Netflix and Algorithms

Following my previous posts on Netflix, I have been reading a detailed analysis in Ed Finn's book, What Algorithms Want (2017).

Finn's answer to my question Does Big Data Drive Netflix Content? is no, at least not directly. Although Netflix had used data to commission new content as well as recommend existing content (Finn's example was House of Cards) it had apparently left the content itself to the producers, and then used data and algorithmic data to promote it. 

After making the initial decision to invest in House of Cards, Netflix was using algorithms to micromanage distribution, not production. Finn p99

Obviously something written in 2017 doesn't say anything about what Netflix has been doing more recently, but Finn seems to have been looking at the same examples as the other pundits I referenced in my previous post.

Finn also makes some interesting points about the transition from the original Cinematch algorithm to what he calls Algorithm 2.0.

The 1.0 model gave way to a more nuanced, ambiguity-laden analytical environment, a more reflexive attempt to algorithmically comprehend Netflix as a culture machine. ... Netflix is no longer constructing a model of abstract relationships between movies based on ratings, but a model of live user behavior in their various apps Finn p90-91

The coding system relies on a large but hidden human workforce, hidden to reinforce the illusion of pure algorithmic recommendations (p96) and perfect personalization (p107). As Finn sees it, algorithm 1.0 had a lot of data but no meaning, and was not able to go from data to desire (p93). Algorithm 2.0 has vastly more data, thanks to this coding system - but even the model of user behaviour still relies on abstraction. So exactly where is the data decoded and meaning reinserted (p96)?

As Netflix executives acknowledge, so-called ghosts can emerge (p95), revealing a fundamental incompleteness (lack) in symbolic agency (p96).

---

Ed Finn, What Algorithms Want: Imagination in the Age of Computing (MIT Press, 2017)

Alexis C. Madrigal, How Netflix Reverse-Engineered Hollywood (Atlantic, 2 January 2014)

Previous posts: Rhyme or Reason - The Logic of Netflix (June 2017), Does Big Data Drive Netflix Content? (January 2021)

Walking Wounded

Let us suppose we can divide the world into those who trust service companies to treat their customers fairly, and those who assume that service companies will be looking to exploit any customer weakness or lapse of attention.

For example, some loyal customers renew without question, even though the cost creeps up from one year to the next. (This is known as price walking.) While other customers switch service providers frequently to chase the best deal. (This is known as churn. B2C businesses generally regard this as a Bad Thing when their own customers do it, not so bad when they can steal their competitors' customers.)

Price walking is a particular concern for the insurance business. The UK Financial Conduct Authority (FCA) has recently issued new measures to protect customers from price walking.

Duncan Minty, an insurance industry insider who blogs on Ethics and Insurance, believes that claims optimization (which he calls Settlement Walking) raises similar ethical issues. This is where the insurance company tries to get away with a lower claim settlement, especially with those customers who are most likely to accept and least likely to complain. He cites a Bank of England report on machine learning, which refers among other things to propensity modelling. In other words, adjusting how you treat a customer according to how you calculate they will respond.

My work on data-driven personalization includes ethics as well as practical considerations. However, there is always the potential for asymmetry between service providers and consumers. And as Tim Harford points out, this kind of exploitation long predates the emergence of algorithms and machine learning.

 

Update

In the few days since I posted this, I've seen a couple of news items about autorenewals. There seems to be a trend of increasing vigilance by various regulators in different countries to protect consumers.

Firstly, the UK's Competition and Markets Authority (CMA) has unveiled compliance principles to curb locally some of the sharper auto-renewal practices of antivirus software firms. (via The Register).

