Contenu connexe Similaire à Getting Started in Big Data-Fueled E-Commerce (20) Getting Started in Big Data-Fueled E-Commerce1. Getting Started in Big Data
Fueled E-Commerce
European Outdoor Summit
Stockholm, 17 October 2013
jason.radisson@echtzeit.net
This document contains confidential material and ideas proprietary to Echtzeit GmbH. This document may not be reproduced in any form or by any
means or disclosed to others or used for purposes other than for this discussion. It may not be disclosed to any third party even for the purposes of
evaluation, except as expressly authorized by Echtzeit GmbH in advance for each case. This document is intended to be delivered orally and does
not represent a complete record of that discussion.
2. E-commerce is entering another wave of disruption, Big
Data is a driver
Universalists and market-places
USP
1. Same-day
delivery
Food/Non-Food
Price, broad selection,
convenience
2. Big data
Category specialists
Sporting
Goods
Curation: i.e., deep
selection, specific
expertise and service
HABA
Brands
Pharmacy
Cosmetics
Brands
Outdoor
Fitness
Brands
Electronics
DIY
3. Proliferation of
incubators
Furnishings
Brands
Artisan
Flash
Ethnic
Organics
Fresh
OEMs and niche retailers
Electronics &
Media
Video & TV
Grocery
Home &
Garden
Garden
Clothing,
Shoes &
Accessories
Merchandize and
business model
innovation in the long-tail
1. With push to same-day delivery aggregators and grocery players will take on the universalists and market-places in 2014-15.
Everyone will be looking to the long-tail items to subsidize increasing fulfillment costs.
2. Big Data advantages universalists who can aggregate long-tail demand. They’ll use it to move into curation.
3. Increasing disruption to traditional consumer businesses drives demand for innovation, leading to a spike in e-commerce
incubation activities. Lots of venture funding chasing similar concepts (project a, 7 ventures, rocket internet, REWE ventures, etc.)
> Everyone, especially category specialists and brands, will need to step up as gross margins get squeezed, even as OPEX rises
1
3. We define Big Data as smart applications that create a
sustainable competitive advantage in e-commerce
Desired business
outcome (e.g., 7d
Marketing ROI)
Big Data:
• Is the umbrella term for a class of business applications that learn from millions
of interactions and automatically adjust to customer intent and market context
• Key Big Data apps in e-commerce: 1) SEM/O, 2) Loyalty & Onsite
Merchandizing, 3) Dynamic Pricing/Offers, 4) Inventory & Fulfillment
• Creates a break-away competitive advantage as more visits = more split testing
volume = smarter systems & teams = more demand/visits
(…)
2. ‘Cold-start’
phase and initial
semi-automatic
optimizations
3. ‚Hill-climbing‘ phase of a Big Data application implementation
1. Manual processes
and ‘gut’ decisions
time
© Echtzeit GmbH 2013, all rights reserved
2
4. Valuation of the Big Data opportunity in e-commerce is
straightforward, the mechanics are well understood
Percent of total online sales driven by run-time applications in algorithmic
merchandizing and customer marketing, in %**
35
7-30
2-3
Assume ecommerce
business of €400
to 500M p.a. in
turnover and
10% incremental
sales (lift) for
early-stage Big
Data implementation
Potential for €4050M in
incremental
sales for an midsize e-commerce
division
*
12-15
100+
2-3
Number of loyalty/merch
strategies in portfolio
*eBay range from average business day (7%) to peak holiday shopping season day, such as Cyber Monday (30%)
** To convert percent of total online sales (PTOS) to lift use LIFT = PTOS/ (1-PTOS)
Source: Amazon numbers published in HBS case ‘eBay Inc. and Amazon.com’ from 3 April 2012; eBa, SportScheck estimated
© Echtzeit GmbH 2013, all rights reserved
3
5. But, there are several challenges with executing a Big Data
strategy in Europe … motivated us to found a new company
Incumbent perspective in DACH
Companies need Big Data
applications and processes and
can’t readily build/buy them
5. Open-source systems require specific data-science and
engineering skills. EU has yet to build talent pool
4. Legacy infrastructure scales expensively and slowly (2-3 year
cycles). Can’t keep up with data-volumes or open-source innovation
3. For data-privacy, time-zone and cultural reasons, it is easier to
do business with local partners, rather than Silicon Valley startups
2. Marketers have a tougher time employing playbooks and hillclimbing strategies (less traffic and MVT knowhow)
1. Big Data applications require automating business processes and
an IT product-focus at board level. Change-resistance is a factor
© Echtzeit GmbH 2013, all rights reserved
4
6. Establishing a fact-basis early-on helps to validate the
opportunity and clear out internal change hurdles
Our agile implementation model
1. Establish facts and quantify
latent opportunity
(2-3 months)
2. Pilot the application and
playbook of strategies
3. Deploy application at scale
(4-6 months)
(4-6 months)
Build application and playbook of successful strategies
Run
Audits of customer base and
item catalog performance
Initial playbook of algorithmic
‘strategies’
Optional: overhaul of metrics,
descriptive segmentations
(e.g., CLV, psychographics)
Piloting algorithmic strategies
with Tiger Team. Develop the
application
MVT reporting, including ROI
and other causal metrics
Implement performance
management process d/w/m
Deploy the application with
initial champion portfolio
Realign resources on testing
challengers
Automate causal
performance reporting
Quantify opportunity gap
and establish fact basis for
change
Demonstrate effectiveness of
Big Data approach vs.
