Expedia Business Analytics: Updated 2004 April 28 2005

2 Table of Contents Background and Objectives Critical Questions Methodology and Analytic Approach Analysis Checklist Summary of Results and Key Findings Next Steps Appendix

3 Project Objectives To develop a tighter, more robust understanding of the performance of individual channels AND the relationship between various marketing inputs and resulting business performance/outputs. How do various marketing inputs impact and influence business performance? How do various marketing inputs impact and influence each other? Within the above, understanding the impact and performance of our offline media efforts are of particular interest, especially given our inability to track this channel through more directly observable methods.

4 Desired Outcomes Findings from this effort will be used to: and monitor our performance over time . Suggest optimal marketing investment allocations both within and across marketing programs. Develop forward - looking modeling capabilities to allow for more accurate forecasting and business planning.

5 Critical Questions What are the specific effects of various marketing activities on overall transactions ? Understanding the relative performance of each activity at driving various metrics (reach, transactions, GP) independent of all other activity will be essential to get to our desired end goal of using these findings to help us optimize our marketing mix. overall and various marketing activity on new vs. repeat customer metrics ? Do different media have greater impact on driving existing customer behavior? Can we develop separate equations to predict and optimize driving new vs. repeat business?

Methodology

7 Segmentation Analysis Using large behavioral segments to determine the impact of overall and various marketing activities. Given the limitation on the data we have with regards to business and leisure, we are only able to analyze marketing impact on overall transactions and purchasers (new vs. repeat). Initial Plan: Four Square Segmentation Analysis New Customers Returning Customers Leisure 1 2 Business 3 4 Final Approach: Four Square Segmentation Analysis New Customers Returning Customers Leisure 1 2 Business

8 Data Inputs

9 Data Audit and Transformations Data Cleansing Identified data gaps and outliers. Attempted to run correlations and analysis on raw data corrupted/missing values and to eliminate seasonal spikes. Compared Donovan data to CMR data to confirm media GRPs Further analysis confirmed results were similar using Donovan or CMR. Compared Donovan data to media plans to verify accuracy. We are confident that data sets used closely depict Expedia and media activities during the period studied (05/2001 10/2004).

10 Data Transformations After Before Exponential smoothing of the data in this manner removes seasonality, day of week, holidays as well as random non - picture of business patterns. TOTAL TRANSACTIONS

11 Data Transformation: Daily vs. Monthly vs. Weekly Daily data was too granular to obtain correlations. Monthly data was too blunt for statistical analysis -- 42 data points could not adequately describe a truly multivariate business. Weekly data provided sufficient data points (185) to normalize seasonality and describe correlations between activities while allowing us to introduce multiple variables into equations.

12 Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions - Key Questions: 1). Is there a seasonal behavior? 2). What is the conversion behavior? 3). What is the correlation? 4). Is there a day of week behavior? Marketing Influence on Consumer Behavior - Key Questions: 1). How does each marketing activity (and various combinations thereof) affect transactions? 2). How do offline marketing activities impact the results of online marketing activities? Analysis Checklist

13 Consumer Behavior: New versus Repeat Purchasers - Key Questions: 1). What are the new vs. repeat purchaser trends? 2). What is the difference in transactions between new vs. repeat purchasers (enterprise)? Marketing Influence on New versus Repeat Purchasers - Key Question: 1). How does each marketing activities (and in combination) affect new vs. repeat purchasers? Analysis Checklist

Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions - Key Questions: 1). Is there a seasonal behavior? 2). What is the conversion behavior? 3). What is the correlation? 4). Is there a day of week behavior? Marketing Influence on Consumer Behavior Consumer Behavior: New versus Repeat Purchasers Marketing Influence on New versus Repeat Purchasers Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions

15 Seasonality: Overall Visitors/Shoppers/Confirmers 2002 and 2004 also show seasonal bump in Feb - Mar. Visitors Shoppers Confirmers Metric: People

