Want to Improve Your Sales Forecast? Check Your Company’s Facebook Feed.
Skip to content
Operations Data Analytics Strategy Sep 6, 2017

Want to Improve Your Sales Fore­cast? Check Your Company’s Face­book Feed.

Social media data can help pre­dict con­sumer demand.

Social media data can help sales forecasts.

Morgan Ramberg

Based on the research of

Ruomeng Cui

Santiago Gallino

Antonio Moreno-Garcia

Dennis J. Zhang

You’re scrolling through Face­book and see a post from your favorite cloth­ing store show­cas­ing a great pair of jeans. You like” it, per­haps even leave a com­ment that you are eager to buy a pair, and then scroll on.

Add Insight
to your inbox.

We’ll send you one email a week with content you actually want to read, curated by the Insight team.

How is that store putting your likes and com­ments to use? It’s prob­a­bly using them to shape its social media mar­ket­ing strat­e­gy. But it is much less like­ly that the retail­er is using that data to make oper­a­tions deci­sions, such as how many pairs of those jeans to man­u­fac­ture or whether to mark down prices.

That could change. In a recent study, Anto­nio Moreno, an asso­ciate pro­fes­sor of oper­a­tions at Kel­logg, found that social media data can improve sales fore­casts. When researchers incor­po­rat­ed infor­ma­tion about a cloth­ing company’s Face­book inter­ac­tions into pre­dic­tion mod­els, they could more accu­rate­ly esti­mate pur­chas­es the fol­low­ing week. 

Using advanced algo­rithms was key to the improve­ments, mean­ing sim­ply col­lect­ing social media data is not enough: com­pa­nies should also upgrade their fore­cast­ing techniques.

It’s impor­tant to get new data but also use more sophis­ti­cat­ed pre­dic­tion method­ol­o­gy,” Moreno says.

The study did not reveal why the Face­book infor­ma­tion improves fore­casts. Moreno spec­u­lates that the data might reflect how much atten­tion cus­tomers are pay­ing to the brand, as well as good or bad word of mouth. 

By intro­duc­ing social media data, we can do better.”

But com­pa­nies may care more about the technique’s effec­tive­ness than the mech­a­nism behind it.

If some­thing works,” he says, some­times they might be able to live with­out know­ing why it works.”

Using Social Media Data

The idea of min­ing social media data to guide oper­a­tions is in its infancy.

For instance, com­pa­nies might show a forth­com­ing shirt in two col­ors and see which one gen­er­ates more clicks. The com­pa­ny could then use that infor­ma­tion to decide which col­or to pro­duce. But it’s still not main­stream,” Moreno says.

And while aca­d­e­m­ic stud­ies have explored whether social media posts boost sales, lit­tle research has been done on using the data for inter­nal oper­a­tions decisions.

Moreno decid­ed to inves­ti­gate that idea with Ruomeng Cui, at Emory Uni­ver­si­ty, and Den­nis Zhang, at Wash­ing­ton Uni­ver­si­ty, both of whom are for­mer Kel­logg doc­tor­al stu­dents, along with San­ti­a­go Galli­no at Dart­mouth College.

The team worked with an online cloth­ing com­pa­ny. Most of the firm’s social media – dri­ven traf­fic came from its Face­book page, which had more than 300,000 fol­low­ers at the time of the study. But to fore­cast sales, the com­pa­ny was large­ly rely­ing on basic infor­ma­tion such as its over­all sales growth and week­ly or sea­son­al pat­terns — such as the ten­den­cy to sell more on weekends. 

Moreno’s team wrote soft­ware to extract infor­ma­tion about the company’s Face­book posts from Jan­u­ary to July 2013. The final data set includ­ed more than 171,000 users, 1,900 com­pa­ny posts, about 25,000 com­ments, and a quar­ter-mil­lion likes.

Next the researchers used lan­guage-pro­cess­ing soft­ware to cat­e­go­rize each com­ment as pos­i­tive, neg­a­tive, or neu­tral. In addi­tion, the team obtained inter­nal data on the company’s sales and adver­tis­ing cam­paigns dur­ing that time.

Train­ing the Fore­cast­ing Models

Using what they gath­ered, the researchers pro­duced two sets of sales-fore­cast­ing mod­els: the base­line fore­cast, which includ­ed only inter­nal com­pa­ny infor­ma­tion, and a sec­ond fore­cast that com­bined inter­nal and social media data.

