Tag Archives: Recommendation agents

Why Recommendation Agents Should Let Us Participate

“I see you are looking at our infinite range of stuffed animals, may I help you find what you need?” They are the salespeople of the online retailers; recommendation agents (RA’s). By capturing our perceived preferences based on browsing patterns or interests, RA’s aim to understand our needs. Not an unnecessary luxury of any sort, as the complexity and amount of information we are confronted with often exceeds our limited information-processing capacities and thus the benefits of RA’s can turn into costs. (Dabholkar, 2006; West et al., 1999). If there would be a Maslow pyramid for online shopping needs, it would be the bottom layer; a basic need, indeed.

However, one recommendation agent does not fit all. Different websites use different types of RA’s and the extent to which we can interact with these systems is heavily influenced by the interface design and its dialogue initiation process. Ranging from extensive questionnaires to not even a “hello, I’m here”, the possibility to participate in a two-way dialogue depends on the online salesperson you have encountered. But does the quality and quantity of customers’ input really matter?

In their lab based experiment, using existing RA’s in a controlled setting, Dabholkar and Sheng (2011) show that greater consumer participation in using RA’s leads to more satisfaction, greater trust and higher purchase intentions with respect to the recommended products and the system itself. Existing research already elaborates on the effects of participation in decision making on satisfaction, trust and purchase intentions in the offline and online context (Driscoll, 1978; Chang et al., 2009; Yoon 2002). In addition, much research has been conducted in the RA field, but upon this point failed to combine these two topics.

A great strength in Dabholkar and Shengs’ research, is the fact that there is a significant importance in understanding these relationships in the RA field as they are of huge strategic importance to online marketers. Therefore this topic is highly relevant. Moreover, by adding the dimension of financial risk, the authors are able to also identify that higher product prices moderate the need of participation in the RA context. This gives marketers insight for which products their recommendation agents should have high/low levels of possible interaction and therefore are able to personalize their RA’s per product and possibly increase purchases.

But, there are also a few limitations that need to be taken into account. One could argue that the used sample is non-representative for the online shopping population, as it completely consisted of college students with an average age of 21.91. Although the authors highlight the fact that the largest share of the Internet population is aged 18-32, it is not unthinkable that a student’s perception of financial risk differs from a middle aged person with substantially more spending power. Besides, students perceptions of trust in the online shopping context may be not completely representative, as they grew up with the Internet.

Summarizing, Dabholkar and Sheng give great insights in the effects of consumer participation in RA’s on satisfaction, trust and even purchase intentions. However, generalizability at this point is questionable, so further research across different age groups needs to be conducted to validate these results. But for now; Does your customer base primarily consist of students? Then it is time to revaluate your online salespeople. Get them to communicate with us, we would love to talk!


Chang, C. C., Chen, H. Y., & Huang, I. C. (2009). The interplay between customer participation and difficulty of design examples in the online designing process and its effect on customer satisfaction: Mediational analyses. Cyber Psychology & Behaviour, 12(2), 147-154.
Dabholkar, P. A. (2006). Factors influencing consumer choice of a ‘rating web site’: An experimental investigation of an online interactive decision aid. Journal of Marketing Theory & Practice, 14(4), 259-273.
Dabholkar, P. A., & Sheng, X. (2011). Consumer participation in using online recommendation agents: effects on satisfaction, trust and purchase intentions. The Service Industries Journal, 32(9), 1433-1449.
Driscoll, J. W. (1978). Trust and participation in organizational decision making as predictors of satisfaction. Academy of Management Journal, 21(1), 44-56.
West, P. M., Ariely, D., Bellman, S., Bradlow, E., Huber, J., Johnson, E., . . . Schkade, D. (1999). Agents to the Rescue? Marketing Letters, 10(3), 285-300.
Yoon, S. J. (2002). The antecedents and consequences of trust in online-purchase decisions. Journal of Interactive Marketing, 16(2), 47-63.

Skyscanner: Your digital traveling buddy

Skyscanner is a search-website for traveling. You can easily compare flight tickets from almost all flight companies on price, traveling time and location. Their mobile app is downloaded over 40 million times and it has over 50 million returning customers who use Skyscanner every month. It is because of the quality of the recommendation system (RS) that they have perceived a customer satisfaction rate of 8.2 (trustpilot). In this blog you will learn what the three key success components are of Skyscanners’ RS.

Continue reading Skyscanner: Your digital traveling buddy

The Grid: A revolutionizing way of creating websites

Here is an interesting perspective: the internet itself is one of the largest, if not the largest, collection of consumer-created content. We all use the internet, and the majority of us participate in creating content for it as well, be it simply through updates on social media or actually creating websites or blogpost such as this one. If we follow that logic, then one could argue that the process of creating web content is a service that could be designed to optimally support co-creation. This is exactly the way the Grid looks at the internet as well. This start-up is building a new way to create websites and content, by focusing purely on the simple needs of the content creator.

