Strength in Numbers - ebusiness.mit.eduebusiness.mit.edu/sponsors/common/2011-AnnualConf/erik.pdf · Wal-Mart – inventory management Harrah’s - customer management Consistency

© 2011 MIT Center for Digital Business. All rights Reserved.

Strength in Numbers: How do data-driven decision-making

practices affect firm productivity?

May 19, 2011

Erik Brynjolfsson and Heekyung Kim MIT Sloan School of Management

Lorin Hitt University of Pennsylvania, Wharton School

1

2 © 2011 MIT Center for Digital Business. All rights Reserved.

The Nanodata Revolution

Clickstream/Page views/Web transactions

Mobile phone/GPS/Location data

Email messages

RFID (Radio Frequency Identification), Bar Code Scanner Data

Web links/Blog references/Facebook

ERP/CRM/SCM transactions

Real-time machinery diagnostics/engines/equipment

Google/Bing/Yahoo Searches

Stock market transactions

Twitter feeds

Wikipedia updates

Etc….

2


Examples of Data-Driven Decision-making (DDD)

Wine Chemistry Housing Sales Truck Routing

3


Research Questions

Do more data-driven decision-making practices improve firm performance (productivity, profitability, and market value)?

What makes a firm more data-driven?

4 4


Literature

• Data-driven decision-making (Davenport, 2009; Loveman,

2003; Lavalle et al., 2010)

• Information technology and firm performance (Weill 1992;

Dewan et al, 1997; Brynjolfsson et al., 1995, 1996, 2002;

Bharadwaj et al, 1999, 2000; Bloom et al., 2008; many others)

• Codified knowledge and organizational learning (Nelson and

Winter, 1982; Zander and Kogut, 1995)

5 5


Key Findings

• Data-driven decision-making (DDD) may explain a 4-6% of

output and productivity, controlling for traditional inputs and IT

use.

• DDD is also correlated with other performance measures

• return on assets, return on equity, asset utilization, market

value), controlling for other firm-specific characteristics.

• Firms with more consistent business practices are more data-

driven

• Younger firms also tend to be more data-driven.

6 6


2009 Digital Advantage survey overview

5

8

5

13

10

41

7 2

10

*

* Manu-

facturing

*

Health Care and

Social Assistance

Finance and

Insurance

Wholesale

Trade

Utilities &

Natural

Resources

Transportation

and

Warehousing

15

34

25

133

46

68

3

6

*

*

*

*

*

*

*

*

SOURCE: Digital Advantage Survey

Companies (330 total respondents; 179

matched to Compustat and IT use data)

Respondents by category

Percent

Respondents by revenue group

Count

7 7


Data-Driven Decision-Making (DDD)

How are decisions made for the creation of a new product or service? (1 to 5 scale: Experience and expertise=1, Data=5)

To what extent do the following statement describe the work practices and environment of your entire company.

• We depend on data to support our decision making (1: Describes not at all, 5: Completely describes)

• We have the data we need to make decisions

(1: Describes not at all, 5: Completely describes)

8 8


Data-driven decision making I: Typical basis for a new product/service

Percent of respondents

SOURCE: 2009 Digital Advantage survey

Minerals, Oil & Gas, Utilities, and Construction

0

25

3831

6

* *

Manufacturing

12

20

3025

13

* *

Wholesale/Retail Trade, Transport, Accommod./Food

12

3629

1212

* *

Information

1818

27

9

27

* *

Finance and Insurance

08

71

138

* *

Professional and Other Services

8

19

50

158

* *

9





19

50

25

60

Manufacturing

25

48

19

71


21

49

28

20

Information

9

45

27

18

0


25

46

25

40


11

70

1144

Not at all Completely






Data-driven decision making II: Use data to make decisions in the entire company

10

11 © 2011 MIT Center for Digital Business. All rights Reserved. 11




6

63

1913

0

Manufacturing

8

44

2719

1


7

47

33

14

0

Information

99

4536

0


13

63

21

40


0

50

2723

0







Data-driven decision making III: Have data we need


Estimation of the impact of DDD on productivity

Ln(Sales)it = β0 + β1 Ln(Materials)it + β2 Ln(Physical Capital)it

+ β3 Ln(IT Labor)it + β4 Ln(Non-IT Labor)it + β5 (DDD) i

+ Other controls

i: firm

t: year (2005-2009)

Sales, Physical Capital, Employee from Compustat

IT Labor from a job-posting site (Tambe and Hitt, 2008)

Non-IT Labor = Employee – IT Labor

Other controls = 1.5 digit NAICS industry, year,

employees’ human capital

(importance of typical employee’s education,

% of employees using PC/Emails, and/or Avg. workers’

wage)

> 0 ?

