Personalisation, Profiling, or Pain-point?

A deep dive into the murky waters of Big Data Analytics and Marketing. 15min read.

Big Data and AI are powerful phenomena that businesses are leveraging to target consumers more effectively. Mass customisation is here to stay, but where does one draw the line in aggregating, assimilating, and analysing personal data to promote products?

The internet – and all its weird and wonderful content – is generally funded by advertising. Our digital footprints are making interactions between businesses and consumers ever more individualised, more ubiquitous. Does it irritate us to have to accept cookies? Do we mind being mined if advertisements are relevant? For, in today’s world, that’s the way the cookie crumbles…

Figure I. Anybody Want An Internet Cookie? (Eubanks, 2014)

This blog investigates the implications of highly targeted marketing, for both the consumer and corporations alike. They say marketing is all about the Ps, but the Ps that come to mind on this subject are Personalisation, Profiling, and Pain; erring on the benign, insidious, and uncomfortable side respectively.

First, let’s explore exactly what facilitates highly targeted marketing.

Big Data/Machine Learning Big data analytics (BDA) influences marketing dramatically. Big Data refers to “data sets that are so large (terabytes to exabytes), unstructured, and complex that require advanced and unique technologies to store, manage, analyse, and visualize” (Xu et al., 2016:1562; Ameen et al., 2021). BDA captures customers’ emotional, social, sensory and cognitive responses from multiple touchpoints e.g. social media, online reviews (Ameen et al., 2021:3). In conjunction with Machine Learning (ML), BDA uses probability to offer insights into behavioural patterns, which can be deployed to predict future behaviour. Predictive models exploit patterns to identify risks and opportunities. Deep Learning process even more data, using a multi-layered structure of algorithms called neural networks, mimicking the brain. Without it, Google Translate would be as primitive as ten years ago.

Many hold – the more data, the better – leading to lower estimation variance and thus greater prediction accuracy (Chawla, 2020). To caveat this, Data Science Lead, Sarkar emphasises the “Garbage In Garbage Out” principle: “if we have a huge data repository with features which are too noisy or not having enough variation to capture critical patterns… any ML models will effectively be useless” (Chawla, 2020). Broussard (2018) emphasises the social construct of data (computers and math too): “We might imagine that data springs into the world fully formed from the head of Zeus. We assume that because there is data, the data must be true.” When O’Neil (2006) considers the “sloppy and self-serving ways that companies use data,” she is reminded of phrenology – a pseudoscience of old, whereby “doctors” would probe for indentations on a patients’ skull to predict mental traits.

Figure II. Phrenology head from The Household Physician, 1905 (Wills, 2019)

“Phrenology was a model that relied on pseudoscientific nonsense to make authoritative pronouncements, and for decades it went untested. Big Data can fall into the same trap.” Despite the limitations, for many companies in a hyper-competitive environment, BDA can seem like the silver bullet. 97.2% of American Fortune 1000 companies are investing in BDA and AI and worldwide revenues for BDA were projected to reach $274.30 billion by 2022 (Petrov, 2021).

Figure III. Projected growth of BDA Revenues, (IDC, 2021)

So, what exactly is the purpose of the investment?

Personalisation

The cornerstone of each business is the target audience. To be successful, every business must first identify the audience and market it wishes to target. “In markets where rich data are commonplace and digital channels make personalized offerings easy to deliver, ML methods are propelling large-scale context-dependent personalization and targeting to a new level.” (Ma and Sun, 2019). Segmentation is becoming more fine-grained, with hundreds of accurately carved-out microsegments replacing a few big, coarsely defined segments. The brand you liked on Facebook, products searched on Google, apps downloaded on to your smart device, are increasingly incorporated into your virtual persona.

If you managed a small shop and knew each customer personally, you would already have these insights and would use them to better your business. In building individualised profiles and combining them with predictive analytics, thus delivering a personalised experience, Bill Stensrud contends “vendors are using big data to try to acquire the consumer” (Bollier, 2010; Kitchin, 2014:120). Supermarkets like Tesco and Walmart link checkout sales to customers’ loyalty/credit cards, using insights to devise better pricing strategies, store layouts, advertising campaigns, product placement (including efficiency measures too) (Bollier, 2010; Kitchin, 2014:120). Manyika et al. (2011) suggest BDA could increase retailers’ operating margins by 60% (Kitchin, 2014:120).

