Artificial Intelligence

AI-powered marketing 101

“How is AI going to change marketing?”, “What is AI-powered marketing?”, “Should we use AI in marketing?”… All these are questions that I get asked a lot. By clients at the office, by strangers on forum boards, and even by my friends at restaurants and house parties.

The reason why this question is so frequent, is because there is still a lot of confusion around it. Academics have their own definition, which is somewhat restrictive; while vendors seem to be labelling anything they sell with “AI” to sell more of it.

The truth is, it doesn’t have to be complicated at all! Months of answering these questions allowed me to build a comprehensive approach to guiding people through what AI-powered marketing really means, how it is being used, and why it matters.

First, let’s start with the definition I typically give:

AI-powered marketing is a set of emerging marketing practices that rely on artificial intelligence techniques to improve the understanding of customers’ behavior. Thanks to it, businesses can create better customer experiences that drive revenues up (i) and optimize marketing spend (ii).

In other words, AI-powered marketing is the solution to the #1 problem that marketers have been facing ever since marketing emerged: how to get a precise understanding of customers.

This precise understanding is the most critical component of any successful marketing plan1 Marketers previously had essentially two ways of finding answers: market research, and gut instinct.

Both are very imperfect. Market research takes a lot of time and money, and can potentially be biased. Gut instinct is unreliable and virtually impossible to scale.

True, marketers also had massive amounts of data collected over the years (e.g. loyalty program history for each customer), but the data was too hard to be leveraged due to how large, dispersed and dirty the data sets were.

Artificial intelligence brought means of analyzing such data sets, thus giving marketing departments the opportunity to use this data to serve the purpose of understanding customers better.

AI-powered marketing was born.

The reason why it’s a game changer is because it dramatically changes how marketing approaches the matter of understanding with customers. AI-powered marketing allows to shift from an understanding based on identity (i.e. age, gender, zip code…), to a more precise understanding based on behavior (i.e. purchases, online activity, returns…).

To give you a more practical idea, here is non-exhaustive list of various AI-powered marketing techniques that are currently in use:

  • Recommendation engines that can tailor product recommendation at the individual-level. Amazon and Netflix are the most famous examples, with an estimated 35% of Amazon’s sales coming solely from recommendations shown to shoppers2
  • Dynamic pricing models that make prices fluctuate in real-time depending on the state of both demand and supply. Uber’s price surge is the most famous example. Yet, it could soon expand to other industries, like gas stations3
  • Propensity models that can predict individual consumer’s behavior. MAIF (French insurance company) is using one to predict the likelihood of a customer leaving to the competition in order to take action before it happens4
  • Sentiment analysis models that scan what people say about any given product to identify potential shortcomings and opportunities. Arby’s (US fast food chain) realized that its consumers were in love with their sauces thanks to it. Consequently, they launched the sauces as standalone products and leveraged the insight to fuel their marketing campaigns5
  • Programmatic advertising models that automatically bids in real time for online ad spaces based on how likely it is that the viewer will end up buying whatever is being displayed.

Get new articles delivered to your inbox.

Opt-in here to be notified by email as soon as new articles are published. That’s about one email per month, and zero spam.

  1. To illustrate how important it is, please refer to the following case studies I wrote: How market research revealed consumer insights that saved Pampers’ from going bankrupt in China, How Cialis has beaten Viagra by leveraging a better understanding of consumers’ purchase drivers.

  2. More here.

  3. PriceCast Fuel is a technology built by a2i, which enables gas stations to adjust their prices in real-time based on the current state of demand and supply, more about their technology here.

  4. This project has been conducted by Artefact, a French startup agency that specializes in predictive marketing. Check them out here. Or Target (US retailer) that predicts the pregnancy of its consumers[footnote]This certainly is the most famous business case in predictive marketing. I wrote about it here.

  5. More here

Dynamic pricing is eating the world

Since all men are equal before God, all men should pay the same price for the same goods. That was essentially the idea that the Quakers had in mind when they started selling their products at fixed prices. The price tag concept was born, it was mid-18th century.

Since then, setting and paying fixed prices have become the normal way of doing in business. True, some industries got rid of the practice along the way (eg. airlines, hospitality, rentals, insurance, online display ads…), but most economies have remained built around the very concept of price tag.

A shift might be just around the corner though. Lately, dynamic pricing seems to have extended way beyond the usual yield-management suspects. Uber is using it for cabs, Disney, for theme parks and Marks&Spencer for retail.

