Tip:
Highlight text to annotate it
X
Nicholas Bloom, Associate Professor of Economics at Stanford
University and the Business School, explains a rare field
experiment conducted between 2008 and 2010 that provided
evidence that good management practices add value to the
bottom-line of manufacturing firms. The experiment by
Stanford and World Bank researchers was conducted in a
textile manufacturing center in India. Bloom's presentation
was at a September 2010 academic conference, honoring his
co-author John Roberts, the John H. Scully Professor of
Economics, Strategic Management and International Business
at the Stanford Graduate School of Business.
So the idea is to try and run a field experiment on
management practices to try and see if we can get some
causation from management performance in. Going to Scott's
challenge earlier, it is an open question as what matters,
and
[inaudible] saying earlier as complementarities, can't we
show there are any practices that even in small changes have
an impact. So the way we try to go around addressing this
question is we took a bunch of large firms, whereby large
firms we mean 300-person firms, in Mumbai that work in
textile. So why Mumbai? Well, Mumbai is the commercial
capital of India. India is a very nice place to do field
research in. It's a developing country, and it's pretty cost
effective, and it's nice to go out and work there. And we
pick one industry which is broadwoven textiles, cloth, so
basically it's clothing, shirting and suiting. And we
randomly selected these large firms into treatment and
control plants. The treatment plants got six months of free
consulting from another consultancy firm, not McKenzie,
Accenture in this instance. And the control guys got a
month. So why did they get a month? They got a month because
one of the classic firms in field experiments is you have to
get data of the control group. So there is always a problem
you face. That's very easy to deal with the treatment guys.
You go and give them something that they want. You collect
data before and after. The control guys, you have got to get
before and after data. So in that case, we had a very heavy
treatment for six months for the treatment guys, basically
light treatment for the control guys and an ability to
collect data. We are going to collect weekly data on all the
plants we have down for two years on a whole range of
outputs. And I will show you this in a lot more detail but
basically, the (inaudible 2:36) of things we found. One of
them in some - well, given the discussion earlier, it's not
obvious actually how predictable it is. It depends kind of
what angle you come from. But you see very large increases
in profits and productivity. And because of the randomized
nature of the intervention, you can argue this is causal. A
lot of it just comes through better quality control, and I
will show you a set of kind of much more detailed evidence.
But basically, these firms are in, what I call, the kind of
[inaudible] model of production. They make massive amounts
of cloth. They make it as fast as they can. Then about 20%
of their manpower is involved in fixing and repairing at the
end. And part of the intervention is trying to move them
slightly more towards a more modern manufacturing process.
If you find defects, fix it on the spot so it doesn't
permeate. And a second issue is around decentralization of
power in firms. John Zillich, as we discussed earlier, has
also been involved quite heavily in - there is an
organization in the firm. What's one of the things that was
less obvious to us and just turned down as we run the
experiment is better management practices have lead to
pretty have decentralization within firms. Again, talking to
Bob on this, I guess my interpretation of it and what we see
on the ground is the better management practice improves
informational flow within the firm. So the owners have felt
more relaxed about letting plant managers to take decisions
in part because the owners know what's going on, in part
because they can spot a theft. And then finally, there has
been a big increase in computerizations. So as you have
needed to collect and process data more, you can use more
computers. So let's back up, what do these places look like.
Here is a picture of one of the factories. I put this up to
give you an idea. These are large factories. Typically, this
is one building. You can see it's in a gated compound with
several floors, particularly two buildings per factory. Most
firms, medium firms have two factories and a head office in
Downtown Mumbai. So I put this up to kind of highlight, and
these are complicated organizations. My sense in terms of
manpower complexity, and there may be a similar size to
something like the GSP. I don't quite know how many
employees the GSP has but that big, they operate 24 hours a
day. There is no way you can run them in an ad-hoc manner.
Here is a second plant. Again, you can see a large
complicated facility. Here's the first stage of the
production process coming back to talking about CAD/CAM, in
terms of CAD/CAM being an old technology, you can see some
of this stuff looks like it's left over from an imperial
India. I think there is some pretty old equipment lying
around. This is the first stage of making fabric. So when
you make fabric, what happens first is you get all these
pools of yarn. So the input to this is yarn thread.
