Quick Measurements of Long Term
Effects in a Growth Environment
April 2017 ● All Things Data
Saar Golde, Chief Data Scientist
Transportation is undergoing a once-in-a-century transformation
2017 -> 2030: Networks of On-Demand Shuttles
(dynamic, shared, electric, autonomous)
1900: 20mm horses and horse carriages 1900 -> 1930: 15mm Model T cars
2017: 1.2bn private cars
10mm buses
Via: Software for operating networks of on-demand shuttles
On-demand
Passengers book using a smartphone app
Convenient
Passengers are picked up within minutes at a
nearby corner – a “virtual bus stop”
Shared
Passengers are seamlessly matched with
others heading in the same direction
Fully Dynamic
Vehicle routes and schedules are
updated in real time
((( )))
Pick-up
Drop-off
((()))
Via is the world’s first on-demand shuttle network
operating at scale
shared rides since
launch (2013)
13M+
rides per week
250,000
Venture funding
$137M
members across
NYC, Chicago, DC
700,000+
Via efficiency vs
competition
3x
Major global partnerships
launching in 2017
Global
Economics of Ridesharing
Rapid Growth
Low Margin Repeat Business
#ATD2017
The Approach: use Expected Life Time Value
• Low	margin	and	repeat	business:	Need	to	measure	long	term	effects
• For	example,	life	time	value
• Rapid	growth:	most	users	are	active
• Can	only	measure	life	time	value	for	a	small	subset	of	users
• Quick	measurement
• Can’t	wait	till	rider’s	end-of	life
#ATD2017
The Challenge: Measuring Marketing
Campaign Effectiveness
Expected Life Time Value
Expected	sum	of	(discounted)	future	stream	of	value	(revenue	/	profit)
Three	main	components:
• Life	expectancy	(duration)
• Stream	of	value	(amount	per	unit	of	time)
• Correlation	between	life	expectancy	and	the	stream	of	value
#ATD2017
Estimating the Effect of a Marketing
Campaign on Expected Life Time Value
#ATD2017
Incremental Value Added Life Expectancy Correlation B/W the two
Key Points for ELTV Estimation
• Control	group(s)	needed
• Stream	of	value	(amount	per	unit	of	time)
• Quickest	and	most	direct
• Life	expectancy	(duration)
• Slower	to	estimate	but	just	as	critical
• Machine	learning	speeds	up	estimates
• Correlation	between	life	expectancy	and	the	stream	of	value
• Hardest	/	slowest	to	measure
• Tricky	to	incorporate	directly	into	ELTV
• When	positive	can	be	omitted	to	find	lower	bound	on	campaign	effectiveness	
#ATD2017
Refine&
Iterate
Questions ?
Comments !
Thank You!
#ATD2017

Saar Golde - All Things DATA 2017

  • 1.
    Quick Measurements ofLong Term Effects in a Growth Environment April 2017 ● All Things Data Saar Golde, Chief Data Scientist
  • 2.
    Transportation is undergoinga once-in-a-century transformation 2017 -> 2030: Networks of On-Demand Shuttles (dynamic, shared, electric, autonomous) 1900: 20mm horses and horse carriages 1900 -> 1930: 15mm Model T cars 2017: 1.2bn private cars 10mm buses
  • 3.
    Via: Software foroperating networks of on-demand shuttles On-demand Passengers book using a smartphone app Convenient Passengers are picked up within minutes at a nearby corner – a “virtual bus stop” Shared Passengers are seamlessly matched with others heading in the same direction Fully Dynamic Vehicle routes and schedules are updated in real time ((( ))) Pick-up Drop-off ((()))
  • 4.
    Via is theworld’s first on-demand shuttle network operating at scale shared rides since launch (2013) 13M+ rides per week 250,000 Venture funding $137M members across NYC, Chicago, DC 700,000+ Via efficiency vs competition 3x Major global partnerships launching in 2017 Global
  • 5.
    Economics of Ridesharing RapidGrowth Low Margin Repeat Business #ATD2017
  • 6.
    The Approach: useExpected Life Time Value • Low margin and repeat business: Need to measure long term effects • For example, life time value • Rapid growth: most users are active • Can only measure life time value for a small subset of users • Quick measurement • Can’t wait till rider’s end-of life #ATD2017 The Challenge: Measuring Marketing Campaign Effectiveness
  • 7.
    Expected Life TimeValue Expected sum of (discounted) future stream of value (revenue / profit) Three main components: • Life expectancy (duration) • Stream of value (amount per unit of time) • Correlation between life expectancy and the stream of value #ATD2017
  • 8.
    Estimating the Effectof a Marketing Campaign on Expected Life Time Value #ATD2017 Incremental Value Added Life Expectancy Correlation B/W the two
  • 9.
    Key Points forELTV Estimation • Control group(s) needed • Stream of value (amount per unit of time) • Quickest and most direct • Life expectancy (duration) • Slower to estimate but just as critical • Machine learning speeds up estimates • Correlation between life expectancy and the stream of value • Hardest / slowest to measure • Tricky to incorporate directly into ELTV • When positive can be omitted to find lower bound on campaign effectiveness #ATD2017 Refine& Iterate
  • 10.