3. Apple Stores have more than 2x
Sales/Square Foot than their nearest
competitor.
(source RetailSails:
http://www.retailsails.com.php53-12.dfw1-1.websitetestlink.com/site-
4. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
* Note, this isn’t necessarily the lowest price
5. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
Given our product set, which products are customers
demonstrating the most interest in? Which ones are they likely
to be interested in next season?
6. Provide me products I want…
Historical
Product
Sales Analytics Informed
Merchandising
Targeted Upsell in
Customer Store
Demographics
Targeted Offers Online
Customer
Research Targeted Social Media
Advertising
Social
Media
7. “…..he was able to identify about 25
products that, when analyzed
together, allowed him to assign
each shopper a “pregnancy
prediction” score. More
important, he could also estimate
her due date to within a small
window, so Target could send
coupons timed to very specific
stages of her pregnancy.”
8. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
Given our customer buying patterns, demographics, and
migration patterns, what are the best locations for our retail
locations? Should we offer different types of retail locations
oriented at different types of buyers?
9. … at a place convenient to me …
Purchasing Retail Location
Patterns Optimization
Store Differentiation
(i.e. Walgreens)
Migratory
Patterns Optimization of Product
Mix per Retail Location
Offline/online Targeted Physical Print
purchasing Advertising
trends
Mobile Advertising
Social
Media
10. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
How can I arrange layout of products that customers want?
How can I do so in a way that maximizes the likelihood that
customers will purchase higher margin products?
11. … where products easy to find…
Video capture of Heat map of which
in-store shopping square meters have
behavior highest rev/margin
Further insight into
customer preferences
around product
Offline/online
purchasing Insight into how to
trends position products in
specific stores
Insight into what to
Audio analysis of offer people online after
what people say an offline visit
about products in
store
12. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
Are customers having negative experiences in stores? Can we
analyze comments in reviews of selected locations to know
whether our customers are getting the service they expect?
13. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
Can we better predict what customers want in the store based
on what they have browsed for online? How about offering
them things online that people like them have looked at or
purchased in the store?
14. … friendly people who anticipate my
needs…
Video capture of
facial Greater understanding
expressions/emot salespeople’s non-
ion of staff verbal communication
skills
Insight into what
Social media communication modes
analysis of sell what products
good/bad
experiences
15. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
Can we adjust hours and sales associate schedules based on
predicted traffic flow? Based on level of activity our in-store
cameras manage to pick up?
16. … at a time convenient to me …
Sales by hour
Further insight into
trends over time
what business hours
for which locations
Insight into what
Online people tend to plan as
purchases purchases versus
(planned) v impulse purchase
offline (impulse)
17. What Makes Me Likely to Buy?
Provide me products that improve my life…
…found in a place convenient to me
…in a store where the products are easy to find
…where they are provided by friendly people
…who are able to anticipate my needs
…at a time convenient for me
…for a price I am willing to pay*.
Can we quickly adjust pricing based on convenience,
scarcity/abundance, or demographics in order to optimize
margin? Can we predict what a given type of customer will pay
in a more sophisticated way?
18. … for a price I am willing to pay.
Real time
inventory levels
Further per store
per store
pricing optimization
Supply chain
adjustments
Buyer’s ability to
pay
20. What is Agile Analytics?
Agile Analytics is the application of data science…
…to pressing business questions
…which are predictive in nature
…where solutions are usually not obvious
…involving data that is often diverse, messy, and high volume
…where feedback lends itself to continuous improvement
…for which answers have significant business impact.
21. What is Agile Analytics Not?
Data Warehouses
Consolidate data, get “one true version” of
the truth.
Business Intelligence
Drive reports from data. Allow users to explore
data and drive their own reports and needs.
Good at describing the past, but inadequate for
predicting the future.
Analytics
Using advanced maths, statistics, machine
learning, monte-carlo simulation, and other
advanced techniques to drive insight from
data.
22. What is a Data Scientist
Like many popular buzzwords, “data scientist” is already becoming
diluted. When ThoughtWorks uses the label Data Scientist, we are
describing someone with at least three of these qualities:
The depth and expertise in mathematics to apply the appropriate
statistical techniques to solve a problem
A strong blend of mathematical and development skills to enable
them to implement analytical models
Expertise in machine learning techniques and technologies
Expertise in a the use of analytical techniques in a specific domain
To ensure that our people meet these qualifications, we’ve hired
individuals with advanced degrees, specifically PhD’s in Physics or
Mathematics with research experience in applying statistical methods
23. What Makes Agile Analytics Different
Traditional Analytics Agile Analytics
Often depends on data being in a perfect Data as it is, not how we wish it to be.
state. Delayed for years while waiting for Understand that there will never be a
long running Enterprise Data Warehouse perfect data warehouse. Data growth is
projects to finish. fast outstripping the ability of a data
warehouse group to make it perfect.