In e-commerce, accelerated exponentially by the global pandemic, billions of customers are not visible, thus “recapturing this level of insight is extraordinary” (Stephenson, 2018:41). Furthermore, using BDA, marketers are more likely to discover the customers’ thoughts and opinions on the brand “behind closed doors.” This is carried out by sentiment analysis, which uses text analytics and natural language processing, to discern whether underlying opinion is positive, neutral, or negative (Rajanarthagi, 2021). Companies can subsequently refine their value proposition, or positioning accordingly. Combing the internet to glean nuggets of information that customers do not share formally, is surely an innovative way of operating.

Stephenson speaks candidly on how to leverage BD to gain a competitive advantage. Further to the ‘given,’ (personas based on static features e.g. gender, age), Stephenson urges marketers to broaden segmentation criteria to include whether they read or write reviews, which topics and hashtags they mention, what filters they frequently deploy. Yet preferences and necessities are dynamic. An MBA student today filtering accommodation for class at The Shard by price, perhaps upon landing a high-powered job, my filtering behaviour will change (but probably not).

Figure IV. Personalisation ad (Snovio, 2021)

This means that effective targeting not only identifies the right match – value proposition and customer – but also delivers at the right time in the right context. In e-commerce, AI enables a recommendation based on multiple competing objectives, such as customer satisfaction, maximum revenue, inventory management, etc (Stephenson, 2018:51). That an algorithm can “listen” in real-time, holistically assessing business needs, almost parallels a skilled leader’s role. This balancing act is miraculous. Nevertheless, can AI replicate the same level of personalised care as humans?

As Head of Sales for a film sales agency, I went above and beyond to make clients feel special; accompanying them to screenings, events, and taxi ranks. We became familiar over dinner and wine, them sharing information, often unprompted. While many wanted to connect on social media, I did not pounce on this opportunity to find out more information to sell them films. Besides, personal information does not always translate into what someone may buy. My clients often bought based on the market for which they would be acquiring rights. In this regard there is not necessarily a causal link between interests and purchase. Some assumptions are wildly off. Some – like recommendation algorithms from Spotify and Netflix – are uncannily apt.

My clients proffered information based on our relationship. They seemingly did not mind my access to their private life online. In a bizarre parallel, people do not mind major platforms suggesting similar playlists, romcoms, or authors. Why? Firstly, because many platforms offer unprecedented convenience. They have won us over, but are not complacent, hence the innovation and value-add in content suggestions. Secondly, we have a relationship, and we trust they will wisely store our data. Finally, assumptions on leisure preferences don’t generally offend. Naturally, this changes when assumptions are intimate in nature.

Profiling

US retailor Target, was an early adopter of BDA. You have likely heard the controversy whereby Target’s algorithms made sensitive inferences about people. Analysing historical buying data, Target statistician Pole noticed that patterns started to develop reflective of becoming pregnant. Controversy stemmed from an uninformed father receiving discount coupons on baby products, assuming Target was encouraging his daughter to become a mother, only to learn she was pregnant (Duhigg, 2012). Target acquiesced that this sort of behaviour was problematic, however, adapted accordingly! Becoming ever more cunning, Pole states “we started mixing in all these ads for things we knew pregnant women would never buy, so the baby ads looked random.” (Duhigg, 2012). Apparently, a pregnant woman will use the coupons and become a loyal customer, if she does not suspect being spied on.

This was 10 years ago. Now online shopping is “the new norm,” we “interact” regularly with smart speakers, and BDA and ML has advanced exponentially. This all means that the scope has widened for marketers to peer into our lives and perform profiling, which The Cambridge Dictionary defines as “the activity of collecting information about someone, especially a criminal, in order to give a description of them.” There is a fine line between personalisation and profiling, but the latter evokes personalisation through obtrusive, nefarious ways.

Cambridge Analytica (CA) is a prime example. CA harvested personal information from over 87 million Facebook users through an external app in 2015 (Ma and Gilbert, 2019). This helped to flesh out psychological profiles, which was leveraged in political campaigns. Psychographics focuses on predicting traits measured by the Big Five personality scale: openness, conscientiousness, extraversion/introversion, agreeableness, and neuroticism. Traditional demographics-centric target marketing shows cleaning products to housewives, whereas psychographic-based targeting shows products relating to safety to especially neurotic people. CA whistle-blower Wylie declared: "We exploited Facebook to harvest millions of people's profiles. And built models to exploit what we knew about them and target their inner demons" (Ma and Gilbert, 2019).