Secondly, new banking rules in India for repeating payments. Among other things, this creates challenges for free trials and introductory offers. (via Tech Crunch)

---

Machine Learning in UK financial services (Bank of England / FCA, October 2019)

FCA confirms measures to protect customers from the loyalty penalty in home and motor insurance markets (FCA, 28 May 2021)

Tim Harford, Exploitative algorithms are using tricks as old as haggling at the bazaar (2 November 2018)

Joi Ito, Supposedly ‘Fair’ Algorithms Can Perpetuate Discrimination (Wired Magazine, 5 February 2019)

Duncan Minty, Is settlement walking now part of UK insurance? (18 March 2021), Why personalisation will erode the competitiveness of premiums (7 September 2021)

Manish Singh, Tech giants brace for impact in India as new payments rule goes into effect (TechCrunch, 1 October 2021)

Richard Speed, UK competition watchdog unveils principles to make a kinder antivirus business (The Register, 19 October 2021)

Related posts: The Support Economy (January 2005), The Price of Everything (May 2017), Insurance and the Veil of Ignorance (February 2019)

Related presentations: Boundaryless Customer Engagement (October 2015), Real-Time Personalization (December 2015)

Data Strategy - Richness

This is one of a series of posts looking at the four key dimensions of data and information that must be addressed in a data strategy - reach, richness, agility and assurance.

---

In my previous post, I looked at Reach, which is about the range of data sources and destinations. Richness of data addresses the complexity of data - in particular the detailed interconnections that can be determined or inferred across data from different sources.

For example, if a supermarket is tracking your movements around the store, it doesn't only know that you bought lemons and fish and gin, it knows whether you picked up the lemons from the basket next to the fish counter, or from the display of cocktail ingredients. And can therefore guess how you are planning to use the lemons, leading to various forms of personalized insight and engagement.

Richness often means finer-grained data collection, possibly continuous streaming. It also means being able to synchronize data from different sources, possibly in real-time. For example, being able to correlate visits to your website with the screening of TV advertisements, which not only gives you insight and feedback on the effectiveness of your marketing, but also allows you to guess which TV programmes this customer is watching.

Artificial intelligence and machine learning algorithms should help you manage this complexity, picking weak signals from a noisy data environment, as well as extracting meaningful data from unstructured content. From quantity to quality.

In the past, when data storage and processing was more expensive than today, it was a common practice to remove much of the data richness when passing data from the operational systems (which might contain detailed transactions from the past 24 hours) to the analytic systems (which might only contain aggregated information over a much longer period). Not long ago, I talked to a retail organization where only the basket and inventory totals reached the data warehouse. (Hopefully they've now fixed this.) So some organizations are still faced with the challenge of reinstating and preserving detailed operational data, and making it available for analysis and decision support.

Richness also means providing more subtle intelligence, instead of expecting simple answers or trying to apply one-size-fits all insight. So instead of a binary yes/no answer to an important business question, we might get a sense of confidence or uncertainty, and an ability to take provisional action while actively seeking confirming or disconfirming data. (If you can take corrective action quickly, then the overall risk should be reduced.)

Next post: Agility

The Exception That Proves the Rule

My thin clean-shaven friend @futureidentity is reassured by messages that appear to be misdirected.

Have we reached peak hipster? I've just been sent an advertisement for a beard brush - right-handed model.

— Robin Wilton (@futureidentity) April 1, 2016

The ads I get are mostly really poorly targeted; am I doing #privacy right, or is the industry just not very good at it?#CISNOLA

— Robin Wilton (@futureidentity) June 7, 2016

This kind of thing in my TL gives me faith that targeted advertising algorithms still suck. https://t.co/JOkuybTXNn

— Robin Wilton (@futureidentity) March 3, 2018

But when I read his latest tweet, I thought of the exception that proves the rule. Fowler defines five uses of this phrase: I'm going to use two of them.

Firstly, when an advert is exceptionally badly targeted, we notice it precisely because it is an outlier - an exception to the normal pattern or rule. Thus reinforcing our belief in the normal pattern - the idea that many if not most messages nowadays are moderately well targeted. This is what Fowler calls the loose rhetorical sense of the phrase.

Secondly, adverts aren't necessarily misdirected by accident. Conjurers and politicians use misdirection as a form of deception, to distract the audience's attention from what they are really doing. (Some commentators regard the 45th US President as a master of misdirection.)

This is how Target does it, so the pregnant customer doesn't feel she's being stalked.

Then we started mixing in all these ads for things we knew pregnant women would never buy, so the baby ads looked random. We’d put an ad for a lawn mower next to diapers. We’d put a coupon for wineglasses next to infant clothes. That way, it looked like all the products were chosen by chance. (Forbes)

So just because a marketing message appears to be a random error, that doesn't mean it is. Further investigation might reveal it to be carefully designed to foster exactly that illusion in a specific recipient. And if it turns out to be targeted after all, this would be what Fowler calls the secondary rather complicated scientific sense of the phrase.