business as usual
Achieve scale via automation.
Realign processes.
Continually improve
© Echtzeit GmbH 2013, all rights reserved
5
7. Focus your digital marketing efforts first on the frequency
upside sweet spot. This is where you can drive ROI at scale
14d Lift in %
Response in %
250%
25%
200%
20%
150%
15%
100%
10%
50%
5%
0%
0%
b. ONE
PURCHASE
c. TWO OR 3
PURCHASES
d. FOUR TO
11
PURCHASES
e. 12 TO 49
PURCHASES
f. 50 TO 149
PURCHASES
g. 150 TO 349
PURCHASES
h. MORE
THAN 350
MerchLift
120%
83%
58%
22%
4%
2%
-16%
LoyaltyLift
205%
129%
73%
45%
26%
38%
140%
MerchResponse
2%
4%
6%
11%
17%
20%
22%
LoyaltyResponse
3%
5%
8%
14%
20%
20%
16%
•
In general, your actions will be most effective in the sweet spot of frequency* upside.
•
Specifically, your strategies will speak to discrete opportunities in loyalty and
merchandizing. For this category, there should be about 30-40 maximum.
* Recency R is an accelerator, M monitization is almost a constant for a given consumer’s wallet
© Echtzeit GmbH 2013, all rights reserved
6
8. We believe a Tiger Team drawing from Business, Data
Science and Infrastructure is best
Organizational model for building and implementing any Big Data application in e-commerce
Business
• Own the results
• Generate and prioritize
hypotheses (‘challenger’
strategies) to maximize
long-run returns from
the Big Data portfolio
4-5 from Business
Planning and
Campaign Ops
Application
Development
Infrastructure
Operations
8-10 Engineers*
2-3 Engineers
• Build and maintain the
application plus the algorithms
and data that power it
• Build and maintain APIs
• Generate datasets for BI
• Build and maintain scalable
infrastructure (run-time and
backhaul) at 5-9s uptime
• Deployment of applications and
updates
* New engineers typically from either Computer or Data Science track and will need to be trained on any gaps during first year
© Echtzeit GmbH 2013, all rights reserved
7
9. At SportScheck we built a recommendations application to
mitigate cart abandonment in real-time as a first step
Challenge
• SportScheck is Germany’s leading sporting goods
retailer with ca. €500m in revenues, 60M online
visits, and 20M visits p.a. to its 16 physical stores.
• The online business is growing steadily at some
5-10% p.a. But compared to Amazon, with 2030% CAGR in EU, there is a significant
opportunity gap.
• SportScheck’s customer marketing,
merchandizing and post-sales processes are
otherwise largely manual.
Approach
• We selected a white-space business opportunity,
mitigating cart abandonment (ca. 50% incidence
with no pre-existing treatment), as first use-case,
and worked in a cross-functional Tiger Team.
• Listening began in May and the system went live
in early July.
• We implemented our pixel, began logging clickstream data and training our models.
• We went live with a minimal implementation (2-3
simple real-time strategies) on the homepage.