16 Seasonality: Shoppers By LOB Observation: seasonality varies by LOB Hypothesis: we maybe able to exploit seasonal behaviors within specific LOBs Implication: test weighting the LOB messaging (advertising, offers, etc.) toward specific LOBs based on seasonal behavior. For example, - Jan/Feb weight towards packages - May/Jun/July weight towards hotel/car - Aug/Sept/Oct weight towards air - Nov/Dec weight towards hotel 2004 2003 Metric: People

17 The above chart shows a correlation between weekly transactions by LOB. The high level of correlation between air, car, hotel and packages indicates we can combine all LOBs for analyzing media impacts. Standard deviation measures the spread of data from the mean value. Therefore the higher the % of STDV from mean implies greater variability. Packages and hotel have the highest % of STDV from mean. Hypothesis: This suggest they are most likely to be influenced by outside factors (such as seasonality, Implication: Advertising plays a smaller role with these LOBs. Mean and standard deviation of weekly transactions Seasonality: Transactions By LOB Metric: Units

18 Conversion: Visitors/Shoppers/Confirmers At Enterprise Level Observation: For last two years, the conversion rates (visitors to shoppers, shoppers to confirmers and visitors to confirmers) have declined Question: Why are people coming to the site but not shopping? V:S V:C S:C Metric: People

19 Overall Shoppers/Confirmers (45 Day Snapshot) The observed examples show a slight difference in patterns between shoppers and confirmers. Hypothesis: Shopping process could last from a few days to one week for many customers. Implication: The window of opportunity to convert shoppers to confirmers could range from a few days to one week. During this timeframe, a series of personalized offers (based on items shopped) could be communicated through the web site and email to increase shopper/confirmer conversion rate. 45 Day Snapshot: Jan 1 Feb 15 2004 Examples where peak days for confirmers occur after the peak shopping days. An example where the does not coincide with that of confirmers. Shoppers Confirmers Metric: People

20 Conversion: Visitors to Shoppers By LOB Observation: Comparing 2003 to 2004, there is a significant declined in visitors converting into air shoppers. By contrast, visitors are converting at a higher rate to hotel. Question: Can we make ourselves more relevant to consumers for air shopping again? If so, how? Metric: People

21 Conversion: Shoppers to Confirmers By LOB Overall, the percent of visitors that did not shopped increased by 2%. Hypothesis: With a larger user base, customers maybe visiting in the post purchase phase. Implication: Determine the key reasons why ~20 - 22% of visitors do not shop and the activities they engage in. Identify opportunities to get those who visit and do not shop, to shop. For hotel, there is an increase in conversion from visitors to shoppers, however the conversion rate from shoppers to confirmers is declining. Indicating that more visitors are using the website for shopping and pricing comparison but fewer are confirming. For air and car, there is a decrease in conversion from visitors to shoppers, however, the conversion rate from shoppers to confirmers is increasing. Indicating that visitors who shop for air and car are more likely to convert. Hypothesis: Air and car consumers increasingly are realizing the parity across category. Implication: Need to make air on Expedia more relevant again to consumers. Conversion: Shoppers to Confirmers Conversion: Visitors to Shoppers Metric: People

22 Conversion: Searches to Transactions Overall and by LOB A shopper can have multiple searches and a confirmer can have multiple transactions. This chart shows the conversion of searches to transactions. The searches to transactions conversion rate is a metric that monitors at an activity level versus users level. Overall: Searches to transactions conversion rate varies based on convenience and price. To get the best deal or avoid a poor decision, consumers require more shopping as the perceived cost of an item increases - Perceived cost can be a combination of 1). Price of item, 2). Importance of the decision and 3). Perceived risk (example: advance payment). Searches/Transactions Data: 2003 - 2004 Metric: Units

23 Conversion vs. Correlation Rate at which one activity (action) leads to another Conversion Measures the degree through which the behavior of one activity can be predicted from another (i.e., when the behavior of one activity changes, the other is likely to make a corresponding change) - Suggests some causal relationship but not necessarily a direct cause and effect relationship. Correlation Guide to correlation value:

