Harikesh Nair
What makes big data big is not its volume
but its impact on decision making!
This month, I feature an interview with Dr. Harikesh
Nair, Professor of Marketing, Graduate School of Business, Stanford
University. Specializing in marketing analytics, Dr. Nair
blends applied economic theory and econometric tools with marketing data
to better understand consumer behavior and to improve the strategic marketing
decisions of firms.His research is a must read for anyone interested in big
data and marketing through social media
1. Last year you delivered a short talk
entitled “Building Better Employee Incentives with Big Data.” Can you summarize
the highlights of that talk? How would you define Big Data?
I think the “bigness” of data should not be defined by the amount or the
quantity of data or the speed at which it can be collected, but by its ability
to make big impacts on decision-making and outcomes when utilized properly. For
me, big data means the fact that
information and analytics can be combined in new ways with domain knowledge
that managers already had, to make large impacts on the organization. That is
what I consider big data.
A lot of companies have made large investments in collecting data on how
the companies interact with customers. We refer to that as CRM, or customer
relationship management systems. We now can track a lot about customers—what
the touch points are, what the customer did. That has generated a revolution in
how consumer interactions with companies are analyzed. What I was discussing in
the video was a similar revolution occurring within the firm. Specifically, in
much the same was as customer interactions with the firm can now be modeled, we
can think about how employees behave, how they interact with one another within
the organization, and model employee behavior in a granular way. All of that
can be tracked. We can conceive of the firm as an internal market. I think that
is a huge change that will occur in the near future: the notion of big data
will enter within a firm, an organization.
The fact of the matter is that right now, we really do not have a good
understanding of what makes an employee good. We really do not have a
systematic understanding of why some employees are good while others are not
and what drives productivity. For instance, consider whether teams make
employees more productive. Some employees may not perform well when working in
isolation, but exceed expectations when grouped together into a team. Who
should we group together? Similar or dissimilar people? What works better? We
really don’t know. The advantage of the new scenario is we can now track who
works with whom, how they interact with each other, and see whether
productivity was higher or lower under different team configurations. Also
interesting is the fact that all of that information can be combined with
traditional HR data on who received a high salary, who was given a bonus, who
received a commission. And you can match these data with actions employees
took. Then you can ask questions like, “If you give this person so much
commission, will you increase his productivity this much?” In other words, we
can now quantitatively assess the value of a commission.
2. In what other areas do you feel Big Data
can be particularly useful in improving corporate performance? How can managers
and staff systematically collect and analyze such data?
One analogy offered by a colleague is that big data is
a natural resource like oil. Oil in its original form is not useful. It first
has to be extracted and stored. Then, it has to be cleaned. Ultimately, it must
be refined in a way making it useful for consumption in a vehicle or jet
engine. Data are similar. The first step is data collection. But simply
collecting and storing large quantities of data somewhere is not useful. The
data need to be systematically cleaned and rendered in a form suitable for
answering a business question. And finally - and I think this is the really big
gap for an organization - data in and of itself does not “speak” or “tell a
story”. It has no value until it is combined with some business and domain
knowledge and interpreted correctly. The design of the right experiment or the
right question to ask, and when to use a database to broach it, and when not to
in order to avoid so called “analysis paralysis,” are all managerial questions.
So, I think the mere collection of data is not going
to change much. It’s when the data are used by intelligent individuals who
understand how to use analytics in combination with domain knowledge that it’s
going to be powerful. I think the most important thing for companies is to try
and find those kinds of individuals. This is not easy - the human-capital
scarcity is very real. There are good statisticians who may not know the
business context. There are people who are trained and credentialed in
particular aspects of business but who don’t have the training to think
quantitatively, or do not know how to think in a data-oriented manner. You need
to find people who can do both. Just one of the two pieces is typically insufficient.
Corporations need to focus their attention on this human-capital dimension.