With profiling, there is a sense that businesses are more predatory, arguably exacerbated by cookies. First party cookies permit website owners to collect analytics data, remember language preferences, and generally deliver a positive user experience. Indeed, Stephenson champions Big Data and algorithms for analysing and responding to user behaviour in real-time: “this ability at scale, is what the big data ecosystem provides. Your customers are continuously expressing preferences as they type search terms and subsequently select or ignore the results.” Conversely, third-party cookies, constitute "third party" MarTech. Browsers of websites are often unaware of the dozens, if not hundreds, of trackers that are collecting their information, to subsequently sell to improve targeting.

It is likely data from third parties are fed into algorithmic profiling, “that can be used as an indicator to classify a subject as a member of a group” (Hildebrandt, 2008; Mann and Matzer, 2019). These categories are formed from “probabilistic assumptions” (Leese, 2014: 502) thus are de-individualised (Mann and Matzer, 2019). Many researchers have pointed to the decision of a loan application, not assessed on the basis on individual risk, rather postcode. The Wall Street Journal simulated visits to Staples.com in 2014 from US postcodes (more than 42,000) discovering prices varied for a third of Staples products, depending on factors including income and proximity to competitors (Klein, 2014).

The idea poor people are made to pay more (e.g. for staplers) is confirmed by O’Neil (2016:203) who states payday lenders and “their ilk” commence by targeting “the easiest targets, the low hanging fruit,” because “they have less access to information” and are “most desperate.” Postcodes can function as “indirect prox[ies] to other indicators such as the socio-economic or racial composition of one’s neighbours [leading] to concerns about social sorting and discrimination” (Mann and Matzer, 2019). Price discrimination is illegal based on race, religion, etc, however, indirect discrimination is prevalent online yet extremely difficult to detect. In the context of algorithmic profiling, especially where machine learning is used to create new inferential categories, this problem exceeds even the issue of indirect discrimination (Mann and Matzer, 2019). The famous Blackbox problem with AI, is that it is impossible to determine how the algorithm came to its decision.

Figure V. AI’s Blackbox Problem (Worldline, 2021)

Pain

“The allure of the technology is clear - the ancient aspiration to predict the future, tempered with a modern twist of statistical sobriety,” states AI ethics expert Pascale, “yet in a climate of secrecy, bad information is as likely to endure as good, and to result in unfair and even disastrous predictions” (Broussard, 2018:115).

The internet is awash with examples of unprincipled targeting. After one lady’s mother passed away, she googled gravestone prices, thereafter, was virtually besieged (Nudson, 2020). Similarly, another woman who was diagnosed with premature ovarian failure at 29, and will never have biological children, states she still gets pregnancy-related advertisements 5 years later (Nudson, 2020). Tia Dole, of a suicide prevention organization for LGBTQ youth, says that targeted ads can put people at risk if, for instance, an adolescent shares a computer with a family that doesn’t accept their sexuality (Nudson, 2020). When advertisements remind someone of something that they really want but cannot have, or surface an uncomfortable secret, it can cause myriad mental health problems. Clearly here, the brands presented are resented, rather than desired. Online “surveillance” to sell products can lead to consumer backlash” (John et al., 2018).

In 2018, Amazon filed a patent related to identifying users' physical and emotional well-being based on speech data (enabled by Deep Learning). If a user coughs, Alexa may suggest a chicken soup when you probe for recipes. It would be a boon if Alexa becomes so advanced that she detects throat cancer, but O’Neil and Broussard indicate how technology often gets it wrong, resulting in unintended consequences.

Article 22 of GDPR stipulates that citizens have the ‘right’ not to be subject to automated decisions. Despite this, Leese (2014: 500) argues that “data-driven forms of profiling produce a distinct form of knowledge that appears dynamic and implicit, and thus continually escapes the scope of the regulatory legal regime” (Mann and Matzer, 2019). That regulation is forever catching up with technological innovation, like Alexa’s well-being recognition, is another intractable problem.

Final Thoughts

Evidenced by the blurriness of content between the headings, targeted marketing can meander between Personalisation, Profiling, and Pain. The examples of discrimination are in the minority, but sizeable nevertheless. Thankfully, bias and privacy invasion, is being taken more seriously.

Apple’s Safari and Mozilla’s Firefox already block cookies, but now Google’s Chrome will phase out the usage. Tech start-up, Gener8, which enables users to monetise their data, caused a furore in The Dragon’s Den recently. Users are shown targeted ads from Gener8 (based on preferences) and they earn (or get promotions) as a by-product (Parson, 2021). Paradoxically, Gener8 identified a real and widespread customer pain-point, turning cookies against parasitic MarTech firms, and empowering the user.