So what exactly is the Grid? In its simplest form, the Grid allows website owners to only concern themselves with the content they want to put on their website, the Grid will take care of the rest. And for this ‘rest’ part, the Grid has designed an artificial intelligence system which will analyse the content you would like to place, and designs your website optimally according to both the new and already existing content. It for instance looks at colors in a picture you want to post and then changes the colors surrounding that picture on your website to match them. It will adapt to the purpose of your website as well, be it a blog, corporate landing page or a webshop. Basically, as a website owner you do not have to posses technical and design knowledge to be able create content. And this is important, since this allows more people to actually create content. As is also mentioned in the paper by Randall et al. (2005), when customizing goods (or a website in this case), some user might be able to deal with parameter-based interface, where you can adjust every detail but do need to posses some expertise to understand all the details, however not everybody is capable of this. For them, need-based interfaces are better, which might not offer all possibilities but are a lot more user friendly. The Grid is basically a needs-based interface for content creation.


The Grid’s A.I. system basically makes recommendations to improve your website, 24/7. However, while in normal recommendation system these recommendations are made as suggestions, the Grid’s A.I. skips this step and executes the recommendation right away. In order to do this, there has to be a certain level of trust between the user and the system. Due to the Grid continuously running A/B tests in order to find the design best suited for the visitors of that specific website, the Grid’s A.I. learns and builds that trust relationship.

Now what the Grid is doing is really interesting, as it’s a completely new way of looking at website creation. However, it also makes me wonder about the next step. Perhaps in time, it might be possible that every website customizes itself to a specific visitor. For instance, if your computer knows you have bad eyes, or like pictures, your websites will be displayed with larger headings or with more focus on images. In a sense, you would create a personalized internet. Food for thought for sure.

For more info on the Grid, check out this website.

Searching in Choice Mode: the faster and superior way?

good bad choice

With the rise of the internet, consumers can consider an ever growing amount of alternatives of a certain product they intend to buy.  Sequentially, recommendation systems have become a widely observed phenomenon in electronic commerce. Much is known about the effect of these product recommendation on the outcome of the purchase decision of the consumer, but research now also concludes that the process of the product search is transformed in the presence of recommendation systems. EUR Professor Benedict Dellaert  and Gerard Haubl from the University of Alberta explain how we traditionally approach searching for a product and how RAs transform the way we search. We trade in the well documented normative model of consumer search (i.e. Hauser & Wernerfelt, 1990), where we continue to search for additional alternatives until we believe that the potential improvement of our product selection, which the newly examined alternative might represent, does not weigh up against the effort of search we need to put in, for a model where we search in choice mode. In choice mode, we let the recommendations guide us on which alternatives to consider and we continuously compare the alternatives we have assessed among each other. Furthermore, we consider a smaller set of alternatives than we would without the RAs and consider these in more depth. The higher the variability in the (perceived)  attractiveness of the alternatives, the stronger this effect becomes, a finding that contrasts the prediction of normative search theory (see Weitzman, 1979). grfiek Luckily for us, the research indicates that the choices we make using the choice mode match our preferences a lot better than the normative model search, following the average match of 82% with the RA versus a mere 47% without the RA. On top of that, we save time as we do so. Though, there might be a fly in the ointment; more recent research from Adomavicius, Bockstedt, Curley & Zhang (2013) reports that the recommendation system can manipulate consumer preferences and can give rise to a bias. The consumer may choose for a particular alternative, because he believes that its high position on the recommendation system means that that particular alternative is a ‘correct’ answer to the uncertainty he faces. Whether that is true depends greatly on the quality of the recommendations. Also, the consumer might be biased when reporting its satisfaction with the product, as subconsciously he adapts his preferences to the product characteristics of the product bought using a recommendation system. In the end, it seems we can make truly better and more time-efficient product choices with the help of recommendation systems, on the conditions that we critically asses the quality of these systems and remain loyal to our original preferences.


Resources :

Adomavicius, G., Bockstedt, J. C., Curley, S. P., & Zhang, J. (2013). Do recommender systems manipulate consumer preferences? A study of anchoring effects. Information Systems Research, Vol. 24, No. 4, pp. 956-975.

Dellaert, B. G.C.,  Häubl, G., (2012) Searching in Choice Mode: Consumer Decision Processes in Product Search with Recommendations. Journal of Marketing Research, Vol. 49, No. 2, pp. 277-288.

Hauser, J.R., & Wernerfelt, B ., An Evaluation Cost Model of Consideration Sets,  Journal of Consumer Research, Vol. 16, No. 4 (Mar., 1990), pp. 393-408 Weitzman, M., Optimal Search for the Best Alternative, (1979), Econometrica, Vol.  47, No. 3, pp. 641–54.