12


Productivity and Data-Driven Decision-Making (DDD)

13

Dependent

variable =

Ln(Sales)

DDD 0.0475** (0.019)

Ln(Material) 0.501*** (0.042)

Ln(Capital) 0.0991***(0.023)

Ln(IT-Employee) 0.0852***(0.022)

Ln(Non-IT

Employee)

0.224***(0.032)

Constant 1.133***(0.182)

Industry and

Year Control

Yes

Number of Firms 189

Observations 682

R-squared 0.92

Robust standard errors

were clustered around

firms. ***p<0.01,. **p<0.05,

*p<0.1.

Industry classification was

based on NAICS 2 digit for

manufacturing and 1 digit

for other industries .


Why do some firms adopt DDD more than others? What are the drivers of DDD?

1. Adjustment Cost: Firms with a higher adjustment cost have high

organizational inertia and do not find it optimal to

make an organizational change (Nelson and Winter,

1982)

- Constructed from 7 survey questions:

Please rate whether the following factors at your

company facilitate or inhibit the ability to make

organizational changes: 1) financial resources; 2)

skill mix of existing staff; 3) employment contracts; 4)

work rules; 5) organizational cultures; 6) customer

relationships; 7) senior management involvement

14



2. Firm Age: -: Older firms have high inertia and cannot make

organizational change (Hannan and Freeman, 1977,

1984, 1989; Bresnahan, Greenstein and Henderson,

2010; others) => Cov (firm-age, DDD) < 0

15



2. Firm Age: -: Older firms have high inertia and cannot make

organizational change (Hannan and Freeman, 1977,

1984, 1989; Bresnahan, Greenstein and Henderson,

2010; others)

=> Cov (firm-age, DDD) < 0

+: Selection on productivity – survived firms are more

productive than exit firms due to more resources,

better adjusting ability to environment, learning-by-

doing (Haltiwanger et al. 1999).

=> Cov (firm-age, ε) > 0 -> underestimation not overestimation

16


Why do some firms adopt DDD more than others?

What are the drivers of DDD?

3. Consistency of Business Practices

Cases and Literature

CVS – Enterprise IT system over 4,000 retail stores. (McAfee, 2008;

Brynjolfsson and McAfee, 2009; Brynjolfsson 2009)

Wal-Mart – inventory management

Harrah’s - customer management

Consistency of business practices across their branches let their firms

gain a higher performance through data-driven decision-making.

Thus, firms with consistent business practices have more incentive to

adopt DDD in the first place.


Construction of Consistency Measure

Survey Question Scale

Looking across your entire company, please rate the

level of consistency in behaviors and business processes

across operating units

(HR survey q1)

1-5

Regarding the first core activity of your company, the

consistency within business unit

(HR survey q9a)

1-5


consistency across functions (e.g., sales, finance, etc)

(HR survey 9b)

1-5


consistency across geographies

(HR survey q9c)

1-5

Effectiveness of IT in building consistent systems and

processes for each operating unit

(IT survey q13b)

1-5

18


DDD drivers can be potential instrumental variables (IV).

Productivity DDD

1.Adjustment Cost

2.Firm Age

3.Consistency

Instrument Variable (IV)

19


Productivity and DDD: OLS and IV

OLS IV

DDD 0.0475** (0.019) 0.064* (0.035)

Ln (Material) 0.501 ***(0.042) 0.504***(0.034)

Ln(Physical

Capital)

0.0991***(0.023) 0.0979***(0.023)

Ln (Non-IT

Employee)

0.224 ***(0.032) 0.224*** (0.032)

Ln (IT-Employee) 0.0852***(0.022) 0.0844***(0.022)

Industry and Year

Control

Yes Yes

R-squared 0.92 0.92

Overid Test:

Hansen’s J

0.68

Hausman Test 0.58

20

Robust standard

errors were

clustered around

firms. ***p<0.01,.