The list goes on and on, and expands beyond these somewhat famous examples. However, the fact that dynamic pricing has been under the spotlight lately, doesn’t mean that it’s a new concept or idea. The truth is, most industries have been using some form of dynamic pricing way before it became the trendy buzzword it is today:

  • Coke has been priced at different price points in each distribution channel for years, ranging from a few cents a bottle in supermarkets, to over $10 in luxury hotels.
  • Aspirin is notoriously more expensive in gas stations, than at a chemist because if you are considering getting some in a gas station, you must be facing an emergency of some kind, and willing to spend more.
  • Even markets’ fruits sellers typically give deep discounts when closing time approaches in order to stimulate demand and clear their inventory that would otherwise perish.

What is even more noteworthy, is that at their very core, these examples are very similar to the modern applications of dynamic pricing. All of them attempt to get a sense of consumers’ willingness to pay at a given moment, based on several data points such as competitive intensity, inventory, location, consumer segment, urgency, etc.

Yet, the modern landscape of dynamic pricing is different in two important regards, which are useful to understand what made possible today’s widespread adoption of dynamic pricing. First, dynamic pricing implementation have become much more accessible than it was before. Second, dynamic pricing models have become much more sophisticated than they were, giving marketers great hopes about its potential.

Dynamic pricing for all

Three barriers used to block the way for most businesses to start using dynamic pricing: complexity, perceived risk of backlash and lack of relevant data. For most industries, these difficulties have been overcome during the past decade though:

  • It became cheap and easy to experiment with dynamic pricing: When American Airlines started their yield management program in 1985, they had to build everything from scratch. Now, one can simply give a call to any SaaS company and start experimenting on the spot for a few thousand dollars or less.
  • Consumer acceptance increased – risk of PR backlash decreased: The first-movers took the risk of PR backlash (and sometimes got burned, like Uber). Consumers have now come to understand that dynamic pricing happens, which lessens the risk of major PR backlash for newcomers.
  • Marketers are trying to find ways to leverage the data they have at hand: Marketing and IT departments have collected huge sets of data over the year, without necessarily knowing precisely what to do with it. The recent breakthrough of AI made it possible to analyze these super large data sets, thus enabling the creation of efficient dynamic pricing models.

These changes are what allowed industries that would have stuck with their former fixed price models otherwise. Consider cinemas, an industry that is far from being cutting-edge in MarTech. Many of them are now adjusting prices in real-time based on pre-sales, time of booking, weather and various other demand drivers. None of them built a dynamic pricing model from scratch though. They simply phoned Smart Pricer, a software company that does that for their niche.

The truth is, roughly anyone can do somewhat advanced dynamic pricing today. Even the most non-techy WordPress site owner can do it: there’s a $129 plug-in that unlocks the feature!

From guesses to measurements

The second interesting change, lies in the capabilities and promises of dynamic pricing models. More precisely, in the fact that we shifted from extrapolation-based techniques, to measurement-based techniques.

Extrapolation techniques rely on extrapolating an averaged observation about the behavior of a few consumers to the whole consumer base. For example, the fruit seller knows that most people will be willing to buy more fruits if the price drops. So, he drops the price for everyone, by a number that seems reasonable to him.

Measurement techniques are different. They rely on accurately measuring consumers’ willingness to pay based on the analysis of several impactful data points. For example, Amazon’s dynamic pricing algorithms can precisely measure the traffic volume it gets on individual product pages, and what the conversion rates are for each product. If it detects a surge or a decline, it can adjust the price and A/B test different price levels until it finds the sweet spot.

Below is a graph that maps the price evolution of a footwear deodorizer. The price spiked seven times over six months, up to x2 the normal retail price. It turns out that each of these spikes happened following the publication of articles about that very footwear deodorizer in major press titles. These articles most likely resulted in traffic spikes on Amazon, who raised the price more or less depending on how big the traffic increase was, from an extra $5 to an extra $9.

IMG - Amazon Dynamic Pricing Case Study Example

The ability to personalize price points for each individual consumer is the other big innovation brought by measurement-based techniques. Just like regular dynamic pricing, the idea of personalized pricing is not new per se, insurance companies have been doing it for ages when charging smaller premiums from “good drivers” for example. Also, it is not yet as widespread as regular dynamic pricing due to higher technical requirements. Yet, the few cases I heard about make it sound very promising!

Consider the story of Mavi. They are a premium jeans brand retailer that increased their orders by 68% thanks to similar techniques. More precisely, they are personalizing the discounts they give to their customers, based on each customer’s profile. The top predictor of the discount level most likely is the average order value. What is really interesting though, is that they also included the predicted customer lifetime value as a predictor of the discount level.

I find this addition really exciting as I believe it has the power to change what pricing is used for. Pricing has traditionally been about three things: positioning your offering, avoid leaving money on the table, and clearing out your inventory. By taking customer lifetime value into account when it comes to setting prices, one adds an extra use case for pricing. That is, using pricing as a customer retention tool.

Get new articles delivered to your inbox.

Opt-in here to be notified by email as soon as new articles are published. That’s about one email per month, and zero spam.