© Echtzeit GmbH 2013, all rights reserved
CASE STUDY
Results
• It's early days and the work
shows great promise. We are
generating a couple percent
lift in conversion-rate & sales
• We are implementing several
enhancements which will livetest in Q4 (additional algos,
offers and placements). Each
improvement will generate 50100 basis points in
incremental conversion.
8
10. Takeaways (by page number)
APPENDIX
1. The point of Big Data in e-commerce is to unlock the Long Tail and enable competition on price, selection
and convenience USPs. Current industry dynamics and competitive forces – for example, same-day
delivery, Big Data, proliferation of incubators -- are such that the middle market will continue to be
squeezed.
2. Big Data is the key enabler for category specialists to compete in this and the next wave of e-commerce.
The way to see it is as a set of smart systems that learn from interactions with millions of customers and
automate your core business processes. There are four classes of applications: 1) SEM/-O, 2) Loyalty &
Onsite Merchandizing 3) Dynamic Pricing/Offers, 4) Inventory/Fulfillment
3. There is a huge opportunity in getting this right: initially a 10% improvement in top-line.
4. The main challenge is, most companies can’t staff this capability alone and suppliers aren’t set up yet to
help. By the time they are it may be too late. Change hurdles are non-trivial.
5. Best practice is an agile and interdisciplinary ‘Tiger Team’ approach for getting started in Big Data fueled ecommerce. First you audit, then you pilot, last you scale/automate.
6. In general, a great high-ROI first target for your first Big Data pilots as ‘frequency upside’ segment. SEM/-O
fills this bucket with high potentials and specific strategies are selected from
loyalty, pricing, merchandizing, fulfillment, etc. in real time to migrate these customers to higher frequency
levels and keep them there.
7. Team should be staffed with a triad of a) business, b) application development and c) infrastructure
engineering. Most important hire is the application development engineering lead. Locate this team where
ever s/he has best access to raw talent. A more cautious approach is to first rent/buy another company’s
work, thus establishing a baseline for how much value a given Big Data application can add at your
company. Negotiate a minimum performance level with any application provider to ensure self-funding.
8. We’re having success at SportScheck, where each ‘strategy’ equals a 1% improvement in site revenue
© Echtzeit GmbH 2013, all rights reserved
9
12. Frontends
It’s possible to incorporate an open-source Big Data
platform into a corporate IT landscape
APPENDIX
Enterprise CRM
• Unica, Aprimo
• Sugar
• Salesforce …
•
•
•
Enterprise BI
Microstrategy
Cognos, BO
Tableau …
Live
Web-shops
E-Mail
Mobile Apps
SEM / Ads
Social
Agents/ Call
Centers
Data
sources
Production File Real-Time
System
Applications
Service Bus
Merch/Re
co
Loyalty &
Offers
Pricing
Optimizer
SEM
SEO
Customer
Authentication
Item/
Customer
tables
BI Data
Cubes/
MDX
Run-Time Applications (e.g., Couch, Hbase)
Core DWH
(e.g., MPP-database on commodity
hardware)
Backhaul processing (MapReduce, Mahout, job management
framework)
Hadoop File System (HDFS)
AppDev
•
•
•
•
•
Maven
Hive
R
PIG
Mahout
ETL (Runtime & bulk)
Reference /
Master Data
Click-Stream
Monitoring
Social Media
© Echtzeit GmbH 2013, all rights reserved
Machine
(Server) Logs
Marketing
Outcomes
On-Device
collecting
Offline
Channel Data
Billing &
Payments
11
13. Bio
APPENDIX
• I’m based in Munich and founded Echtzeit (means ‘real-time’ in
German) GmbH about a year ago to build Big Data applications for
several of Germany’s largest consumer companies, including
SportScheck, on the open-source Hadoop technologies.
• My first paid Big Data job was as a teenager, programming text-mining
algorithms (e.g., classification, similarity) in Opposition Research for the
winning side in the 1990 Massachusetts gubernatorial race.
• Based on my Tiger Team work on the eBay turnaround and as a
McKinsey consultant I’m also frequently an advisor on digital
transformations, bridging Silicon Valley technological innovation and
European corporate culture of my clients.
© Echtzeit GmbH 2013, all rights reserved
12
Notes de l'éditeur The main challenge is, most companies can’t staff this capability alone and suppliers aren’t set up yet to help. By the time they are it may be too late. Change hurdles are non-trivial.