For both the base­line and social media fore­casts, the team exper­i­ment­ed with a vari­ety of pre­dic­tion meth­ods. Most of the mod­els relied on machine learn­ing, in which the mod­el trains itself to iden­ti­fy which fac­tors are most important.

To assess accu­ra­cy, the researchers used a mea­sure called mean absolute per­cent­age error (MAPE), which cap­tures how much the esti­mate devi­ates from actu­al sales. For instance, a MAPE of 10% would mean that, on aver­age, the model’s esti­mates were 10% off.

The company’s exist­ing sales fore­casts for the next week had a MAPE of 12%. The researchers’ best-per­form­ing base­line mod­el — the one with­out the social media data — brought the error down to about 7 – 9%.

They can actu­al­ly use this social media to learn and make bet­ter decisions.” 

Adding social media data low­ered it even fur­ther to 5 – 7%. Yet, the social media data alone was not enough. When the team plugged social media infor­ma­tion into a poor­ly per­form­ing mod­el, the accu­ra­cy could be even worse than the base­line mod­el with­out the social media information.

The results sug­gest that both the data and meth­ods are impor­tant. By intro­duc­ing social media data, we can do bet­ter,” Moreno says. But it looks like the first step should be hav­ing bet­ter methods.”

Fine-Grained Fore­casts

Future research could explore in more detail why social media improves sales fore­casts. Researchers could also per­form sim­i­lar stud­ies to pre­dict sales for indi­vid­ual prod­ucts, rather than just total sales. And if data could be bro­ken down by geo­graph­ic area, the infor­ma­tion could help com­pa­nies decide how much of a par­tic­u­lar prod­uct to car­ry in, say, Texas ver­sus Idaho.

Moreno notes that the study’s results may not apply to all indus­tries. Social media data is more like­ly to be rel­e­vant to prod­ucts with high­ly uncer­tain sales or indus­tries heav­i­ly influ­enced by trends, such as fash­ion and enter­tain­ment. But for con­sumer goods like break­fast cere­al, sales are already fair­ly pre­dictable, so adding Face­book data may not improve fore­casts much.

Com­pa­nies could also become more strate­gic about their social media posts, in order to specif­i­cal­ly elic­it infor­ma­tion that will help guide their oper­a­tions. For instance, more firms might adopt the prac­tice of dis­play­ing poten­tial prod­ucts and decid­ing what to man­u­fac­ture based on cus­tomers’ reactions. 

They can actu­al­ly use this social media to learn and make bet­ter deci­sions,” Moreno says.

About the Writer

Roberta Kwok is a freelance science writer based near Seattle.

About the Research

Cui, Ruomeng, Santiago Gallino, Antonio Moreno, and Dennis J. Zhang. “The Operational Value of Social Media Information.” Production and Operations Management. doi: 10.1111/poms.12707.

Read the original

Suggested For You

Most Popular

Organizations

How Are Black – White Bira­cial Peo­ple Per­ceived in Terms of Race?

Under­stand­ing the answer — and why black and white Amer­i­cans’ respons­es may dif­fer — is increas­ing­ly impor­tant in a mul­tira­cial society.

Leadership

Why Warmth Is the Under­ap­pre­ci­at­ed Skill Lead­ers Need

The case for demon­strat­ing more than just competence.

Most Popular Podcasts

Careers

Pod­cast: Our Most Pop­u­lar Advice on Improv­ing Rela­tion­ships with Colleagues

Cowork­ers can make us crazy. Here’s how to han­dle tough situations.

Social Impact

Pod­cast: How You and Your Com­pa­ny Can Lend Exper­tise to a Non­prof­it in Need

Plus: Four ques­tions to con­sid­er before becom­ing a social-impact entrepreneur.

Careers

Pod­cast: Attract Rock­star Employ­ees — or Devel­op Your Own

Find­ing and nur­tur­ing high per­form­ers isn’t easy, but it pays off.

Marketing

Pod­cast: How Music Can Change Our Mood

A Broad­way song­writer and a mar­ket­ing pro­fes­sor dis­cuss the con­nec­tion between our favorite tunes and how they make us feel.