How to Predict Demand for Your New Product
Skip to content
Operations Mar 10, 2017

How to Pre­dict Demand for Your New Product

Rely­ing on man­ag­er exper­tise and mar­ket research may not be enough.

By matching historical sales data with appropriate product life cycle curves, one can reduce new product forecasting errors.

grivina via iStock

Based on the research of

Kejia Hu

Jason Acimovic

Francisco Erize

Doug Thomas

Jan A. Van Mieghem

Launch­ing the next line of lap­tops, routers, or sur­veil­lance cam­eras is crit­i­cal for a tech com­pa­ny that does not want to be left behind.

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.

But that doesn’t make it any eas­i­er to turn a prof­it off of them. 

One of the biggest chal­lenges that com­pa­nies face is pre­dict­ing demand for new prod­ucts over time. Over­es­ti­mate it, and risk ware­hous­es full of excess inven­to­ry. Under­es­ti­mate it, and your cus­tomers could leave emp­ty hand­ed — or you might be left with a hefty bill for expe­dit­ed delivery. 

Imag­ine you have a crys­tal ball and you know exact­ly how demand for your prod­uct will go up or down, month by month,” says Jan Van Mieghem, a pro­fes­sor of oper­a­tions at the Kel­logg School. That would make it very easy to pre­pare to meet demand, because if you know your lead times, you just use your crys­tal ball to source the right num­ber of units from the cheap­est source on time, and you can sat­is­fy 100% of demand with no waste. That is the Holy Grail.” 

But crys­tal balls are dif­fi­cult to come by. So Van Mieghem decid­ed to find the next best thing.

In a new study led by Van Mieghem and Doug Thomas of Penn State Uni­ver­si­ty, researchers part­nered with Dell to ana­lyze sales data from over a hun­dred of the company’s prod­ucts. While com­pa­nies often rely pri­mar­i­ly on man­agers’ expe­ri­ence and mar­ket research, the team found that com­pa­nies like Dell could use his­tor­i­cal data from pre­vi­ous prod­ucts to improve fore­cast­ing accu­ra­cy on new prod­ucts by as much as 9%. This trans­lates into mil­lions of dol­lars of savings. 

The Cost of Get­ting Life-Cycle Fore­casts Wrong 

New prod­ucts rep­re­sent 27% of sales across all indus­tries. Giv­en the stakes, it should come as no sur­prise that busi­ness­es invest in fore­cast­ing the life­cy­cles of their products. 

Small fore­cast­ing errors can mean a big hit on prof­its,” says Van Mieghem. When com­pa­nies like Dell face stock-outs, they do emer­gency air ship­ments or use a faster local source such as Mex­i­co instead of Chi­na. Those options are more expen­sive than their tra­di­tion­al sourcing.” 

For exist­ing prod­ucts, com­pa­nies gen­er­al­ly use data on pre­vi­ous sales to cre­ate, or at least influ­ence, a fore­cast for the next sales period. 

But new prod­ucts are trick­i­er, and pre­vi­ous research has found that most firms do not rely pri­mar­i­ly on his­toric sales data. Instead, they lean heav­i­ly on qual­i­ta­tive mar­ket research, or the opin­ions of exec­u­tives. In that sit­u­a­tion,” Van Mieghem says, you typ­i­cal­ly rely on man­age­ment insight and expe­ri­ence, which is not real­ly data-driven.” 

Yes, there is the occa­sion­al prod­uct that is tru­ly ground­break­ing. But what if most new com­put­ers or cam­eras were not tru­ly new? What if, despite an upgrade here, or a new fea­ture there, they resem­bled past prod­ucts in key ways? Wouldn’t this sug­gest that his­tor­i­cal data for oth­er prod­ucts could be used in new prod­uct forecasting? 

Van Mieghem and Thomas teamed up with Kejia Hu of Kel­logg, Jason Aci­movic of Penn State, and Fran­cis­co Erize of Dell to find out. The researchers came up with the idea of cre­at­ing clus­ters of prod­ucts that had sim­i­lar prod­uct life-cycle (PLC) curves — lit­er­al­ly the shape of the curve when the product’s demand over time is made into a graph. Then they could make pre­dic­tions about which clus­ter” a new prod­uct would fall in, based on the pre­vi­ous prod­ucts it most resembled. 

The idea is that for each prod­uct clus­ter we can find the prod­uct life-cycle curve that fits it best and use this curve to fore­cast demand for the new prod­uct,” Van Mieghem says. If a new prod­uct is exact­ly like an old one, you can use the curve just for that prod­uct to pre­dict sales. But if it doesn’t have an exact match, you can use the curve for the clus­ter it fits into.” 