ZEEF: Taking on Google by using… humans.

Recently a new Dutch start-up has been getting some media attention. They have given themselves the name Zeef (the Dutch word for a sieve), a name that will start to make sense if you continue on reading. Zeef has taken on a rather ambitious goal; challenging Google by changing the way people search for information online. Their trick essentially revolves around sieving the information on the web to only show the relevant bits to people searching for a specific keyword. The interesting thing about Zeef is that this sieving is done by humans.

On Zeef, the information you can search through is managed by so-called curators. As a curator, you can create a page about a topic you like, let’s say backpacking. All the content on this page about backpacking is managed exclusively by you, the curator of the page. Curators can then add links to other websites with relevant information on backpacking on this page, and categorize these links into blocks on the Zeef page. So you might find a block with all kind of links about things to take on your backpacking trip and another block will show you all kinds of websites you can use to find hostels. It is also possible to add images or just blocks of text to your Zeef page, but the main aspect is the collection of links to websites relevant to the topic. The idea behind this is that humans are far better capable of deciding whether a website is relevant to this topic than algorithms, such as Google’s, ever will be. Now you might think “How do I know if this random person who created the page on backpacking is actually knowledgeable on the topic?”, and this would be a fair question since anyone can become a curator and setup a page within minutes. Zeef has tackled this problem by allowing other curators to ‘challenge’ a already existing page on the topic by creating their own page on the same topic. So let’s say you come across the backpacking page on Zeef and think you can do better. You can then simply create a page on the topic on backpacking as well, and when someone then searches on backpacking he/she will be able to choose between the two versions. If you like a page you can vote for it, and this way a ranking of multiple posts on the same topic is created.

(A page I created on Zeef: https://the-grid.zeef.com/axel.persoon)


Zeef has a really interesting approach, as it basically argues that a recommendation system by humans is better than a recommendation system based on a algorithm. And there might be some truth in that statement. As mentioned in the paper by Tsekouras & Li, people appreciate the effort made by recommendation agents. I can imagine that this appreciation of effort is even larger if people know the recommendation agents are human. The strength of Zeef lies in its numbers, and perhaps over time we will see a extensive hub of information completely curated by humans instead of algorithms.

Customer Loyalty and Recommendation Agents

Recommendation agents (RA) are giving online customers recommendations for the past few years. Although the first main function of RAs was to reduce information overload, now it’s also used to increase sales.  More and more information is gathered through the internet and especially social media, to improve personalized preference-based recommendations. At the same time, these systems show success measured by online sales and user satisfaction.

Customer loyalty is considered to be a source of competitive advantage and is useful for long-term business success. Research has shown that there is a strong relationship between customer loyalty, firm’s profitability and stock returns. Returning customers are more profitable than new customers and thus good for business. The aim of the study is to identify the effect between various independent variables (e.g. RA Type, Recommendation Quality, Customer Satisfaction, Product Knowledge, and Online Shopping Experience) and on the dependent variable customer loyalty.

Recommendation quality is based on the preferences of the user and the perceived value of the recommended products. This is the outcome of the type of RA, which could be either content-filtering or collaborative-filtering. Also, the impact of the moderating variable Product Knowledge and shopping experience will be measured. When having expertise in a product, this could negatively affect the customer satisfaction when being advised by a recommendation agent. Shopping experience is also hold in account because the more shopping experience a customer has, the more likely the customer is familiar the interaction with RAs, and the more likely the customer is able to use a RA effectively.

The main reasons for the study is that from marketing perspective, the adopted cognitive-affect-conative-action framework of customer loyalty has not been empirically tested in the context of RAs. This framework states that customers become more loyal when going through multiple stages. Every stage represents some sort of loyalty. There has also been done little research assessing the effect of increasingly higher customer expertise on customer loyalty in the presence of RA usage. Thus, central in this research are the moderating effect of product knowledge on the relationship of Recommendation Quality and Customer Satisfaction.

The results showed that the collaborative filtering RA has a higher recommendation quality than a random RA. The recommendation quality has a positive effect on customer satisfaction and customer loyalty. Also, customer satisfaction is positively related to customer loyalty. The results also show that the impact of recommendation quality on customer satisfaction is negatively moderated by customers’ product knowledge. Thus, product expertise negatively affects the perceived value of the outcome of a RA. Shopping expertise does not have an effect the relationship between customer satisfaction and customer loyalty.  70% of the variance in customer loyalty can be explained by customer satisfaction. This research has shown that an effective use of RA positively influences recommendation quality which in turn positively influences customer satisfaction. When users will have increasing levels of product knowledge, it will negatively influence the customer satisfaction with the website.

The increased knowledge about RAs and how it will increase customer loyalty towards your website is interesting for businesses to retain customers. However, retaining customers are likely to get an increased level of product knowledge. Thus, RAs should always be innovated more and more.


