This document discusses user journey analysis for online marketing. It presents several approaches for analyzing user journey data, including sliding time windows, hand-crafted features, higher-order Markov models, and time series analysis. The objective is to inform micro-level decisions like digital ad bidding. Features are generated from online user behavior data within time windows. Models account for offline channels like TV and model awareness decay. Bayesian statistics are used to estimate models, which are optimized based on costs and benefits.
User Journey Analysis - Micro Decisions and Budget Allocation
1. User Journey Analysis –
Between Budget Allocation and Micro Decisions
Meetup - Data Science@Zalando
Prof. Dr. Burkhardt Funk
Institute of Information Systems
Leuphana University
March 30, 2017
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 1 / 30
2. Selected Approaches to analyze User Journey Data
Learning from UJ data doesn’t fit the classical supervised learning
paradigm (M¨orchen, 2006). Learning options include:
Sliding time windows can be applied to convert problem into
supervised learning problem (Dietterich, 2002)
Hand-crafted features have been used in statistical models in
E-Commerce scenarios (Chatterjee et al., 2003)
Higher order Markov models describe problem as a state machine;
here, states refer to user actions (Anderl et al., 2014)
Time series analysis and state space models (Nottorf and Funk,
2013a)
Context specific heuristics are used to derive management
recommendations (Funk, 2014)
Further references (Shao and Li, 2011; Dalessandro et al., 2012;
Abhishek et al., 2012; Li and Kannan, 2014) ... and the list goes on
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 10 / 30
3. Objective
What do we want to use the model for?
Find most efficient online channels on an aggregated level: Still
top-prio in marketing management
Inform decision-making on a micro level, e.g. decide whether and
what to bid in programmatic advertising
Let’s assume it’s the latter: all clear now? No, e.g. conversion/visit
and time frame
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 12 / 30
4. Generating Features
A
A
V
V A
ST O
S
t
Ad view
delivered
User seen in
RTA auction
User saw
TV ad
Paid search
contact
User visits
website
𝒗𝒊𝒔𝒕 0 00 0 0 0 1 1 1
30min Window
1 22 2 0 1 1 1 1𝑋𝑖𝑠𝑡
𝐴
Session 1 Session 2
0 20 1 0 0 0 0 0𝑋𝑖𝑠𝑡
𝑉
0 00 0 1 1 1 1 1𝑋𝑖𝑠𝑡
𝑇𝑉|𝑉
…
1 22 2 2 3 3 3 3𝑌𝑖𝑠𝑡
𝐴
…
…
Features
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 14 / 30
5. Accounting for Offline Channels
p(outcome|observation) =
1
1 + e−z
z = αXSEA + βXDISP + γXV ID+δXTV
Probability incl. TV
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 15 / 30
6. Awareness Decay
The awareness decay can be modelled by ”Γ-distribution” like shape
XTV ∼ tα−1
e−t/β
Model estimation is tricky, but MCMC techniques can help (β ∼ 0.1
and α ∼ 0.2)
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 16 / 30
7. Model Estimation
p(vist|obsist) = f(Xist, Yist)
αi ∼ N(µi, σ2
i )
µi ∼ N(νgi , σ2
gi
)
gi ∼ B(θ(obsi))
. . .
The model is estimated using techniques from Bayesian Statistics,
i.e. MCMC sampling (implementation with JAGS/ Stan)
Why Bayesian Stats? (i) accounting for prior knowledge, (ii) high
flexibility, (iii) extensive diagnostics
But: computationally demanding
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 17 / 30
9. Bibliography I
Abhishek, V., Fader, P., and Hosanagar, K. (2012). Media exposure through the funnel: A
model of multi-stage attribution.
Anderl, E., Becker, I., Wangenheim, F. V., and Schumann, J. H. (2014). Mapping the
customer journey: A graph-based framework for online attribution modeling. SSRN.
Bremer, V. and Funk, B. (2017). Analysing clickstream data: do paid and organic search
affect each other? International Journal of Electronic Business, 13(2):205–215.
Chatterjee, P., Hoffman, D. L., and Novak, T. P. (2003). Modeling the clickstream:
Implications for web-based advertising efforts. Marketing Science, 22(4):520–541.
Dalessandro, B., Perlich, C., Stitelman, O., and Provost, F. (2012). Causally motivated
attribution for online advertising. In Proceedings of the Sixth International Workshop on
Data Mining for Online Advertising and Internet Economy, page 7. ACM.
Dietterich, T. G. (2002). Machine learning for sequential data: A review. In Joint IAPR
International Workshops on Statistical Techniques in Pattern Recognition (SPR) and
Structural and Syntactic Pattern Recognition (SSPR), pages 15–30. Springer.
Funk, B. (2014). Kanal¨ubergreifende werbewirkungsanalyse in echtzeit. In Realtime
Advertising, pages 41–51. Springer.
Hastie, T., Tibshirani, R., and Friedman, J. (2001). The elements of statistical learning.
Springer series in statistics Springer, Berlin.
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 29 / 30
10. Bibliography II
Li, H. and Kannan, P. (2014). Attributing conversions in a multichannel online marketing
environment: An empirical model and a field experiment. Journal of Marketing
Research, 51(1):40–56.
M¨orchen, F. (2006). Time series knowledge mining. PhD Thesis - University of Marburg.
Nottorf, F. and Funk, B. (2013a). A cross-industry analysis of the spillover effect in paid
search advertising. Electronic Markets, 23(3):205–216.
Nottorf, F. and Funk, B. (2013b). The economic value of clickstream data from an
advertiser’s perspective. In European Conference on Information Systems.
Shao, X. and Li, L. (2011). Data-driven multi-touch attribution models. In Proceedings of
the 17th ACM SIGKDD international conference on Knowledge discovery and data
mining, pages 258–264. ACM.
Stange, M. and Funk, B. (2014). Real-time-advertising. Business & Information Systems
Engineering, 56(5):305–388.
Stange, M. and Funk, B. (2016a). How big does big data need to be? In Enterprise Big
Data Engineering, Analytics, and Management, pages 1–12.
Stange, M. and Funk, B. (2016b). Predicting online user behavior based on real-time
advertising data. In Proceedings of the European Conference on Information Systems.
Prof. Dr. Burkhardt Funk (IIS@Leuphana) User Journey Analysis March 30, 2017 30 / 30