Upstream, this is happened in an industry called spinning
which takes kind of bundles of cotton and spins it into
yarn. And this machine rotates at an incredibly high speed
and basically wines all these yarn on to something called a
warping. So that's stage one. This thing would spit out
several warpings a day. These warpings are then
[inaudible]basically into the second stage which is called
weaving, and here are these warpings. These things rotate
around, and this shutter goes back and forth very fast and
puts the cross threads. So that's the weft. So all fabric
has a warp and a weft thread, and once you put both of those
through, coming out at the far end of this machine is the
fabric. So rims of fabric come out very fast. Because of the
way this factory is set up, because their quality is
basically very poor, the third stage is a massive mending
and repair area. So they will have huge rooms full of people
who are spreading cloth out over these backlit wooden
structures, even the fact that they are manual. And I was in
China earlier about a couple of months ago, and some of the
Chinese factories have automatic things that pull this cloth
down. And these guys are pulling this cloth down manually,
spreading out, looking finding defects, mending it, pulling
a lot of cloth down further. It's very time intensive and it
takes up about a fifth of the manpower. And also, other
downside is they can't fix anything. They are throwing away
on average about 7.5% of the output because the quality is
so poor. So in fact, I actually asked
[inaudible] to raise an anecdote. I remember he told me a
story about when - well, he told me story about when you
were the stamping in there, I think it was the hubcaps, that
this isn't a probably unique to India.
So when I was an undergraduate, I worked in General Motors
factory where we were making bumpers, and I was on the
repair section where if a bumper came through that had a
flaw, I was supposed to take it off the main line and put it
on to the repair line where it would go by a repair machine,
get polished and get put back on to main. This is after it
went through copper plating and then nickel plating, and it
was after the nickel plating and before the chrome plating
that they would do this repair. And one day, I went there,
and there was a flaw in one of the presses and every bumper
that came down the line had a flaw in the same place. And I
was holding the entire production line from the main line on
to the repair line where these racks may go by and told they
were repaired, and the repair line got completely filled.
There was no where left to put the flawed bumper. So I
pushed the Stop button on the line, and the plant manager
was with me and told me you never push the Stop button on
the production line, we never push the Stop button. But
anyway, I got read out for doing this, and I said, "Well,
what am I supposed to do?" "Put them on the floor, put them
anywhere but don't push the Stop button." So that's the
story.
[Informal Talk]
That anecdote struck me as so representative of what's going
on in these factories. I will come back to it in a minute.
But the same defect is just going through and through and
through, and they are just producing the same defect just
continuously. And this is why 20% of the guys, and they are
fixing the same thing. You can imagine how frustrating that
is. The factories were just generally disorganized and
dirty. Here is various shots from factories. I mean we
visited - we have 20 factories we visited. You could never
get a shot like this from a U.S. factory, I mean to the
extent that you could ever even find this, there is no way
they would let you take a photo of it and put it up on their
presentation. I mean out there, it's no big deal, I am
taking photos, this kind of thing. And it's very hard to get
high levels of efficiency when your factory is in general
chaos. The plant floor is often disorganized. The aisles are
blocked. I mean this is a complete reverse of lean
manufacturing. You can see, you need to get equipment up and
down the aisles, you need to get warpings in and out, you
need to get when you
[inaudible] the fabric at the end, you need to get the
fabric rims out and again hey, here there were some table
and chairs, they couldn't really explain what it was here
for. Apparently, they have been here to mend something
historically. They kind of left it around. There is no piece
of equipment. No, I mean it's like endemically terrible. You
had this all out, you can see, it can get pretty low levels
of efficiency. And other thing you notice is the repair
system, it was kind of very old school in the sense if you
just repair stuff when it breaks down. So again, the whole
modern manufacturing there is you repair stuff in advance,
you have routine maintenance. Why? In part because it's more
efficient probably to repair stuff in advance rather than
wait till it breaks down, but also primarily it reduces
variability. So a big problem is you are trying to deliver
an order and your machine breaks down at the last minute,
you generate huge variability in output. Inventory rooms are
pretty chaotic. So here is just bags of yarn. Yarn a big
deal for them. It's expensive. Labor is cheap in India but
capital is expensive. Materials are relatively expensive,
not much ordering. Here is another yarn store. We didn't
actually try and do anything particularly high level and
complex, we went for basically extremely basic, low-hanging
fruit, that kind of stuff. But at least our belief is there
are some complementarities between my comeback to them but
even individually they are good. So things like preventive
maintenance is carried out for the machines is carried out
following the manufacturer's recommendations. They don't
make up their own maintenance schedule. They do it as it
says in the manual. The shop floor is marked where the
machine should be. Some stuff about quality defects, when
you have a defect, you record it, you analyze it, etc. There
is stuff that in European and U.S. factories would be pretty
standard. In these factories, frankly, as a
[inaudible] pretty not standard, some stuff in inventory
control, etc. So here is a bit of data. So what we did is we
recorded the adoption of these practices before, during and
after the intervention. So here is months before the initial
diagnostic phase. So what happens is all the plants have one
month of initial diagnostic phase by attributing control and
the diagnostic phase goes in there and basically collects
data including bills of back series of data dating back a
year and then gives them a set of recommendations on what
they should do, so tell them you need to put an inventory
control, quality control, etc. The treatment guys then get
another four months' implementation to help them carry it
out. The control guys would basically leave and come back
one day a one month to collect further data. So what you can
see is in advance of the consultants turning up, they are
adopting about 25% of those practices. So the stuff I showed
you, they typically don't have quality control, they
typically don't have routine maintenance, etc. The
consultants go in, month zero. And several things, some of
them are kind of surprising. So less surprising, the
treatment guys improve so they adopt a lot of these
practices. The consultants are sitting there on the ground.
