Brace for impact?
The impact of external negative news on the customer purchase journey within the travel industry.
Marc van Eck
MSc Marketing Intelligence Thesis defense
0
Introduction
Customer well-being
More and more important for firms to make the customers’
experience excellent.
Customer purchase journey
Part of the experience. Internal touchpoints are well investigated.
Externalities are known to exist but little research investigate them.
Context: travel industry
This thesis investigates the customer purchase journey within the travel industry.
Negative news
External to the customer journey.
Negative news will be investigated
in how it has impact on the journey.
Research questions
Main question
What is the impact of negative news (plane crashes) on the customer purchase journey in a travel industry context?
Sub question one
What is the impact of negative news on the attribution of customer journey touchpoints?
Sub question two
How enduring is the effect of negative news on the customer journey?
Literature
1. Firms are not isolated
According to Tax, McCutcheon & Wilkinson (2013, p. 461).
Research by Lemon & Verhoef (2016, pp. 78–79) proves that past experiences can influence current experiences
2. News as externality lacks research
Research that exists focusses for example on the effect of news on stock volatility (Conrad et al., 2002; Eagle & Victor, 2013)
3. Risk aversion
Possible explanation for altered behavior when crises happen is risk aversion (Eeckhoudt & Hammit, 2004;
Kahneman & Tversky, 1984; Pandey & Natwani, 2004). 4. Customer journey stages
Focus in this research is on the pre-purchase stage and purchase stage (Lemon & Verhoef, 2016).
5. Touchpoints
Touchpoints are categorized in two different types: the firm-
initiated touchpoint (FIT) and the customer-initiated
touchpoint (CIT) (Wiesel, Pauwel & Arts, 2011).
4
Conceptual model
FITs
Purchase CITs
Pre-purchase stage Purchase stage
Negative news External of the
customer journey
Internal of the customer journey
Time
(H1, H4)
-
-
(H1, H4, H5)Touchpoints
Non-linear-(H3)
-
(H2)Research design
• Event-based, online data from Dutch travel agencies collected by GfK à internal data
• Google Trend data as proxy for the impact of negative news à external data
• No missing data. Tested for outliers, one journey excluded
• Dataset is transformed from longitudinal panel data to time-series data
6
Research design
• Vector Autoregression (VAR) models to investigate impact of shocks
• VAR assumptions: non-stationarity &
cointegration, optimal lag length and non- autocorrelation
• Visual outcomes with IRF plots
• Bayesian Causal Impact (BCI) model as controlling technique
• Does not use external data, models synthetic counterfactual instead
𝑦 " = 𝛼 + 𝛽 ' 𝑥 '" + 𝛽 ) 𝑥 )" + 𝛽 * 𝑥 *" + 𝛽 + 𝑥 +" + 𝛽 , 𝑥 ," + 𝜀 "
𝑌 " = Α + 0
12',..,5 5
𝛽 61 𝑌 "71 + 0
12',…,5 5
𝛿 1 𝑅𝑆𝐼 "71 + 𝜀 "
𝑡 = 1, 2, … , 𝑇
𝑌 " =
𝑃𝑂 "
𝑃𝐶 "
𝑆𝐹 "
𝑆𝐶 "
𝐴𝐹 "
𝐴𝐶 "
Where for time 𝑡,
𝑃𝑂
"= amount of purchases (own, focal brand)
𝑃𝐶
"= amount of purchases (competitor brands)
𝑆𝐹
"= amount of FITs
𝑆𝐶
"= amount of CITs
𝐴𝐹
"= attribution of FITs
𝐴𝐶
"= attribution of CITs
Results: VAR
• Dickey-Fuller test: multiple variables show non-stationarity à solved by taking the first difference
• Optimal lag length is defined at (𝑝 = 6), but model has autocorrelation (p < .05)
• VAR (𝑝 = 8) has no autocorrelation and better AIC (table 5)
• All VAR equations are significant (table 6), no differences found between attribution methods