**p<0.05, *p<0.1.

Industry

classification was

based on NAICS 2

digit for

manufacturing and

1 digit for other

industries.


Does DDD improve the other performance measures?

1. Return on Assets: Pretax Income per

total assets

2. Return on Equity: Pretax Income per

equity

3. Asset Utilization: Output per total

assets


Interpretation Return on Asset Return on Equity Asset Utilization

Dependent

Variable=

Log(Pretax Income) Log(Pretax Income) Log(Sales)

OLS 2SLS OLS 2SLS OLS 2SLS

DDD 0.068

(0.049)

0.19 *

(0.11)

0.059**

(0.029)

0.088

(0.063)

0.066*

(0.034)

0.034

(0.062)

Log(IT-

Employee)

0.069

(0.054)

0.070

(0.053)

-0.041

(0.037)

-0.041

(0.036)

0.051

(0.035)

0.049

(0.035)

Log(Total

Asset)

0.69***

(0.07)

0.64***

(0.08)

0.42***

(0.05)

0.43***

(0.06)

Log(Equity) 0.90*** (0.04) 0.89*** (0.04)

Number of

Firms

174 174 174 174 179 179

Number of

Observations

568 568 565 565 682 682

R-square 0.76 0.76 0.85 0.85 0.84 0.84

Controls: Industry, Year, Log(R&D expense), Log(Advertising expense), Log(Capital),

Log(Total number of employees), Log(Market share), Importance of employees’ education


Does DDD increase market value?

Market Value = ∑ βiAi

(Market Value of firm = Sum of Value of Each

Asset, Ai)

(e.g. Hall, 2001; Hall et al., 2000; Baily et al.,

1981; Brynjolfsson et al., 2002)

Market Value = ∑ βiAi + α x DDDi x Ai

Coefficient (α and βi) is an indicator of how much

investors value a firm with each type of asset

Can DDD be thought of as an asset?


Market Value and DDD

Dependent variable = Market Value

Property, Plant and Equipment - Total (Net) 1.772*** 1.750*** 1.719*** 1.717*** 1.748***

(PPE) (0.495) (0.454) (0.429) (0.431) (0.458)

IT-Employee 8.262*** 6.348*** 7.598*** 7.983*** 7.505***

(2.003) (1.649) (1.635) (1.864) (1.714)

Other assets 0.191*** 0.202*** 0.192*** 0.192*** 0.210***

(0.034) (0.031) (0.029) (0.033) (0.026)

DDD x IT-Employee 3.097**

(1.267)

DDD x Employee 0.123*

(0.073)

DDD x PPE 0.304

(0.379)

DDD x Other assets 0.238*

(0.127)

Constant -5,494 -4,487 -5,060* -5,953* -5,332*

(3360.000) (2799.000) (2818.000) (3396.000) (3066.000)

Observations 676 676 676 676 676

R-squared 0.753 0.769 0.771 0.758 0.77

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

24

Market Value = β0 + β1(Physical Capital) + β2(Computer Capital) + β3(Other Asset)

(Brynjolfsson, Hitt, and Yang, 2002)

Computer Capital = f(IT) (Tambe and Hitt, 2008)


Conclusion

1. Data-driven decision-making (DDD) may explain a 4-

6% of output and productivity, controlling for

traditional inputs and IT use.

2. DDD is also correlated with other performance

measures (return on assets, return on equity, asset

utilization, market value), controlling for other firm-

specific characteristics.

3. Firms with more consistent business practices are

more data-driven; Younger firms tend to be more data-

driven.

25

Documents

Strength in Numbers - ebusiness.mit.eduebusiness.mit.edu/sponsors/common/2011-AnnualConf/erik.pdf · Wal-Mart – inventory management Harrah’s - customer management Consistency