They are Accenture guys. There is one person for every two
plants. It's pretty intensive. They are there for several
months. They manage to get through and adopt quite a lot of
them but they don't manage to get into
[inaudible] anywhere near 100%. So one big challenge is one,
they have experts that are very expensive of that. I will
show you in a minute and show a kind of proven impact. They
don't listen to them. The control plants improved but much
less since this is how they are going to get identification.
There is a third group of plants that are interesting which
are called the excluded plants. So these firms own multiple
plants each. And we don't actually provide treatment control
to every plant. So if Kathryn and Jonathan are each on a
plant in one of these firms, you may have Jonathan as a
treatment and Kathryn as completely exempt. She is neither
treatment nor control. So if G would be an excluded plant
and we go into those as well and collect management data.
And what you see is this is increasing to some extent and
this is entirely copying. So we are not sending consultants
on the ground. They are just basically replicating it over.
Again,
[inaudible] as many questions answered as it solves in a
sense that you know why is it they don't copy a lot of this
over. So they are copying some of it, it's very slow
adoption. But certainly, this changes over time. One of the
things that's kind of interesting is we know from the
anecdotal I guess PE and kind of consulting that it takes
several years. I mean Jonathan, it's coming back too
earlier, it takes just a long time to change practices. We
were in that for a year and it's just very tough to put in
place big changes. What do we do? So what are the areas we
see improvement? So one of the things I mentioned is
quality, the biggest single improvements in terms of
productivity really come from better quality. So beforehand,
massive amounts of manpower, 20% of the manpower spent on
repairing defects, about 7.5% of output is thrown away. How
do they fix this? Well, it's a set of kind of very simple
systems. Historically, they have these things called "
quality log book". So every time there is a defect, they
write it down in the checking stage who the weaver is, the
piece number, the machine number, etc. They write down some
information and then there is a quality grade. So A is the
top grade, it means perfect. Below that, by the time you get
to D, it's basically rejected. So A, B means pretty good
piece of fabric but there is some small defect in it. They
record this because buyers will often ask for refunds and
say hey my piece of fabric had a hole in it and I want my
money back. And they have to go look at their log books and
work out whether to give the money back or whether the buyer
is basically trying their luck. But they never really
systematically analyze it. So one of the things consultants
get in to do instead is change the recording format to
something that looks like this. It's actually fast the
recorder in and again it has a weaver piece number, etc. but
the key difference is they have columns for each of the
major defects. So here there is a 2 in the broken
[inaudible] which means there are some broken
[inaudible] defects in the fabric. And this is a big change
for them because they computerize it. So this they can now
feed into the computers. Every day, they now meet and they
analyze the quality, defect numbers. And this is extremely
effective for improving quality because you can look at
defects, you can spot them in quick time and fix them. So
you know a good example of this would be stain. So a lot of
the fabric will have a staining problem. So particularly,
[inaudible] or white fabric that has oil stains in it, I
don't think it does, but if it had oil stains in it, there
would be a problem in to try and sell it. Oil stains sound
easy to fix but they are not actually that easy to fix
because they come from multiple sources. So one thing can be
that the guy handling, the weaver hasn't used gloves and
other thing can be oil docks have broken and stuff leaking
out with someone kind of overfilled the oil valves. Once you
have the quality/defect data, it's much easy to fix. So you
can see this one weaver, one shift, one piece of machinery,
they fix it, as soon as they fix it, they deal with it. So
you don't get the kind of the problem Paul was talking about
thousands of defects running through. And here is in essence
the output data for quality. There is something called the "
quality/defects index" which is a weighted average of
quality defects, a very high number means bad quality. So
here is bad quality up here, here is good quality. We
normalize in red the control and in black the treatment
plants before the experiment. I mean there is a bit of
movement around here and you can see the standard areas are
reasonably
[inaudible] example. But after the experiment, the treatment
plant's quality improves while the consultants are here and
there is a kind of
[inaudible] significant ongoing downward drift. So what's
happened is quality/defects full above 50%. And this is a
low and it's an enormous improvement for the firm because
they can save in terms of the manpower, increase output
effectively about 3%. So you can do similar - I would say
you can show
[inaudible] statistically significant, etc. So then you can