24 For all intents and purposes all five measures are equivalent - Visitor = Shoppers (= Searches) = Confirmers (= Transactions) Virtually identical correlations were seen when data was examined on a daily or weekly basis. Correlation: Overall Visitors / Shoppers / Confirmers / Searches / Transactions Visitors Shoppers Confirmers Correlation: 0.92 Correlation: 0.97 Correlation: 0.93 Searches Transactions Correlation: 0.94 Correlation: 0.93 Correlation: 0.92 Metric: People

25 Observation: The correlation value for all LOBs (except for packages) is lower than Enterprise, suggesting that: - At the Enterprise user level (regardless of LOB), shopping behavior is highly predictive of purchase behavior over time the conversion rate move in sync. - At LOB level some shopping behavior compared to other is less predictive of purchase pattern shopped for. Hypothesis: Shopping happens at an individual product level vs. bundling. Implication: Shopping and decision making process is independent by product line. Develop cross - sell strategy based on the initial product purchased for air/hotel/car. Correlation: Shoppers to Confirmers By LOB vs. Metric: People Data: 2001 - 2004

26 Observation: Across all LOBs, the correlation between searches and transactions is high. Which means searching behavior is a good predictor of purchase behavior. Correlation: Searches to Transactions Overall and by LOB Metric: Units Data: 2001 - 2004

27 Day of Week: Overall Visitors/Shoppers/Confirmers The chart above shows that the average number of visitors and shoppers are relatively constant Monday through Wednesday. This indicates that shopping takes place at the beginning of the week and decision making tends to take place towards the end of the week. Implications: - Vary offers based on day of week. Much more competitive offering early in the week to get more shoppers to purchase right away and less competitive offer at the end of the week when shoppers are more likely to decide. Metric: People Data: 2003 - 2004

28 Day of Week: Shoppers to Confirmers by LOB Packages is the only LOB that is not in sync with enterprise pattern. This suggests that it could be a Question: Should we examine shoppers and confirmers by time of day to determine if we should develop a media strategy plan around that behavior? Metric: People Data: 2003 - 2004

29 Day of Week: Overall Shoppers/Confirmers vs. Searches/Transactions Number of shoppers remains the same Mon though Wed, however, the number of searches declines steadily. Indicates that more searches per shopper are conducted at the beginning of the week and fewer towards the end. Implication: Early in the week when shoppers are more likely versus decision making, there are more opportunities to present and generate interest on various travel options through site merchandising. Conversion: Searches to Transactions Conversion: Shoppers to Confirmers Metric: People vs. Units Data: 2003 - 2004

30 All lines of business follow the same pattern of searches to transactions within a week, however Car and Hotel have a greater (in terms of percentage points) spike with Friday conversions. Packages show a huge spike on Friday and Saturday. Hypothesis: The huge spike on Friday could be due to an increase in purchase of last minute weekend travel packages. Need to further investigate the type of packages that have high conversion on Friday. Air Packages Car Hotel Day of Week: Searches to Transactions By LOB Metric: Units Data: 2003 - 2004

31 Lagged Time: Overall Searches to Transactions The chart above correlates transactions with prior searches. Findings indicate that there are two types of shoppers: - those who search at Expedia and transact within a day - those who search and transact within 1 - 2 weeks The peaks represent the weekly shopping pattern The depth of the valleys represents the extent to which people take their time to decide Implication: Segment our shoppers based on decision making window and a test series of communication messages to increase search to transaction conversion rate. j Time Lagged: Searches vs. Transactions 0 0.2 0.4 0.6 0.8 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Days from Searches Correlation: Transactions with Days from Searches Metric: Units

32 Car and Hotel have a shorter decision making time between searches and transactions than Air and Package Indicated by the fact that the moving average for Hotel/Car is always higher. Greater average daily correlation indicates that decisions are closely associated with more recent searches. Searches vs. Transactions Timing 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Days from Searches Correlation: Transactions to Day(s) from Searches Air/Package Hotel/Car 1 per.mov.avg.(air/package) 1 per.mov.avg.(hotel/car) Lagged Time: Searches to Transactions By LOB Metric: Units

33 Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions Analysis Checklist Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions - Key Questions: 1). Is there a seasonal behavior? 2). What is the conversion behavior? 3). What is the correlation? 4). Is there a day of week behavior? Summary of Key Findings and Implications:

Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions Marketing Influence on Consumer Behavior - Key Questions: 1). How does each marketing activity (and various combinations thereof) affect transactions? 2). How do offline marketing activities impact the results of online marketing activities? Consumer Behavior: New versus Repeat Purchasers Marketing Influence on New versus Repeat Purchasers Marketing Influence on Consumer Behavior

35 Online Marketing Activities Most of the online activity is Search Engine Keywords and Contextual Marketing Averaging 96% of total online spend and trending upwards. Source: Avenue A Metric: Dollars Spent Percent

36 Offline Marketing Activities Highest variability in general media is in broadcast (TV and Radio) absence of consistent presence in Spot TV combined with National TV to create an analyzable variable. Source: Donovan Metric: Dollars

37 Expedia vs. Competitors: Correlation between Marketing Spend and Website Unique Visitors Metric: Dollars/Units Notes: - The data source for this analysis is ComScore (Unique Visitors) and CMR Stradegy (Marketing Spend). Only monthly unique visit o r data is available. The date range for this data set is 2002 - 2004. A total of 36 data points is used for the analysis. - s is. - CMR marketing spend data is rate card information. CMR has limited tracking of online marketing spend (i.e., search engine ke y word spend is not tracked). In addition, CMR marketing spend will not reflect any value - added media allowances or partnership arrangements. Purpose: - a tion values to put some context to low versus high correlation. Observation: - Expedia and hotels.com show the highest correlation between marketing spend and number of unique visitors. Hypothesis: - Relatively speaking, a correlation value of 0.5 (in this category) is could be considered high. Implication: - r esults of our analysis how high is high?

38 Correlation: Marketing Activities And Transactions Overall and By LOB Online Marketing Activities: - Search engine keyword has the highest correlation with transactions. - can be used as a short term predictor to monitor overall transactions. - Utility (example: directories such as superpages) are sites where shoppers seek local travel agencies/hotels has a high correlation with transactions. Offline Marketing Activities: - Magazine has the highest correlation with transactions. - Broadcast and Magazine combined has a high correlation across all LOBs. - Spot TV lack of consistent presence (too few data points, only 22) to draw a conclusive read on the correlation value. Data: 01/2001 thru 10/2004 Metric: Dollars/Units

39 Correlation: 0.6 R²: 0.35 Regression formula: Transactions = 264,541+ 0.054*(Broadcast & Mag.) Every 18.38 dollar spent in offline marketing spend delivers one transaction. Inflection point is at about $1.5MM in weekly spend. Weeks with $1.5MM to $3MM are above the regression line. Offline Marketing Activities Influence on Transactions Metric: Dollars/Units

40 Correlation: 0.57 R²: 0.32 Offline Marketing Spend (Broadcast and Mag.) versus Gross Profit 0 5,000,000 10,000,000 15,000,000 20,000,000 25,000,000 30,000,000 $0 $1,000,000 $2,000,000 $3,000,000 $4,000,000 $5,000,000 $6,000,000 Offline Weekly Spend (Broadcast & Mag.) (Data: 01/2001 thru 06/2004) Gross Profit Observations Predictions Regression formula: Gross Profit = 6,592,733 + 2.33*(Broadcast & Mag.) Every 0.43 dollar spent in offline marketing spend delivers one dollar in gross profit. Offline Marketing Activities Influence on Gross Profit Correlation: 0.57 R²: 0.33 Metric: Dollars/Units

41 The chart above shows that offline marketing activities are predictive of transactions attributed to search engine keywords. Offline marketing activities are very likely to have an influence on search engine keyword performance. Hypothesis: Search engine keyword activities (attributed transactions) can be a metric to get an early read on the performance of offline marketing activities. Implication: Develop a robust predictive model to predict how the change in spend can influence the change in search engine keyword activities. Correlation: Between Offline Mktg Activities and Transactions Attributed to Online Mktg Activities Data: 05/2001 thru 10/2004 Correlation: Offline Mktg. Activities vs. Click Through Reservations (Transactions Attributed to Online Mktg. Activities) 0.53 0.54 0.32 0.24 0.04 (0.40) (0.25) (0.24) (0.05) (1.00) (0.80) (0.60) (0.40) (0.20) - 0.20 0.40 0.60 0.80 1.00 Online Total Search Engine Keywords Contextual Marketing Travel Pop Unders Utility Lifestyle / General Interest Local Motion Networks Metric: Dollars/Units