---

Related posts

• Responding to Uncertainty (August 2008)
• 85 million faces (Oct 2016)
• Rhyme or Reason - The Logic of Netflix (June 2017)

Sources

Charles Duhigg, How companies learn your secrets (New York Times, 16 Feb 2012)

Kashmir Hill, How Target Figured Out A Teen Girl Was Pregnant Before Her Father Did (Forbes, 16 Feb 2012)

Wikipedia: Exception that proves the rule, Misdirection (magic)

Rhyme or Reason - The Logic of Netflix

@GuyLongworth, who teaches philosophy at Warwick, is puzzled by the Netflix recommendation algorithm, linking Annie Hall with Son of Saul.

Having seen both, I can only think that this must have to do with rhyme.

— Guy Longworth (@GuyLongworth) June 29, 2017

Philosopher Guy's appeal to rhyme rather than reason seems to be based on the view that the two films have nothing else in common. But this is rather contradicted by the fact that he has actually seen both. Netflix has correctly surmised that people like Guy might possibly be interested in both films.

The first thing to understand about recommendation algorithms is that they are not solely (if at all) based on the intrinsic similarity of two products, but on what we might call relational similarity. If I tell you that people who like pizza also like ice-cream, that is primarily a statement about the "people who like". You might try to explain this statement by observing that pizza and ice-cream both have a high fat content, but then so do lots of other foods.

And when someone has just eaten a pizza, it is perhaps more likely that they will go on to eat ice-cream next, rather than eating another pizza straightaway.

Would it be virtue signalling of me to reveal that I resisted the lure of the second pizza?

— Guy Longworth (@GuyLongworth) June 22, 2017

The second thing to understand is that recommendation algorithms work by trial and error. Netflix wants to know if Guy will accept its suggestion to re-watch Annie Hall, and this feedback will add to its knowledge of Guy as well as its knowledge of relational similarity between films.

Trial and error works better if you have a diverse range of trials. If you watch a couple of films in a particular genre, and then Netflix only ever shows you suggestions within that genre, it will never discover that you might be interested in a completely different genre as well. And you will never discover the full range of Netflix offerings, which could result in your abandoning Netflix altogether.

Diversity of suggestion adds to the richness of the experimental data that are generated. How many members of the "people like Guy" category respond positively to suggestion A, and how many to suggestion B? Todd Yellin, Netflix VP of Product, told journalists in March that "we are addicted to the methodology of A/B testing".

What is genre anyway? In the past, genres (in book publishing, music, film, video games) were defined by the industry or by experts. In 2013, Netflix employed over 40 people hand-tagging TV shows and movies. But a data-driven approach allows genres to emerge organically from the patterns of consumption. Netflix (and Amazon and the rest) will be much more interested in data-defined genres than in industry-defined genres.

In her rant against the Netflix algorithm, @mehreenkasana makes two apparently contrary complaints. On the one hand, Netflix offers her content that is nothing like anything she has ever watched. She dismisses one suggestion with the words "I’ve never watched a show in a remotely similar vein." On the other hand, she doesn't see how Netflix can offer her challenging experiences. "Intensely curated experiences, whether you’re looking to explore movies or to meet people to date, remove one of the most critical aspects of a rich experience: risk, as in going out of your comfort zone."

But as @larakiara explains, "personalization is key to ensuring users keep coming back. But there's also the problem of over-personalization, so Netflix has to introduce variants."

Thus we can see Netflix as an embodiment of at least three of @kevin2kelly's Nine Laws of God.

• Control from the bottom up
  
• Maximize the fringes
  
• Honor your errors
  

"A trick will only work for a while, until everyone else is doing it." (Remember Blockbuster.)