à Relatively little variance explained for RSI and sales of focal brand
à Sales of competitor brands, Amount of FITs and Markov Attribution of FITs are explained considerably well
Table 5. Edgerton-Shukur F test and AIC scores
Model (lag) Edgerton-Shukur F test p-value AIC
VAR (! = 1) 3.5898 .000*** 45,241.11
VAR (! = 2) 2.5244 .000*** 44,984.61
VAR (! = 3) 2.3351 .000*** 44,836.57
VAR (! = 4) 2.1739 .000*** 44,751.5
VAR (! = 5) 1.8081 .000*** 44,631.21
VAR (! = 6) 1.308 .001** 44,441.21
VAR (! = 7) 1.1609 .049** 44,364.58
VAR (! = 8) 1.1541 .057* 44,296.71
* p<.1, ** p<.05, *** p<.01
Table 6. Estimation results per equation for VAR (! = 8)
Equation y-variable R
2Adj. R
2F-statistic p-value
RSI .374 .297 4.856 .000***
POS competitor .852 .834 47 .000***
POS focal brand .344 .264 4.271 .000***
Amount of CITs
1.546 .489 9.776 .000***
Amount of FITs .863 .846 51.08 .000***
Markov Attribution of CITs
1.467 .402 7.147 .000***
Markov Attribution of FITs .789 .763 30.47 .000***
* p<.1, ** p<.05, *** p<.01 |
1First difference variable
8
Results H 1
• Decrease expected in amount of touchpoints in pre-purchase, not found in IRF plots à H 1 not supported
• Amount of CITs decreases over time, but Amount of FITs shows a small positive trend on long-term
à H 3 only fully supported for Amount of CITs
Results H 2
• Decrease expected in amount of sales in purchase stage, not found in IRF plots à H 2 not supported
• Small positive trends on long-term
à H 3 therefore not supported for these
variables
Results H 4
• Attribution for both FITs and CITs will decrease as a result of a shock, only found for CITs à H 4 partially supported
• Attribution of CITs decreases to almost zero over time, but Attribution of FITs shows a small positive trend on long-term à H 3 only partially supported for Attribution of CITs, not for FITs
Results H 5
• CITs in the pre-purchase stage will decrease more severely in attribution effectiveness as a result of a shock than FITs,
found in IRF plots à H 5 fully supported
10
Results: Bayesian Causal Impact
• No significant outcomes
Table 8. Bayesian Causal Impact model outcomes Model response
variable Observed
response value
2Predicted response
value
2Absolute effect
2p-value POS competitor 6.3 (319.0) 6.2 (316.2) 0.054 (2.779) .470 POS focal brand 0.31 (16.0) 0.32 (16.45) -0.009 (-0.448) .480 Amount of CITs
12.4 (122.0) -3.6 (-183.5) 6 (305) .477 Amount of FITs 46 (2332) 78 (3989) -32 (-1657) .331 Attribution of CITs
13,179 (162,128) -609 (-31,079) 3,788 (193,206) .480 Attribution of FITs 2,301 (117,354) 2,615 (133,379) -314 (-16,025) .414
1
First difference variable |
2average value (cumulative value)
Discussion
• Impact of shocks on attribution is different for CITs and FITs
• Endurance of the effect of a shock is variable dependent. On average:
• There is a significant impact of negative news on the customer journey:
CITs
Short term Long term
FITs
Short term Long term
(Mortality) risk aversion? Customers happy to view presented FITs in first days.
Positive contribution to conversion as attribution increases
15.2 days till stable and calm
(min. 10 days, max. 20 days)
11 days intense period
(min. 8 days, max. 15 days)
Variable specific Increase in perceived societal risks (Janic, 2000)
has a shorter effect than expected
Indeed an effect on population as a result of shock events, in line with Von Dem Knesebeck (2015)
Altered behavior is very volatile following Eagle & Victor (2013) but short in duration
𝑥
12