Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

How to start with cross-sell analysis - Pavel Jašek

Podczas Blueffect - I Ogólnopolskiego Kongresu Efektywności Pavel Jašek przedstawił:

- jak za pomocą wskaźników dostępnych w Google Analytics ocenić potencjał sprzedażowy strony WWW,
- jak wykorzystać pozyskane dane do optymalizacji procesów sprzedażowych.

  • Login to see the comments

  • Be the first to like this

How to start with cross-sell analysis - Pavel Jašek

  1. 1. Slide notes will help you understand what was said during the conference. 1
  2. 2. Geddy Van Elburg’s presentation mentioned the importance of average ordervalue. In this presentation you’ll learn if leveraging your AOV can be beneficialfor you. 2
  3. 3. Everybody wants to be like Amazon and cross-sell products like crazy.Amazon makes 20-30% of its sales from recommendations.Only 16% of people go to Amazon with explicit intent to buy something.Source: Toby Segaran, Ravi Pathak 3
  4. 4. Not a lot of website take care of their related products, even if their e-commerce platform allows to manage relations easily.The weakest point is human patience to fill in related products. Like you cansee in this case.I completely understand, why they don’t care about it. It is quite difficult tomanage connections of 20 000 products.Start with a minimum level of categories which should be perfected. But whatproducts to connect together? 4
  5. 5. There are several levels of cross-selling connections which you can readabout in every theoretical article.What we found effective is to analyze your historical transactions and seewhat customers wanted naturally.Like in this example, where we don’t cross sell ovens with induction hobs orfridges, but with some basic baking tins. But it makes sense.When you’re buying an oven, you probably want to bake in it. So why don’tyou grab some nice tin that you can be sure that fits into your oven? 5
  6. 6. So far you haven’t seen anything advanced. We just want to recommend tothe customer some products that are really relevant.If you want to start connecting dots even in your store, use your current dataand knowledge.It is possible to understand what products and product categories to focus onfrom historical data and knowledge of business context. 6
  7. 7. You may think that analyzing your data is something too complicated.Online marketers are usually scared of any advanced analytics that is notdirectly visible in Google Analytics as it is too technical or too robust toaccomplish.Don’t worry about that. I want to show you basic step-by-step tutorial how toanalyze your data.Get your hands dirty and dive into data! 7
  8. 8. You all probably use e-commerce tracking in Google Analytics for trackingyour orders or other type of transactions.That’s great, because you already have a huge amount of data to analyze.I personally love using Excellent Analytics as a tool to get my e-commercedata into Excel where I do two basic things:1) I start looking at the data. You can see that some of the rows have different color, because those categories were sold within the same transaction. This will be the base for our cross-selling analysis.2) The second point is that I clean the data. In Czech republic there is some kind of recycling tax for electric good that the customer has to pay. For cross-selling analysis I’m not interested in those fees, so I wipe them out. 8
  9. 9. Don’t worry, I don’t want you to start programming not even trying to readwhat is on the screenshot.But we need somehow to analyze what products or product categories wererelated in our orders. So I recommend you to use the software called R. It iscompletely free and although it looks very technical, it is quite easy to use astatistical library that will help you find strong associations in your orders.On my last slide I put a link to an excellent and short article about how to doprecisely this analysis. I’m sure you will make it in less than one hour.This is the real result from one of our client’s e-commerce data. Some minutesbefore we were talking about ovens and other products for baking. You cansee that customers who are buying installed ovens are also buying inductionhobs. The confidence metric shows us that it is not that true vice versa, socustomers buying induction hobs are not buying installed ovens that much.You can see that it is very clearly stated what product categories are related.We don’t have customers that are buying from completely different categories,but only very related goods. This is very important fact to recognize! 9
  10. 10. Now how do you know that it is worthy to start relating your products?We’ll go back into Excel and with an easy pivot table we’ll divide ourtransactions according to the number of items in them.If we want to estimate how can sales go up with stronger cross-selling, weshould simulate a decrease of items with only one item.From my simulation you can see how higher average order value helps yoursales.I chose 12 % as a very conservative estimation based on the confidencemetric that I just showed you. 10
  11. 11. Now it’s your turn. As a homework after this presentation, you should try thecross-sell analysis yourself.In case you see some associated products, don’t start changing your sitecompletely. At first, try connecting manually products in top categories. In mycase from the example shown before a minute, I would start with ovens, GSMphones and compact digital cameras.If you are sure you can’t handle it manually, there are some services that canease you the work.Lift suggest is one of the examples. It is made by an American-Indiancompany Tatvic and they are running those codes in R software on theirservers and if you insert just a small code on your product pages, it willautomatically serve related products.In any case, I want to show you how easy you can measure if the cross-selling tools are performing well or not. 11
  12. 12. We’ll switch from e-commerce websites to a slightly different category.Our company Dobry web organizes a large number of public trainings aboutdifferent areas of internet marketing. 12
  13. 13. On the training page there is an order form. Before you submit the order, youcan use a nice box for adding one or more training to the order.On the screenshot you can see that it is a training about Google Analytics andthe visitor is just clicking on a button called Pridat (Add) to add a trainingabout web copywriting. He will get a nice 15 % discount if he orders twotrainings at once.We are measuring every click on these Add buttons with Event Tracking inGoogle Analytics. 13
  14. 14. If you haven’t worked with Event tracking before, I strongly recommend youdoing that. It is very easy way to measure all interactions on your webpagethat are not related to pageviews.As we are tracking every click on Add button and even on the button calledOdebrat (Remove), we can see how many visitors have played with our cross-selling tool. But this doesn’t show us if the tool has helped our visitors to ordermore transactions. The real power of this data in Google Analytics lies in thecapability to be connected to visitor goals using the advanced segmentation. 14
  15. 15. These are very important figures. By using advanced segments we can clearlysee how many visits have used the cross-selling tool and how many of themhave actually purchased a training.Every fourth visit that used cross-selling has converted! It is remarkable!Now you can see how easy it is to measure the performance and efficiency ofcross-selling tools. I’m pretty sure you can manage to do it yourself. 15
  16. 16. What works for Amazon or your competitors doesn’t have to work for you. Tryanalyzing your own orders. It will take you just one hour and I think you’llhave fun as well.By doing so you will get the picture and see if your store has any potential tobe better at cross-selling.You can take an opportunity and make your Google Analytics perfect withEvent tracking.Please, don’t forget to connect your data to the real world. It is really beneficialto get some real feedback from real customers. You’ll justify if your cross-selling tools are appropriate for your customers. 16
  17. 17. Twitter: @paveljasekEmail: pavel.jasek@dobryweb.czFeel free to email me what have you observed in your data and how well isyour cross-selling doing.Thank you! 17
  18. 18. You can download test set of orders and categories: (CSV, 92 kB)Save this file as C:./categories.csvSample code for R:install.packages("arules");library("arules");txn = read.transactions(file="C:/categories.csv", rm.duplicates= FALSE,format="single",sep=";",cols =c(1,2));basket_rules <- apriori(txn,parameter = list(sup = 0.002, conf =0.06,target="rules"), appearance = list(default = "both"));inspect(basket_rules);You can play with sup and conf parameters to adjust support and confidencethreshold. 18