Michael Borokhovich, Avhishek Chatterjee, Jason Rogers, Lav R. Varshney, and Sriram Vishwanath

20 [tasks/hour] 50 [tasks/hour] 80 [tasks/hour]0

200

400

600

800

Aver

age

back

log

[num

ber o

f uns

ched

uled

ste

ps]

1145

68622977

step_flex, 500 wstep_semiflex, 500 w

step_inflex, 500 wstep_flex, 700 w

step_semiflex, 700 wstep_inflex, 700 w

20 [tasks/hour] 50 [tasks/hour] 80 [tasks/hour]0

5000

10000

15000

20000

25000

30000

35000

40000

TAT

[sec

]

54058

15165877517

step_flex, 500 wstep_semiflex, 500 w

step_inflex, 500 wstep_flex, 700 w

step_semiflex, 700 wstep_inflex, 700 w

20 [tasks/hour] 50 [tasks/hour] 80 [tasks/hour]0

10

20

30

40

50

60

70

80

90

Wor

kers

util

izat

ion

[%]

step_flex, 500 wstep_semiflex, 500 w

step_inflex, 500 wstep_flex, 700 w

step_semiflex, 700 wstep_inflex, 700 w

Impact sourcing •  Outsourcing tasks to people in underdeveloped regions •  Samasource is a leading in impact sourcing •  Regions: Haiti, Ghana, Uganda, Kenya and India •  Since 2008 - $3.5M in wages •  More than 30K people lifted out of poverty •  89% of workers pursue additional employment and/or

education after working for Samasource

Customers can get jobs done for 30%-40% less

Samasource provides customers •  Project management •  Consultation on task design •  Connects them with workers Partner agencies in underdeveloped regions provide •  Delivery centers (physical facilities) •  Stable electricity •  Local managers

Work model •  Project is split into smaller tasks •  Tasks are distributed to global work centers

via web-based platform Samahub

Task has several steps Step has several substeps Substep requires a single skill: Ordering may be imposed among steps

Each worker: •  Set of skills, e.g., [2,3,7] •  Timezone and daily working hours, e.g., [9-17]

For each substep find workers •  one subset - many workers •  one step - many workers

STEP_FLEXFor each substep find a worker •  one subset - one worker •  one step - many workers

STEP_SEMIFLEXFor each step find a worker •  one substep - one worker •  one step - one worker

STEP_INFLEXTheoretical Results •  STEP_FLEX - optimal •  STEP_INFLEX – optimal if

steps are inflexible

Making impact sourcing more efficient We improve the efficiency by: •  Data-driven scheduling algorithm •  Proposing certain organizational flexibility:

Flexible steps, Flexible substeps

Scheduling  Algorithm  

Task  

Step  1   Step  2  

Task  

Step  1   Step  2  

Project  

Step  3  

Substeps

… …

Workers

Real data (9.3M tasks). Improvements of STEP_FLEX upon current system: •  x6.5 for all projects •  x8 for real-time projects

Synthetic data. Flexible vs inflexible steps and substeps.

Improving Impact Sourcing via Efficient Global Service Delivery