2. • AI will transform the way merchants do their jobs – “software-as-a-
coach” to empower merchants to focus on the areas where human
judgement makes the biggest impact
• Price Optimization that is AI-powered and experiment-driven will win in
today’s dynamic digitally connected environment
ActionableTakeaways from this session:
3. Source: Oxford Martin Programme on the Impacts of Future Technology
M E R C H A N T ’ S R O L E
W I L L B E A M O N G T H E
M O S T D I S R U P T E D B Y
A . I .
A.I. ISTRANSFORMING HOW MERCHANTS DOTHEIR JOBS
Employment level vs. probability of computerization
4. Source: https://www.bloomberg.com/news/articles/2018-06-13/amazon-s-clever-machines-are-moving-from-the-warehouse-to-headquarters
Amazon began automating retail team jobs several years ago. Under an initiative called “hands off the
wheel,” the company shifted tasks like forecasting demand, ordering inventory and negotiating
prices to algorithms, people familiar with the matter say.
At first, humans could easily override the machine’s decisions. For instance, if a brand notified
Amazon about an upcoming marketing blitz for a product, an Amazon manager could increase the
order in anticipation of demand the algorithm didn’t expect.
But such tinkering was increasingly discouraged as the machines proved their precision, the people
say. Anyone overriding the machines had to justify their decision, and the push to automate made
them reluctant.
5. Amazon Pricing: AI & Experimentation in Practice
$27.98-$35.57 price variation over 12 months
• Price is more discoverable
• Shoppers will buy from
anywhere
• Prices move faster than circulars
and can be updated
THE RESULT: AN INCREASINGLY
DYNAMIC COMPETITIVETHREAT
6. AUTOMATION
(Technology)
CRUTCH
SOFTWARE
40 years of “forms” software Machine learning / AI
LEARNING
(Human)
COACHING
NETWORKS
We need to advance our innate human talents faster
than automation tries to replace them
Capability
Source: Emergence Capital
WHAT FORM WILL A.I. SOFTWARETAKE?
AI-EMPOWERED
AI-OVERPOWERED
7. GATHERING
Shelf-edge experimentation
at store level tests new
prices continuously
COACHING
Guiding UI surfaces top
opportunities for
merchants, automating
the pricing decisions
Merchants focus their attention on the tough judgement
calls where humans outperform
COMPARING
Machine Learning Algorithms uncover current price
elasticity + interaction effects
CREATIVITY
Gather Behaviors
Combine with Peers
Predictive Advice
Human Creativity
AI-POWERED SOFTWARE COACHES
MERCHANTSTO OUTPERFORM IN
A DYNAMIC MARKET
8. HOW IT WORKS: DYNAMIC PRICING AND PROMOTIONS
EXCEPTION-BASED HUMAN DECISION MAKING
AUTOMATE ROUTINETASKS
Inputs
Eversight Pricing Suite
Intelligence
Guard rails keep prices and
discount levels in bounds
RULES MANAGEMENT
Feed intoAI to help
guide experimentation
COMPETITIVE FEEDS
Tests higher and lower prices
and promotion mechanics
DIGITAL
EXPERIMENTATION
Balances sales and margin for
each product toward objectives
GOAL OPTIMIZATION
9. AI-POWERED EXPERIMENTS DYNAMICALLY
CHANGE PRICES ACROSSTHE STORE
NETWORK
• Results uncover “magic price points”
• Performance is tracked at the item and
category level
• Store groupings are designed to ensure
unbiased results
• Experiments are perpetual
Lower PriceHigher Price Control Price
10. BASIC RULES ARE MAINTAINED BYTHE SYSTEM TO SIMPLIFY
AND AUTOMATE ROUTINE WORK
11. STATISTICALLY BALANCED PRICE EXPERIMENTS ARE
AUTOMATICALLY GENERATED
Line Group Name
Control stores
1 – 9
Test stores
10 – 15
Test stores
16-20
Test stores
21-25
LINE GROUP 1 Last period price Lower price Predicted optimal price Higher price
LINE GROUP 2 Last period price Higher price Lower price Predicted optimal price
LINE GROUP 3 Last period price Predicted optimal price Higher price Lower price
LINE GROUP 4 Last period price Higher price Predicted optimal price Lower price
LINE GROUP 5 Last period price Lower price Predicted optimal price Higher price
LINE GROUP 6 Last period price Predicted optimal price Lower price Higher price
LINE GROUP 7 Last period price Lower price Higher price Predicted optimal price
12. Reasons for a merchant to be
alerted to a recommended
price change:
High financial impact
Large percentage price
change
Competitive price change
Multiple changes in a row
Vendor’s cost changed
KVI/Watchlist Item
EXCEPTION-BASED UI COACHES MERCHANTS WHERETO
SPENDTHEIRTIME
13. 2.0 2.5 3.0 3.5 4.0
0.51.01.52.03.02.5
Price
AvgQty
5 Weeks
2.0 2.5 3.0 3.5 4.0
0.51.01.52.03.02.5
Price
AvgQty
104 Weeks
MORE ACCURATE DATA, QUICKER
• Detects demand
curves in 1/20th
the time of
traditional
approaches
• Gives current
information, not
average elasticity
over last 2 years
$1.99
$2.00
$3.59
$1.99
$3.13
$3.29
$3.49
$3.59
$3.75
$3.89
Regression of past POS data Eversight shelf-edge pricing
Product X Product X
ELASTICITY DETECTION Modeled elasticity Current price
14. 4.5 5.0 5.5 6.0 6.5
Price
104 Weeks
0.20.40.60.80.70.30.5
AvgQty
THIS APPROACH ALSO DELIVERS MUCH MORE…
Elasticity for products who
haven’t had price changes before
Correcting for mis-reads
on elasticity
Product Y
Product Z Product Z
ELASTICITY DETECTION
$2.99
2.6 2.8 3.0 3.2 3.4
Price
AvgQty
104 Weeks
1.01.21.41.62.01.8
$2.99
2.6 2.8 3.0 3.2 3.4
Price
5 Weeks
1.01.21.41.62.01.8
Product Y
$2.63
$2.73
$2.86
$2.94
$2.99
$3.10
$3.26
$3.36AvgQty
4.5 5.0 5.5 6.0 6.5
Price
5 Weeks
0.20.40.60.80.70.30.5
AvgQty
$4.99
$5.39
$4.54
$4.86
$4.99
$5.34
$5.44
$5.56
$5.80
15. AND UNCOVERS CROSS ELASTICITIES INTHE WAY
HISTORICAL DATA NEVER COULD
QProduct1 = QAVG – eProduct1 x PProduct1 + eProduct2 x PProduct2
where:
eProduct 1 = -0.29 (97% confidence)
eProduct2 = +0.65 (99% confidence)
CROSS-ELASTICITY DETECTION
2.5 3.0 3.5 4.0 4.5
1.01.52.02.53.0
Price of Product 2
QuantityofProduct
1