42 Behavior Between Offline Marketing Activities and Search Engine Keyword Clicks Offline marketing activities have a high correlation with transactions that can be attributed to keyword search engines. Further analysis (chart above) shows that: - Acceleration in key word activity is closely aligned with ramp up in broadcast spend in 2003 and 2004 among the possible explanations. - Drives more clicks based on top of mind recognition. - a phenomenon we have observed in several industries. Metric: Dollars/Units not saying anything new. Previous slide

43 Offline Marketing Activities Influence on Search Engine Keyword Clicks Correlation: 0.58 R²: 0.34 Regression formula: Search Engine Keyword Clicks = 957,280 + 0.533*(Broadcast & Mag) Every 1.88 dollar spent in offline delivers one search engine keyword click. Metric: Dollars/Units

44 Correlation: Overall Transactions Attributed vs. Unattributed to Search Engine Keywords Transactions attributed to search engine keywords versus non - attributed are closely aligned, indicating that keyword behavior (including click - throughs) may be able to be used to monitor and predict short - term business results. Correlation of search engine keyword attributed transactions with total transactions is better than 0.9. Metric: Dollars/Units

45 Marketing Influence on Consumer Behavior Analysis Checklist Marketing Influence on Consumer Behavior - Key Questions: 1). How does each marketing activity (and various combinations thereof) affect transactions? 2). How do offline marketing activities impact the results of online marketing activities? Summary of Key Findings/Implications:

Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions Marketing Influence on Consumer Behavior Consumer Behavior: New versus Repeat Purchasers - Key Questions: 1). What are the new versus repeat purchaser trends? 2). What is the difference in transactions between new versus repeat purchasers? Marketing Influence on New versus Repeat Purchasers Consumer Behavior: New versus Repeat Purchasers

47 New vs. Repeat Purchasers Data: Enterprise vs. LOB New vs. Repeat LOB When a customer in one line of business makes a purchase from LOB. Example: Customer A makes his first Expedia purchase in Air and then comes back and both purchases. New vs. Repeat Enterprise At an enterprise level, a purchaser can only be accounted Impact on analysis To ensure that New purchasers are not overstated and Repeat understated, the Enterprise level data is used for most of the purchasers analysis. Metric: People

48 New vs. Repeat Purchaser Trends: Enterprise Level Starting in Oct 2003, the weekly number of repeat purchasers exceeded that of new. The trend continues in 2004. Number of repeat purchasers continues to increase whereas there is a decline of new purchasers. New Purchasers Repeat Purchasers Metric: People

49 New vs. Repeat Purchaser Trends: Enterprise Level Also, the chart above shows the percent of new purchasers started to level out in 2003 and started to decline in 2004 Metric: People

50 Air Purchasers Trendline for New Purchasers y = 245.98x + 40699 Trendline for Repeat Purchasers y = 331.67x + 23119 - 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 100,000 2001 wk 1 2001 wk 21 2001 wk 41 2002 wk 8 2002 wk 28 2002 wk 48 2003 wk 15 2003 wk 35 2004 wk 2 2004 wk 22 2001 - 2004: By Wk Count of Purchasers New Repeat Linear (New) Linear (Repeat) New vs. Repeat Purchaser Trends: By LOB Air Identical to enterprise level new versus repeat behavior - online travel Both new and repeat purchasers show similar seasonality but the peaks do not line up - Hypothesis: New purchasers are booking further ahead than repeat. - Implication: Analyze the mean days between booking and travel for new vs. repeat. Lag time between peaks Repeat purchaser volume has passed new Yr: 2004 Yr: 2001 - 2004 Metric: People Air Purchasers Trendline for New Purchasers y = 245.98x + 40699 Trendline for Repeat Purchasers y = 331.67x + 23119 30,000 40,000 50,000 60,000 70,000 80,000 90,000 100,000 2004 wk 1 2004 wk 21 2001 - 2004: By Wk Count of Purchasers New Repeat Linear (New) Linear (Repeat)