What if most new com­put­ers or cam­eras were not tru­ly new? What if, despite an upgrade here, or a new fea­ture there, they resem­bled past prod­ucts in key ways?

Com­par­ing Curves 

Armed with data on 133 Dell com­put­er prod­ucts, the researchers first worked to find out which prod­uct life-cycle curves best fit his­tor­i­cal sales pat­terns, start­ing with the his­tor­i­cal sales of indi­vid­ual products. 

Some of the curves they con­sid­ered were more typ­i­cal, hav­ing four phas­es: an intro­duc­tion, a ramp-up of demand, a more sta­ble peri­od in which the prod­uct is con­sid­ered mature,” and final­ly decline. The researchers also includ­ed curves that were more lin­ear in their analy­sis. These curves resem­bled tri­an­gles and trape­zoids. Their goal was to find a curve that fit the his­toric data well — but not so well that the curve was too spe­cif­ic to that par­tic­u­lar product. 

The researchers found that the best prod­uct life-cycle curve for the major­i­ty of the prod­ucts was not curved at all. It was a tri­an­gle. Demand goes up, then it goes down,” Van Mieghem says. 

This sug­gests that these prod­ucts expe­ri­ence very lit­tle in the way of a mature peri­od. Which makes sense for elec­tron­ics, since there is gen­er­al­ly a new­er, faster, flashier mod­el out before the pre­vi­ous prod­uct has run its course. 

This is very attrac­tive man­age­ri­al­ly,” he says. If you’re a man­ag­er try­ing to pre­dict demand for a new prod­uct and I tell you it’s tri­an­gu­lar in shape, you only need to esti­mate three num­bers: how long you think the prod­uct will sell for, when the peak sales will occur, and how high the peak will be.” 

Jan Van Mieghem teach­es pro­grams in oper­a­tions strat­e­gy, lean oper­a­tions, and sup­ply chain man­age­ment through Kel­logg Exec­u­tive Education.

Next the researchers cre­at­ed clus­ters of prod­ucts with sim­i­lar life cycle curves and found the opti­mal life-cycle curve for the entire clus­ter. They then used this curve — in addi­tion to oth­er infor­ma­tion about the prod­uct that a com­pa­ny might have, such as the sea­son of launch, the planned end of life, and the esti­mate of total PLC demand — to scale the curve and recre­ate a demand fore­cast for each product. 

A Bet­ter Way 

Over­all, the researchers’ fore­casts were about 9% more accu­rate (in terms of Mean Absolute Error, or MAE) when com­pared against the his­tor­i­cal data than Dell’s orig­i­nal fore­casts. This is a siz­able improve­ment that could dri­ve trans­porta­tion and inven­to­ry sav­ings of $2 – $6 per unit on mil­lions of prod­ucts a year. 

Van Mieghem believes the tri­an­gu­lar life-cycle curve should be a good fit for any prod­uct with­out a sig­nif­i­cant matu­ri­ty phase, such as clothes or acces­sories pegged to the lat­est fash­ions, as well as elec­tron­ics. In addi­tion, the clus­ter” approach that the researchers used to iden­ti­fy the best curve should be effec­tive for a wide range of new prod­ucts that resem­ble old­er ones. 

The key, accord­ing to Van Mieghem, is that com­pa­nies like Dell can achieve greater fore­cast accu­ra­cy by com­bin­ing data-dri­ven approach­es with knowl­edge and busi­ness insight from demand planners. 

Using his­tor­i­cal data of pre­de­ces­sors, our method­ol­o­gy will sug­gest the most appro­pri­ate shape of the life cycle curve for a new prod­uct in a par­tic­u­lar busi­ness. A plan­ner can then use that data-dri­ven prod­uct life­cy­cle curve as a start­ing point and adjust with their par­tic­u­lar insights about the spe­cif­ic prod­uct, such as when they know about an upcom­ing pro­mo­tion­al push.”

Featured Faculty

Jan A. Van Mieghem

Harold L. Stuart Distinguished Professor of Managerial Economics, Professor of Operations

About the Writer

Sachin Waikar is a freelance writer based in Evanston, Illinois.

About the Research

Hu, Kejia, Jason Acimovic, Francisco Erize, Doug Thomas, and Jan A. Van Mieghem. 2017. “Forecasting Product Life Cycle Curves: Practical Approach and Empirical Analysis.” Working paper.

Read the original

Suggested For You

Most Popular


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.


Why Warmth Is the Under­ap­pre­ci­at­ed Skill Lead­ers Need

The case for demon­strat­ing more than just competence.

Most Popular Podcasts


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.


Pod­cast: Attract Rock­star Employ­ees — or Devel­op Your Own

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


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.