---

Mehreen Kasana, Netflix’s recommendation algorithm sucks (The Outline, 24 March 2017)

Kevin Kelly, Nine Laws of God. Chapter 24 of Out of Control (1994)

Lara O'Reilly, Netflix lifted the lid on how the algorithm that recommends you titles to watch actually works (Business Insider, 26 February 2016)

Janko Roettgers, Netflix Replacing Star Ratings With Thumbs Ups and Thumbs Downs (Variety, 16 March 2017), How Netflix tests Netflix: The story behind the service’s new two-thumbs-up feature (Protocol, 11 April 2022)

Tom Vanderbilt, The Science Behind the Netflix Algorithms That Decide What You’ll Watch Next (Wired, 7 August 2013)

Wikipedia: A/B Testing

Related posts: Competing on Analytics (May 2010), Emergent Similarity (February 2012), The Nature of Platforms (July 2017), Towards the Data-Driven Business (August 2019), Does Big Data Drive Netflix Content? (January 2021)

Personalized emails (not)

Here's a sample from my email inbox, which arrived yesterday.

Dear Richard
I know how important your organization's big data strategy is. That's why I want to personally invite you to attend our webinar. 

How does he know? Is he basing his knowledge on big data or extremely small data? I'm curious to know which.

And what is his idea of a personal invitation? Does he think that personalization is achieved by having his email software insert my first name into the first line? Gosh, how very customer-centric!

But at least the email arrived at a civilized time. Unlike the one that arrived as I was getting into bed the other night, from an eCRM system whose idea of personalization didn't extend to checking what time zone I was in. I guess one must be grateful for these small mercies.

Autumn Events 2015

Open Group Conference - Architecting the Boundaryless Organization

This conference ran from 19th to 22nd of October in Edinburgh.  My talk was on Boundaryless Customer Engagement, and took place on the Monday afternoon. The material was developed in collaboration with my colleague Andrew Forsyth.

The business value of customer analytics and big data is not just about what you can discover or infer about the customer, but how you can use this insight promptly and effectively across multiple touchpoints (including e-Commerce systems and CRM) to create a powerful and truly personalized customer experience.

For most organizations, mobilizing this kind of intelligence raises organizational challenges as well as technical ones. I talked about how some leading companies are starting to address these challenges, and described the vital role of enterprise architecture in supporting such initiatives.

Key takeaways:

• A reference model for omnichannel consumer analytics and engagement.
• An architectural approach for closed-loop integration across multiple customer touchpoints and diverse data platforms.
• A template business case for building and extending your business and technical capabilities for customer engagement. 

---

Unicom Data Analytics Forum - Exploring the Business Value of Predictive and Real-Time Analytics

Was held at the Kensington Hilton in West London on 2nd December.

My talk was on Real-Time Personalization - Exploring the Customer Genome. Retail and consumer organizations have started to develop more personalized interaction with customers, based on rapid analysis of a broad range of customer attributes and propensities, known metaphorically as “genes”. These may be used to target campaigns more accurately, or to generate the next best action in real-time for a specific customer.

For more details and registration, please visit the Unicom website.

---

Here are the two presentations. There are significant overlaps between the two.

Customer Engagement Open Group Oct 2015 from Richard Veryard

Real-Time Personalization from Richard Veryard

post updated 7 December 2015 

Responding to Uncertainty

How does a system respond intelligently to uncertain events?

"A person may take his umbrella, or leave it at home, without any ideas whatsoever concerning the weather, acting instead on general principles such as maximin or maximax reasoning, i.e. acting as if the worst or the best is certain to happen. He may also take or leave the umbrella because of some specific belief concerning the weather. … Someone may be totally ignorant and non-believing as regards the weather, and yet take his umbrella (acting as if he believes that it will rain) and also lower the sunshade (acting as if he believes that the sun will shine during his absence). There is no inconsistency in taking precautions against two mutually exclusive events, even if one cannot consistently believe that they will both occur." [Jon Elster, Logic and Society (Chichester, John Wiley, 1978) p 84]

Austrian physicist Erwin Schrödinger proposed a thought experiment known as Schrödinger's cat to explore the consequences of uncertainty in quantum physics. If the cat is alive, then Schrödinger needs to buy catfood. If the cat is dead, he needs to buy a spade. According to Elster's logic, he might decide to buy both.

At Schrödinger's local store, he is known as an infrequent purchaser of catfood. The storekeeper naturally infers that Schrödinger is a cat-owner, and this inference forms part of the storekeeper's model of the world. What the storekeeper doesn't know is that the cat is in mortal peril. Or perhaps Schrödinger is not buying the catfood for a real cat at all, but to procure a prop for one of his lectures.