51 Hotel Purchasers Trendline for New Purchasers y = 274.29x + 18821 Trendline for Repeat Purchasers y = 218.49x + 4456.8 0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 2001 wk 1 2001 wk 21 2001 wk 41 2002 wk 8 2002 wk 28 2002 wk 48 2003 wk 15 2003 wk 35 2004 wk 2 2004 wk 22 2001 - 2004: By Wk Count of Purchasers new repeat Linear (new) Linear (repeat) New vs. Repeat Purchaser Trends: By LOB Hotel High growth rate seemed to coincide with advertising/mktg./category focus on Hotel Different from enterprise level purchasers behavior. Number of new purchasers continues to exceed that of repeat. Yr: 2004 Yr: 2001 - 2004 Metric: People Hotel Purchasers Trendline for New Purchasers y = 274.29x + 18821 Trendline for Repeat Purchasers y = 218.49x + 4456.8 0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 2004 wk 1 2004 wk 21 Count of Purchasers new repeat Linear (new) Linear (repeat)

52 New vs. Repeat Purchaser Trends: By LOB Car Different from enterprise level purchasers behavior. Number of new purchasers continue to exceed that of repeat. Yr: 2004 Yr: 2001 - 2004 Metric: People Car Purchasers Trendline for New Purchasers y = 83.185x + 8453.9 Trendline for Repeat Purchasers y = 66.928x + 3361 0 5,000 10,000 15,000 20,000 25,000 30,000 2001 wk 1 2001 wk 21 2001 wk 41 2002 wk 8 2002 wk 28 2002 wk 48 2003 wk 15 2003 wk 35 2004 wk 2 2004 wk 22 2001 - 2004: By Wk Count of Purchasers new repeat Linear (new) Linear (repeat) Car Purchasers Trendline for New Purchasers y = 83.185x + 8453.9 Trendline for Repeat Purchasers y = 66.928x + 3361 0 5,000 10,000 15,000 20,000 25,000 30,000 2004 wk 1 2004 wk 21 Count of Purchasers new repeat Linear (new) Linear (repeat)

53 New vs. Repeat Purchaser: Total Transactions Per Purchaser Transactions per purchaser per week is 8% higher for repeat (1.34) than new (1.24). Repeat customers buy more. Hypothesis: - 8% may seem to be slight difference. But is the gross booking per transaction the same for new versus repeat? Implication: Further analysis to determine gross bookings per transaction (new vs. repeat). New Purchasers Repeat Purchasers Metric: People/Units

54 Hotel transactions per new purchaser is almost the same as repeat. Hypothesis: There is less parity with hotels, all customers (new and repeat) shop around. Implication: Focus retention program on hotel? Develop incentive programs to encourage/reward hotel shopping? New vs. Repeat Purchasers: LOB Transaction Per Purchaser Note: Packages data was not of the data feed. Car New Car Repeat Hotel Repeat Hotel New Air Repeat Air New Metric: People/Units

55 Analysis Checklist Consumer Behavior: New versus Repeat Purchasers - Key Questions: 1). What are the new vs. repeat purchaser trends? 2). What is the difference in transactions between new versus repeat purchasers? Summary of Key Findings/Implications: Consumer Behavior: Purchasers & Transactions

Consumer Behavior: Visitors to Shoppers to Confirmers AND Searches to Transactions Marketing Influence on Consumer Behavior Consumer Behavior: New versus Repeat Purchasers Marketing Influence on New versus Repeat Purchasers - Key Question: 1). How does each marketing activities (and in combination) affect new vs. repeat purchasers? Marketing Influence on New versus Repeat Purchasers

57 All offline marketing activities have a higher correlation with repeat purchasers than new. This suggest that offline marketing is effective at reminding previous purchasers to come back. Online marketing activities are not as consistent: - Pop - unders and travel sites seem to be effective at encouraging new purchasers to shop and buy with Expedia. Spot TV and Lifestyle/General Interest too few data points to draw a conclusive read on the correlation value. Question: Is there a way our advertising msg. could be more effective at acquiring new customers? Correlation: Enterprise Level Marketing Activities And Purchasers Metric: Dollars/People