do similar things for inventory beforehand, and now we have
reorganized the inventory, put it on shelves. We bag it, put
labels, enter into the computer. There are some amazing
things happened when they did it. For example, one the
plants discovered they used to make shirting and suiting
fabric, when they organized their store and recorded all of
it, they had stopped making suiting about two years ago but
they discovered large bags of suiting yarn in the back of
their store. You realize they have held this stuff for two
years, they of course can now set it off. But no wonder they
are carrying excessive inventory. The other thing they do is
to the extent they have excessive inventory to make shade
cards. So these are cards that record what the fabric is,
write a bit of information about the strengths and the
stitch. They send it to the design teams down in Mumbai, and
they design into new products to try and get rid of the
excess inventory. And again, inventory
[inaudible] over time. And then finally, output, so they did
a set of changes in output, marked out the factory floor,
did these things called "snag tags" where they go around and
they put these little labels on that machine so that when
the maintenance men come around, they can pick off the tags,
spot the defects, organize the spare parts, the tools, went
from kind of the very old scrappy handwritten notes to
computerized records. Here's like my favorite photo. You can
tell how well they are looking after their records based on
the large footprint on the record. You can see how carefully
they are monitoring stuff beforehand. They put up these
boards. I mean an interesting question on piece rates they
don't provide extremely high piece rates but they now give
them about 10% piece rate with the board. So again, an issue
I guess on complementarity, there is no point providing
piece rates if you don't provide feedback to the workers
[inaudible] how they are doing. As soon as you put in piece
rates and put them up on the board, it's relatively easy for
them to tell who is doing well and who is not doing well and
cross compare. And again, performance improves. Another
thing that we know, by the way, this is kind of interesting
that you get from the regression tables is in column one is
running OLS, so this is exactly the same as Kathy for
example is going to assist you
[inaudible]. So what we do is we have a panel of performance
data and a panel of management data, and we run regressions.
So we put in basically plant fixed effects and time dummies.
So we are looking at the change in performance versus the
trend versus the change in management versus the trend. And
you see here, in this case, it's not significant but in many
of the indicators, we typically get a significant effect
even in the OLS. In the second column, we go to instrumental
variables, so instrument the changing management with the
amount of treatments. So basically, you are getting an
entirely causal effect. And you see in almost all of these,
these coefficients go up. In this case, the impact on output
is tripled. And the reason this happens, and I think it
comes back a bit to the reason these guys were badly managed
in the first place, is they don't tend to change until stuff
goes wrong. So what's happening is you have these factories,
performance going along on average is even improving, they
are not that receptive to the management changes. As soon as
things start to turn down, performance starts to slump. They
are much more receptive to changing management practices. So
the effects that you see in OLS seem to be a downward by
some of what we get in in terms of running experiments. So
when you are on cross-sectional service, to the extent we
get some at least on our evidence suggests that the true
impact may be even larger. So let's look at bit on the
decentralization and IT. So I mentioned, other changes that
are going on in these factories is decentralization of
information control from the owners, the firms down to the
plant managers. So a typical situation is imagine, I am the
owner and we have Peter and Bob as the two plant managers.
So I am the owner. I live in Mumbai. The town is Tarapur.
It's about three-hour drive outside Mumbai. I typically go
out to Tarapur everyday, spend half the data at Peter's
factory, half the day at Bob's factory because I, as the
owner, don't want to do this particularly, I am spending
huge amount of time commuting them back and forth. I would
much rather let these guys run the thing without me. The
problem is I can't trust them in my absence. So one of the
problems you see for example is a lot of theft. So if I don'
t turn up in the factory, I worry that Bob for example is
stealing parts from the factory reselling them. I don't know
what's going on, it's hard for me to stop this so I have to
come in everyday. If instead you start to have a detailed
performance metrics, I can track things on a daily basis,
it's much easier for me to spot what's going on. I don't
need to turn up as much. I also rely that Bob has a better
ability to run his own factory. So when you give them better
management tracking systems, better information tracking
systems, we see increased decentralization in the sense the
owners don't turn up to the factory as much, the plant
managers have much more autonomy in hiring and investment.
