2018.

[17] Mikko Kotila, Ruben Cuevas Rumin, Shailin Dhar, “Compendium of ad fraud knowledge for media investors”, WFA Report (2016). World Federation of Advertisers,, accessed November 2018.

[18] Vian Chinner, “Artificial Intelligence And The Future Of Financial Fraud Detection”,, June 4, 2018, accessed November 2018.


The CAD Fad: Shanghai-Based Winsun Wants to Print Your Next House


IBM’s Deja Vu in Disruption

Student comments on Machine learning to save millions of dollars of advertising waste at P&G

  1. When considering the challenges facing corporations looking to advertise not only must companies consider the effectiveness of their messaging, but they must also consider whether their advertising is actually reaching their consumers. Facebook and YouTube are platforms that have been able to build multi-billion dollar revenue businesses through the accurate targeting of customers through advertising and in many ways have democratized advertising so that it is not limited to large corporations with multi-million dollar brand budgets. In the wake of their success other corporations have also developed platforms to advertise for everyday websites to mommy blogs. Although on a brand by brand basis this may represent minimal waste, when aggregated this creates billions of dollars of waste. I agree that P&G is likely one of the best firms to look at tackling this challenge. However, given that brand strategy is their core competency, I think that they should outsource this challenge to their media team as a joint venture. If successful, this will ultimately create an opportunity for P&G to purchase their media team and bring the resource fully in house to remain competitive. However, if the results are able to make the entire market more efficient and reduce fraud in the advertising space, this may be something powerful to benefit consumers everywhere and be an opportunity that other companies will purchase if not solely owned by P&G.

    1. Interesting perspective – this is exactly what happened in the last decade, but unfortunately the outsourcing / collaboration with agencies did not work very well. The problem of transparency came out very strong in the last two years and now advertisers are in-sourcing more and more their media operations (including P&G).

  2. Nice work — I have never even considered the assistance of ad fraud, but it’s very clearly something that could become a huge problem as more and more of advertising budgets migrate to highly cost efficient digital advertising. I for one am skeptical that this is a problem that can truly be successfully combated — it seems to me it would be rather simple to use bots to fraudulently click through advertisements and yet extremely difficult to detect this activity. The only hope would seem for big players in the industry (like P&G) to collaborate in an attempt to stay ahead of fraud technology (as difficult as that may be in practice).

    1. The problem of ad fraud is very present in the industry – many fraud detection companies are in the space (e.g. Moat, Integral Ad Science, Double Verify), but it’s always a game of cops and robbers, with fraudsters trying to engineer new ways to extract value from the system, and 3rd parties that try to catch up.

  3. It is a great article. Ad fraud is one of the emerging topics in marketing across most industries. One of the outstanding questions of that, however, is that given the insufficiency of data in this field – is it possible to scale this mechanism using machine learning tactics, especially if we think broadly about other industries .Given machine learning solely depends on data, is there a way out of it given this is a great idea?

    1. The good part is that there are many data in this space – trillions of transactions happen for second, and all of them are assessed by different tech systems to understand how to best allocate media budgets. More than data, the actual problem is speed i.e. how to be able to analyze so much data in milliseconds.

  4. Interesting piece! I would be curious to explore the tensions between social platform providers and actual advertisers in tackling ad frauds. For example, to a social network such as Instagram, is the incentive to reduce or increase the number of people your ads target? Most likely the latter, in which case, they “might” be more open to being lax on fake accounts, etc. P&G on the other hand wants to reach the most “real” accounts with the least amount of marketing dollars, but has very limited power in filtering for these accounts. Unless social media networks genuinely commit to drastically reducing fake accounts, ad fraud is here to stay.

    1. The tension described here is spot on – today, in the industry, buyers and sellers have usually the agreement to leverage 3rd parties that assess if accounts are fake and so execute financial transactions only for real humans.

  5. Interesting read, and a very enlighting topic. I have never considered the money leakage that can come from Ad fraud. On your question, I’m in favor of companies like P&G to develop their own in-house capabilities mainly because companies need to keep up with new trends in the industry (IT specifically) and I’m talking here in general, and not only to combat Ad fraud, but companies need to have the ability to cope with the new trends in the industry with advanced analytics, AI, and so on. On this particular topic, I see the benefit of cooperating with others in the industry to work jointly to come up with more efficient solutions. Something that’s done quite a lot in the Oil and Gas industry via knowledge sharing in conferences and technical papers – to name a few.

    1. Very good point – this is the direction the industry is taking i.e. having media capabilities in-house (including data, tech and analytics), as previously there was a huge value leakage as agencies may have had the wrong incentives (i.e. maximize their profit vs. the cost of their clients).

  6. When I read this article I couldnt help but think of the entrepreurial gap that is exploited by 3rd non-advertising advertisers to cheat the system. But it seems that this is a blessing in disguise as this challenges leading programmers to roll up their sleeves and find solutions to combat the problem, thereby bring machine learning one step forward. Interesting topic which will gain more and more relevance.

    1. Cops and robbers, fraudsters and verification companies. Given the size of the issue and the fast growth of digital media, this problem is definitely becoming more and more relevant.

  7. Ad fraud was a concept I had never heard about – I enjoyed learning about how the rise of technology to better serve customers can also give rise to technology with malicious intent. I wonder if there are ways to filter through non-human ad-viewership in the same way websites currently test for bots (e.g., typing out letters from an image, or choosing from different images ones that contain a particular object). Would be interested to understand what kind of efforts are already in place.

    1. To learn more, feel free to visit the websites of Moat/Oracle, Integral Ad Science, WhiteOps, Protected Media and similar verification companies.

Leave a comment