Massive computing innovations are taking place in the
collection, cleansing, and visualization of data. Great software is available
for these tasks. But this data will not affect the business unless a line
manager uses this information to make a decision. I think the collection,
cleaning, and visualization aspects are easier to solve than this problem—that
is, understanding how this data can be used to drive the business forward. For
this, firms need to train existing managers, hire new people, and empower them
to make decisions as well. Universities have a role to play in proactively
providing more and better training, graduating more people with the requisite
skills. The effort will have to be collective. Presently, there is an acute talent
shortage.
3. Not too long ago, you conducted a Big Data
study of Facebook messages investigating the effect of social media content on
customer engagement. Can you summarize that study for us, highlighting key
findings? To what extent do you think they generalize to countries like Japan?
The purpose of the study was to learn about the role of advertising content
in the context of social media. To understand the background on the study, lets
ask: when a company spends a lot of money on advertising, what should consumers
infer? If the product were of low quality, but many people purchased the item
as a result of high advertising, the firm would be adversely affected. Thus,
the fact that the company is advertising heavily signals to consumers the
product is high quality. Punchline - the presence of heavy advertising by a
company suggests to consumers that the product is high quality. Knowing this,
only really good firms would advertise. Really bad firms would not, thereby
validating consumer thinking. This is the so-called signaling model of advertising
in a nutshell.
This model is unsatisfactory. It implies you do not necessarily have to
advertise heavily in order to get the advertising’s benefits. You simply need
to spend some money to signal that you are a good firm to consumers. Yet we see
that many companies spend significant resources in hiring ad-agencies and
creative directors and others to design compelling advertising messages; that
ad-creative firms have a lot of clients and are making money. Some ads center
on lifestyle features, others state prices. Some ads describe how good you will
feel if you use the product. Yet others simply furnish objective information. All
suggests that the ad-creative matters. Also, many people have tried out Pepsi and
Coke and know how they taste – they don’t need to see ads to know whether Coke
or Pepsi “work for them”. But we see that both companies continue to spend huge
sums advertising. This is clearly not simply about signaling. A limitation of
the canonical model therefore is its failure to assign a compelling role to the
content of the ads. This paper asks how the content of the advertising matters.
We worked with a company to collect a lot of posts that firms are
publishing on Facebook. Then, we used Amazon’s Mechanical Turk (a crowdsourced
labor platform) in combination with a natural language-processing algorithm to
create variables reflecting attributes of the content of the posts. Eg, does
the content mention prices? A sale? The brand? Do the posts contain emotional
content? Is this content emotionally positive or negative? Our key finding was
that information about prices and availability, what we called search attributes, were not effective in
generating engagement, which we defined as “Likes” and comments on facebook.
Emotional content, in contrast, generate a lot of engagement. From this
relatively large-scale study of over one-hundred thousand posts on Facebook, we
concluded that emotional content is extremely important for generating
engagement. In addition, though search attributes like prices alone do not seem
to generate engagement, coupled with emotion, they have a positive effect on
engagement. This main finding of the paper underscores the importance of
emotional or social content for generating interaction. Information provision
alone does not seem to work.
4. How do you think social networks in
specific and the internet in general have changed marketing? What three changes
do you feel are most significant? How well have companies adapted overall? What
can they do to better adapt?
Social networks have changed marketing in a huge way. Their primary
influence has been to create a spillover effect that amplifies the importance
of marketing messages. In the pre-social days, I could target a message to the
user, exerting an effect on that person. The message might affect other users,
but we had no way of knowing. Now, we can target a message at an influential
user. Not only will he or she potentially be affected by that message, but that
user might share the message with others because of the social nature of the
message. Individuals connected to this targeted individual are also affected by
the message, which they, in turn, will share with their friends. Thus, social
networks increase the efficacy of the impact of messages. Further, much of this
is measurable.