Further, regulators worldwide are cracking down on breaches to data protection laws, leading to hefty fines. Effort is being made to focus on interpretable White Box AI. Incumbent tech firms are hyper-aware of AI’s potential bias, and the importance of Diversity and Inclusion. Google doesn’t let marketers target based on sexual interests or personal hardships, and Facebook inhibits advertisers targeting based on race, sexual orientation, and medical conditions.

To keep the internet free – advertisements relevant, helpful, and non-offensive – John et al., (2018) suggest applying norms about information sharing from the off-line world e.g. don’t infer something about someone of an intimate nature. Rightly or wrongly, businesses will utilise BDA and ML to continue making inferences using our data, but it is critical to forge a good, old-fashioned relationship first.

Useful articles in the domain of BDA and ML

Albayrak, N., ?zdemir, A. & Zeydan, E. (2019) An Artificial Intelligence Enabled Data Analytics Platform for Digital Advertisement.

Ameen, N., Hosany, S. & Tarhini, A. (2021) Consumer interaction with cutting-edge technologies: Implications for future research. Computers in Human Behavior, 120 106761.

Broussard, M. (2018) Artificial unintelligence: How computers misunderstand the world. MIT Press.

Chawla, V (2020) More Data Always Better For Building Analytics Models? [online] Available from: https://analyticsindiamag.com/is-more-data-always-better-for-building-analytics-models/

Chithrai, M (2020) How Is Big Data Analytics Using Machine Learning? [online] Available from: https://www.forbes.com/sites/forbestechcouncil/2020/10/20/how-is-big-data-analytics-using-machine-learning/?sh=3bd97e1e71d2

Duhigg, C (2012) How Companies Learn Your Secrets. [online] Available from: https://www.nytimes.com/2012/02/19/magazine/shopping-habits.html

Eubanks, E (2016). Anybody Want an Internet Cookie? [online] Available from: https://www.relationshipone.com/blog/anybody-want-internet-cookie/

Hill, K (2012). How Target Figured Out A Teen Girl Was Pregnant Before Her Father Did. [online] Available from: https://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/?sh=21b9a7df6668

John, L. K., Kim, T. & Barasz, K. (2018) Ads that don’t overstep. Harvard Business Review, 96 (1): 62-69.

Klein, J (2014) We pay a high price when retailers micro-target us. Available at: https://www.wired.co.uk/article/josh-klein

Kitchin, R. (2014) The data revolution: Big data, open data, data infrastructures and their consequences. Sage.

Ma, L. & Sun, B. (2020) Machine learning and AI in marketing–Connecting computing power to human insights. International Journal of Research in Marketing, 37 (3): 481-504.

Ma, A & Gilbert, B (2019) Facebook understood how dangerous the Trump-linked data firm Cambridge Analytica could be much earlier than it previously said. Here's everything that's happened up until now. [online] Available from: https://www.businessinsider.com/cambridge-analytica-a-guide-to-the-trump-linked-data-firm-that-harvested-50-million-facebook-profiles-2018-3?r=US&IR=T#what-data-did-they-get-2

Mann, M. & Matzner, T. (2019) Challenging algorithmic profiling: The limits of data protection and anti-discrimination in responding to emergent discrimination. Big Data & Society, 6 (2): 2053951719895805.

O'Neil, C. (2016) Weapons of math destruction: How big data increases inequality and threatens democracy. Crown.

Parsons, J. (2021) Cookie-blocking browser Gener8 will pay you for seeing adverts. [online] Available from: https://metro.co.uk/2021/04/28/dragons-den-scramble-over-cookie-blocking-browser-that-pays-for-data-14486940/

Petrov, C. (2021) 25+ Impressive Big Data Statistics for 2021. [online] Available from: https://techjury.net/blog/big-data-statistics/#gref

Rajanarthagi. (2021) Role of Big data in digital marketing - The Future. [online] Available from: https://gecdesigns.com/blog/role-of-big-data-in-digital-marketing

Siegel, E. (2020) When Algorithms Infer Pregnancy or Other Sensitive Information About People [online] Available from: https://montrealethics.ai/when-algorithms-infer-pregnancy-or-other-sensitive-information-about-people/

Stephenson, D. (2018) Big Data Demystified: How to use big data, data science and AI to make better business decisions and gain competitive advantage. Pearson UK.

?