Focus on building a perfect predictive
model before trying it out. Not designed Focus on time to market. Get a model out
for iterative learning. there, get feedback, improve it, repeat.
Perfect is the enemy of the good!
Often focused on the software tool, not
the data science that goes into a Think like a startup. Use Open Source
solution. Software involved are often Software. FlightCaster’s founders did not
packages that cost into the millions of seek big enterprise software vendors – yet
USD. they are far superior to large airlines at
predicting flight delays.
Much higher up-front costs – not just for
software licenses, but for Minimize the “cost-to-experiment”. Ramp
implementation. up investment based on results, not
speculation or hubris.
Much higher risk due to the costs – and
more importantly – time spent on the
solution before you see results.
25. Define
Question
Gather
Retest
Information
The
Scientific Publish
Results
Form
Hypothesis
Method
Draw Test
Conclusions Hypothesis
Analyze
Results
26. Define
Question
The Retest
Gather
Information
Scientific
Method:
Publish Form
Results Hypothesis
5/8ths of the steps
in the scientific
method are about
testing our Draw Test
hypothesis and Conclusions Hypothesis
doing something Analyze
with it. Results
27. Define
Question
Agile Retest
Gather
Information
Analytics
: Analyze Idea
Application of the Publish
Results
Form
Hypothesis
scientific
method, lean Test Build
principles, and
agile practices to
analytics. Draw
Conclusions
Test
Hypothesis
Analyze
Results
28. Lean Startup
“The creation of rapid prototypes designed to
test market assumptions, and uses customer
feedback to evolve them much faster than via
more traditional product development
practices.”
… applies to agile analytics efforts as much
as it does to startups in general.
29. Getting Started
Start Small – establish a few smaller areas of focus, seek to get some
results and momentum as fast as possible. Take a humble approach to
this as your organization learns how to apply these techniques. Once
you understand how this works for you, then scale up.
Embrace Failure – seek to validation – or invalidate - your first
hypothesis as soon as you can. Build out a “minimum viable model”.
Don’t be afraid to try something small and fail. Focus on building a
capability to measure what works, so you can more effectively iterate
over the model and make it great.
People over Tools – agile analytics is much more about intellectual
capital than tools, processes, or even data. A small team of data
scientists can be much more effective than millions of dollars in hardware
and software.
Diversity over Size – data is important, but the hype around the bigness
of data obscures the importance of taking advantage of the diversity of
data. Remember you will often get insights from smaller sources of data
that happen to have the inputs that help drive a great predictive model.
Editor's Notes
One of the most acute demonstrations of the powers of predictive modeling is the recent story regarding predictive analytics at Target. This incident is so amazing that it sounds like an urban myth. However it is not. For years, retailers have been employing predictive analytics to find correlations between products and events or characteristics of the consumer. The classic story was the correlation between diapers and beer; the new father stopping on his way home from work picks up the diapers. A strategically placed six pack is immediately appealing to someone stressed from recent changes in their lifestyle.Through various identification techniques, such as cookies, user logins and even credit card data, analysts now have a much more intimate picture of their customers. Retailers are able to not only track purchases, but also the journey to that purchase, and those situations where consumers choose not to buy. Intuit Corporation, makers of Quicken and TurboTax, have created a 200 TB Customer Experience Database that combines advertising traffic with the user’s activity on their site. Through this, Intuit’s Big Data team is able to determine which channels are the most effective, how many visits will occur, on average, before the user buys and other behavioral patterns that have help shaped both their user experience design and their marketing strategy.But back to Target. With the volume of data now available, Their data scientists uncovered strong correlations between the purchase of key items (unscented lotions and skin crème, vitamins) and pregnancy. As a result, they are able to identify pregnant customers and customize offers. The story becomes a bit sensational, and a little creepy, when one of their targets is a 16-year old girl who has not revealed her pregnancy to her parents.
NOTE: This is true whether you are corporate IT or an independently funded startup