And we measure it here in a range of different indicators
and take the average but this decentralization index is kind
of an average of hiring decentralization, investment
decentralization, choice of the plant and products of the
factory manager and this is increasing very dramatically in
the treatment firms versus control firms. The other thing we
look at finally is computers. So these factories are pretty
low tech in terms of computerization. So here's a picture of
actually a relatively high tech factory. Most of the
factories will have one office out on the factory floor with
some guys using computers, here are kind of the old green
screen computers, the average factory have something like
21/2 computers for 3000 people. Here's a much more low tech
setup where one woman with a computer in the corner and a
dot matrix printer. Quite a few of the plants don't have any
computers at all. Again, we see very big effects on
computerization and the reason for this is that in some
sense it's pretty intuitive. If you put in process these
modern practices involving data collection, data processing,
data analysis, you need much more computerization to do it.
So coming back to the discussion of complementarities
through out the day you can see that clear there is a
complementarity between kind of
[inaudible] talking about earlier between capital and
practices.
Audience: It's not just that they needed the computers to
put in those systems but they use the computer much more
intensively than they used to. So there are lot more people
who are looking at computer generated reports and are
running things on the computer.
Yes, it's a very good point. I guess we can highlight that
in part because of this big debate of a skill-based
technical change that you see in the US and Europe, across
the Europe, a big increase in income inequality. And part of
the reason people think this is going on is because of an
increase in the returns to skills. But what is technology
driving is returns to skills, well coming back to what Cathy
was talking about earlier, one that the technology we think
may just be improvements in management practices because
modern management practices are much more IT intensive. The
use of IT requires much higher level of skills than
traditionally for example in these factories so as you
increase, if you adopt modern management practices, you just
see there is more hours spent on the computer which
increases demand for people that can use computers. So for
these factories this has been kind of IT intensive but also
skill intensive. So finally, when were they weren't
introduced before, there are a lot of issues out in the
market so I am going to skip this and get on to kind of
[inaudible] stuff. But basically they have very high
tariffs. Chinese input which are the obvious competitors are
heavily tariffed. It's very capital intensive so hard for
new entrants to enter. And it's very hard for the current
successful firms to expand. They have this big problem about
reallocation. So coming back to the Peter and Bob story, if
I am the owner and these are my two factories, and I have to
spend all my day managing one or the two of them it's very
hard for me to setup a third factory. In fact, the best
managed firms out there, when you ask them why don't you
expand, they are very profitable, why don't they expand and
setup new factories, the primary reason is I don't have
enough time in my day I am already running two factories.
The only reason they manage to do that is basically have
people they can trust, who are they, they are brothers and
sons, etc. So the thing you find out there that seems to
explain firm size is number of male family members, the
people they trust around the firm. So the biggest firms are
the guys with 4 brothers, the smallest firms typically
tended to be you know the brilliant guy but had no brothers
and no sons so he had no one. India has kind of the same
former legal system as the UK but just has a very slow
corporate system. So when you ask them about can they borrow
money, yes, do they have collateral to start off with as an
outsider, no. So in other words, it's impossible and you
will come up with stories of people who have done this. So
most of the owners started off by trading and then they
maybe bought 10 looms and they did well and scaled up. The
kind of US example, a fantastic manager, raises a lot of
money, mortgages his house and goes in and starts a firm,
it's not zero but it's very low. So it's like
[inaudible] everything is moving very slowly. So it's not
like reallocation doesn't happen at all and entry and exit
doesn't happen at all but it's a slow process and so you get
these firms that persist that are badly run that are making
money and not dropping out of business. And the other
question is why don't they improve themselves, well the
evidence we found on the ground is primarily informational.
So when you go in there and ask them about it, they have
mostly heard of it. So there are several hypothesis about
it, one is they have never heard of these practices before,
that's mostly not true. Most of the stuff they have heard of
before, they are aware of it. The big problem seems to be
they don't believe it matters. So for example in the quality
issue you go talk to them and they will say you know you can
improve your quality by introducing quality control systems
and they would go but my quality is good, it's better than
Vinod's down the road, I don't have a quality control
process problem. And in a sense it's true compared to their
local competitors and it's related to the fact they don't
face global competition. But in the bigger scheme of things
their quality processes, many of the other processes aren't
fantastic, and they look like European and I guess US firms
of 20-30 years ago. There is a frontier that's improving
overtime, Indian firms are 30 years behind the frontier,
this is kind of moving them 15 years towards the frontier
and the frontier would keep evolving.
Audience: The fact that the Indians are, so many of these
firms are badly managed and it's an informational problem,
think about the US auto industry. It took it at least 20
years to figure out what the Japanese were doing differently
and to come anywhere close to implementing what was well
documented, it was no secret. But they just initially
wouldn't believe it and then couldn't believe it and then
couldn't do it.