Yoon, V. Y., Hostler, R. E., Guo, Z., & Guimaraes, T. (2013). Assessing the moderating effect of consumer product knowledge and online shopping experience on using recommendation agents for customer loyalty. Decision Support Systems, 55(4), 883–893.

Recommendation Agent 2.0

Just like Don Corleone in the epic Godfather movie, the folks at Washington-based e-commerce giant Amazon are about to make offers to their business partners, the company hopes they can’t refuse. That the firm, like Corleone, relies on ‘capos’ to enforce these offers, can be doubted, but they have a treasure the Mafia never possessed. Amazon’s immense knowledge about their business partners, generated from all sorts of data, takes the firm to a stage where it plans to rely a huge part of its channel domain on its recommendation system. But let me explain.

After establishing overnight delivery years ago and the introduction of the ‘drone delivery service’ last summer, the firm filed a patent in January with the bulky name ‘anticipatory shipping’. The idea behind it could revolutionize the e-commerce environment. According to techcrunch.com the system is the next ‘step towards cutting out human agency entirely from the e-commerce roundabout’ (1). When set up properly the automated collaborative filtering algorithm will learn from the behavior of registered consumers and anticipate what they could possibly be interested in, before the consumers themselves think about it. When matched with a specific product, Amazon plans to wrap the item and send it towards the potential customers before an order has been placed. This means slashing down shipping time by relying on clients’ historical buying patterns, preferences expressed via surveys, demographic data, but also browsing habits, wish-lists and even mouse movements (1). Essentially, the system entirely outsources the consumers’ shopping experience and direct communication with Amazon on basis of an online recommendation agent.

Continue reading Recommendation Agent 2.0

Hostelworld: More than just a place to book your next accommodation

Hostelworld.com Home Page

In the backpacking world Hostelworld has become a household name, which makes sense considering the fact that www.hostelworld.com is the world’s number one hostel booking website, and “the leading provider of online reservations for the budget, independent and youth travel market.” With the mission: “to become the fastest-growing online provider of great value accommodation, using innovative technology to inspire independently minded travellers everywhere,” the company has certainly done just that, now listing over 35,000 properties in 180 countries, with over 3.5 million guest reviews.

The website was created by a hostel owner and IT entrepreneur who realized that at that time, 1999, there was no way to book and pay hostel reservation deposits online. The company started out as a platform connecting backpackers in need of cheap accommodation with budget hostel owners but has the platform has grown to include other types of accommodation including campsites, self catering accommodations, B&B’s and budget hotels.

Hostelworld created “the network effect, with value growing as the platform matched demand from both sides…increasing returns to scale,” (Eisenmann, 2006). One of the ways the website attracted so many users was through their affiliate program with over 3,500 distribution partners, including big names in the travel industry such as Lonely Planet and Ryanair.com.

So how was Hostelworld been able to attract more clients than similar platforms? Continue reading Hostelworld: More than just a place to book your next accommodation

Yummly.com: A Foodies Paradise using Excellent Recommendation Agents


I recently came across an online platform that foodies (like myself) will absolutely love: Yummly.com. This online food platform brings together recipes from all over the web to create the ultimate online cookbook. Searching for recipes that exactly match your taste and needs of the occasion, has never been easier!  Ok, so maybe I am a little behind on the facts, this yummy food platform was created in 2009 and has become the fastest growing food site in the world, with over 15 million users today.

So what is Yummly exactly?  According to their website:

Yummly was launched in 2010 by foodies on  a mission to invent the ultimate kitchen tool. Whether it’s finding a recipe or going to the store, Yummly wants to make it easier for foodies to do what they love – Cook, eat and share! Yummly’s mission is to be the world’s largest, most powerful, and most helpful food site in the World.

Funny fact: founder and CEO Dave Feller got the idea for Yummly because he couldn’t find a way to easily search for recipes that include mustard!

So what has been the success factor of this new platform, that has made it better than the many other recipe websites?

Continue reading Yummly.com: A Foodies Paradise using Excellent Recommendation Agents

Donors Choose

According to the Oxford dictionary, Crowdsource refers to an action in which an individual or a group of individuals “Obtain (information or input into a particular task or project) by enlisting the services of a number of people, either paid or unpaid, typically via the Internet” (1).

Nowadays, there are many types of crowdsourcing, perhaps one of the most popular one is crowdfunding. In which individuals or companies fund their projects by the contribution of small amount of money of a large amount of people, usually via the internet.

One of the most innovative companies in 2014, according to the Fast Company, is DonorsChoose.org, an online charity site, in which teachers from a public school of any part of the United States of America can post a request for materials that students need for their education. Once published in the website anybody who visits the website can contribute to any project.

Continue reading Donors Choose