Businesses often construct imaginary pictures of their customers, inferring their personal circumstances and preferences from their buying habits. Sometimes these pictures are useful in predicting future behaviour, and for designing products and services that the customers might like. But I think there is a problem when businesses treat these pictures as if they were faithful representations of some reality.

This is an ethical problem as well as an epistemological one. You've probably heard the story of a supermarket, which inferred that some of its female customers were pregnant and sent them a mailshot that presumed they were interested in babies. But this mailshot was experienced as intrusive and a breach of privacy, especially as some of the husbands and boyfriends hadn't even been told yet. (A popular version of the story involves the angry father of a teenaged girl.)

Instead of trying to get the most accurate picture of which customers are pregnant and which customers aren't, wouldn't it be better to construct mailshots that would be equally acceptable to both pregnant and non-pregnant customers? Instead of trying to accurately sort the citizens of an occupied country into "Friendly" and "Terrorist", wouldn't it be better to act in a way that reinforces the "Friendly" category?

Situation models are replete with indeterminate labels like these ("pregnant", "terrorist"), but I think it is a mistake to regard these labels as representing some underlying reality. Putting a probability factor onto these labels just makes things more complicated, without solving the underlying problem. These labels are part of our way of making sense of the world, they need to be coherent, but they don't necessarily need to correspond to anything.


Minor update 10 Feb 2019

Coffee Shop

There are some interesting parallels - and important differences - between the businesses that emerged from the coffee shops of the past (see Wikipedia) and those that might emerge from the Wi-Fi enabled coffee shops of the present-day (discussed by Christopher Baus, Niall Kennedy, Om Malik, Stephen O'Grady and others).

What service does the modern coffee shop actually provide? Provide a place where you can fill yourself with caffeine, spend most of the day staring at a laptop, and perhaps meet some friends sometimes? (Apparently some coffee shops switch the broadband off from time, because this is the only way to force people to talk to each other.)

What service did Mr Lloyd provide in his famous coffee shop? Presence! People formed business relationships because they were in the same place at the same time - they didn't spend their whole time talking to people they already knew - let alone talking to people (via mobile phone or Skype) who weren't even there.

Mobile phone companies are surprisingly bad at presence. If I phone someone who happens to be in the same cafe, or the same train, the phone company would rather take my money for an unnecessary call, rather than give me the valuable information that the person I'm talking to is in the same cell. I have visions of technophiles Skyping one another across a crowded cafe, rather than having a proper conversation. Just like the worst kind of office.

But for those of us that aren't so good at striking up conversations with strangers, the technology of presence might help. Suppose the guy on the next table happens to know Stephen (detected via his address book) or reads Stephen's blog (detected by his subscribed feeds). Maybe that gives me an opening for a conversation I might not otherwise have had. I am sure I sit next to people on airplanes, never knowing what interests we have in common. Perhaps I should develop better conversation skills (or so my wife tells me) but in the meantime ...

Telecoms blogger Martin Geddes has been talking about the opportunities (as yet unrealized) from an understanding of conversation and presence, which I mentioned in my earlier post on Personalization and Presence. But these are also opportunities for the coffee shop. Forget LinkedIn, let's have EspressedIn.

Personalization and Presence

To continue the theme of context-aware services ...

In his post on Context-Awareness, Charlie Bess (EDS) has returned to Paul Miller's question - is this just advertising? Charlie has spend some time recently talking about attention management, and there is certainly some value in using context to target various forms of communication more accurately.

However there are many important forms of communication besides advertising. And there are other important forms of service provision, and many other ways in which services can be differentiated according to context.

Charlie himself has blogged about Personalization and Personal Experience. Neither of these are possible without the service providers having access to some elements of the customer's context.

And my favourite telecoms blogger, Martin Geddes, has just posted some great examples of context-awareness on the Telepocalypse blog. In Disappearing Telephony, he talks about the transient context (presence) of a conversation taking place within one's social network. And in They Told Me, he describes IVR systems that are capable of some interesting forms of personalization. Martin says that the IVR vendor TellMe aims to "optimise to meet user goals, not sub-tasks". I shall be very interested to see how this is done.