58 Analysis Checklist Marketing Influence on New versus Repeat Purchasers - Key Question: 1). How does each marketing activities (and in combination) affect new vs. repeat purchasers? Summary of Key Findings/Implications: Marketing Influence on New vs. Repeat Purchasers

Next Steps Client to provide input on analysis and presentation Set up presentation to expanded team at Expedia Discuss frequency and timing of update to analytic template Bi - annual? 2 nd /4 th qtr or 1 st 3rd qtr Discuss weekly monitoring of 4 - to 6 - week trailing data (at least visitors and shoppers)

Appendix Glossary of Terms Definition of Statistical Terms

61 Glossary of Terms Offline Marketing Activities (Radio/Television/Magazine): Marketing spend for each media for each broadcast week. Online Marketing Activities (Banners/Keywords/Contextual/Pop - Unders): Marketing spend for each media by day. Examples for each category:

62 Glossary of Terms Total Searches (Expedia): Search results that are not attributed to any marketing activities. Searches Attributed (Ave A): This data represents raw action tag hits that can be attributed to online marketing campaigns. Searches Unattributed (Ave A): This data represents raw action tag hits. It accounts for any time one of the action tags are hit, regardless of whether or not it was attributed to any marketing efforts. Transactions: Daily transactions numbers pulled using the Expedia dashboard cube user interface. Search Engine Keyword Clicks: Number of clicks generated from each search engine keyword placement. Online Click Reservations: Reservations that can be attributed to click throughs from online marketing activities. Enterprise Level New/Repeat Purchasers: Number of customers new/repeat to Expedia. Transactions By New/Repeat Purchasers: Number of transactions by new/repeat purchasers who were new to Expedia.

63 Glossary of Terms Shoppers/Visitors/Confirmers: - All data is based on page views. So for confirmations for instance, it could be overstated if someone looked at their confirmation page twice - Shoppers are defined as anyone who saw search results. - Visitors are based on cookies so if someone deletes their cookies, they will be counted as another visitor, so visitors could be overstated.

64 Correlation is the degree to which two variables travel together. Positive correlation means that both variables increase at the same time at a similar rate. Negative correlation means that one variable goes down while the other goes up at the same time at a similar rate. Correlation are displayed with a value ranging from - 1.0 to 1.0. A 1.0 correlation indicates perfect positive relationship and - 1.0 indicates a perfect negative relationship. Statistical Term: Correlation Definition Direct Correlation 87.9% Indirect Correlation - 89.7% Typically, a minimum of 25 data points are required to calculate a reliable correlation. Correlation in no way implies causality. Two variables with a high observed correlation could both be affected by a third variable equally.

65 Regression can be defined as a method that estimates the value of one variable when that of other variable is known. Regression formula: y = mx + b x is the dependent variable y is the independent variable The equation of linear regression is to find the one line draw through a XY graph of two variables such that the average distance of the observe variables from that line is minimized. r 2 , is the square of the correlation between the two variables thus if two variables have a correlation of 0.5, the r 2 of the linear regression equation for those two variables will be 0.25. The closer an r 2 is to 1.0, the better the x variable is a t predicting the total variance of the y variable. Statistical Term: Regression Formula

66 The standard deviation is kind of the "mean of the mean." The standard deviation is a measure of how spread out your data is. It tells how tightly a set of values is clustered around the average of those same values. It's a measure of dispersal, or variation, in a group of numbers. Statistical Term: Standard Deviation The following charts will help visualize the concept. Standard deviation indicates: 1). how good is the mean at describing the closeness of the data points and 2). how much inherent variability is there in the data set. Higher standard deviation is often interpreted as higher volatility. In comparison, lower standard deviation would likely be an indicator of stability. In business, greater variability suggests opportunity to influence outcomes. Temperature by Month High Standard Deviation Average Height of NBA Point Guards Low Standard Deviation # of Players