Social networks have also changed the power equation between consumers and
firms. In the past, if you had a terrible experience at a restaurant or on a
plane flight, you could write to the CEO, but then no one else will know of
your experience. Or you can tell a few of your friends at home. Now, you can post
your experience in Facebook for viewing by a vast circle of friends, some of
whom will respond indicating they have had a similar experience. They may
repost your experience and tweet it, too. Then, their friends will read about your experience. It may be included
as part of a thread in a trending story in Twitter, in which case millions of
people may eventually learn about your experience. In this way, social networks
have increased the power of the consumers. They now have the ability to explain
their viewpoint to a large number of people, a form of influence once enjoyed
only by firms with the resources to advertise. That gives consumers more of a
voice.
What can firms do to adapt? I think measurement is particularly important.
Tracking technology has improved. Advertising channels and social-media options
have proliferated. You can advertise on Twitter, post ads on Facebook. You can
select from Facebook social ads to ad feeds. Paid content and social content
are also options as well as Youtube. Measurement technology in the digital
world is advanced compared to the offline world. A large number of firms now
sell sophisticated measurement products. The problem, I believe, is on the people
side. Human capital in marketing organizations is currently inadequate to
effectively leverage all of these tools. People do not know how to use them.
They are overwhelmed by the volume of data flowing in and by the proliferation
of available options. The first step firms can take is to improve their
in-house measurement capability and hire people who understand these tools.
Firms can demand better measurement from agencies providing data to them as
well.
5. Google coined the neologism ZMOT—Zero
Moment of Truth. In contrast to the classic First Moment of Truth, the five to
seven seconds a consumer sees the product on the supermarket shelf, deciding
whether or not to purchase it, Google believes consumers now decide before they
go based on online research. To what extent do you agree with this Googly
notion? To the extent that you agree, what do you think the implications of
ZMOT are for company marketing departments? How will they have to alter their
marketing strategies?
As an academic, I hesitate to give uni-causal explanations for multi-causal
phenomena. Shopping behavior - where and how we choose to buy - is a complex
process. It involves obvious considerations related to price, convenience, and
mental associations to other brands as well as word of mouth and advertising. I
am reluctant to buy into the First Moment of Truth or ZMOT notions for this
reason. I do think both concepts apply, but I do not think either applies
completely to the exclusion of other considerations.
Compared to the past, the sheer magnitude of information and the ability to
access it on the go with a smart phone have changed behavior significantly.
Many consumers are no longer influenced by conventional lifestyle advertising
or simple branding. If you think about what a well-known brand is in the first
place, it is a cue that the product is very good. It comes with the promise
that the quality is good. It tells you that you will not be cheated. If I am
time constrained or cannot find information related to a purchase, I will trust
the well-known brand. Purchasing a well-known brand is lower risk in this case.
Because information is now easy to find and access, this impact is reduced.
In some sense, information in lifestyle advertising and information in branding
are substitutes. As information becomes easier to access, we will access
information from our networks. In that sense, I agree with the Google folks.
Lots of decisions are not driven by cues that you see in the stores, but by
information about products that consumers have researched.
Two strategies will be important in adapting to this new reality. One is to
think deeply about the marginal value of time for people. Time-constrained
people may not find reading lots of information and checking prices an optimal
course of action. The information-acquisition cost is too high for them. Other
individuals may not be all that hungry for information in certain contexts. In
other contexts, they might actively search it out. For this reason, it is important
for companies to target the content and timing of their messages appropriately.
The mass marketing approach has changed to one of a heterogeneity of approaches
targeting to consumers for whom the brand matters.
Given that consumers are more active in seeking and using information, the
second strategy is for companies to play a role in providing that information.
Good companies now need to provide a page explaining what makes them a good
company. Simply stating you are a good company is no longer effective. You now
have to explain why. Companies can create their own web and product-information
pages. Companies need to be in the channels where consumers acquire
information. Many consumers type key words in the Google search engine.
Companies need to be there. This is the second way companies can respond to the
proliferation of information and the reduced transaction cost of acquiring
it.
