%ontents Introduction................................................................................................................................ 1 Business Objectives................................................................................................................ ...1 Data Mining Objectives.............................................................................................................1 Data Set......................................................................................................................................1 Data Modeling............................................................................................................................1 Model in SaS Enterprise Miner 13.1..........................................................................................1 1.
Decision tree – 2 wa......................................................................................................2
2.
Decision tree – 3 wa......................................................................................................3
&o%paring regression wit# and wit#out trans$or%ed variables............................................+ &u%ulative li$t c#art..........................................................................................................+ Evaluation o$ %ultiple %odels................................................................................................... , &u%ulative li$t and cu%ulative percent response values at )- percentile........................., ro$itabilit o$ a roactive *etention lan................................................................................./ ossible Incentives O$$ered........................................................................................................0
Cell2Cell: The Churn Game &ntrodu'tion %e$$2%e$$ is the ( th $ar)est *ire$ess 'o!pan" in the +S, )iin) seri'e to near$" 10 !i$$ion subs'ribers, serin) !ore than 210 !etropo$itan !arkets . 200 'ities /'oerin) near$" a$$ 50 states #he 'o!pan" is 'urrent$" a'in) a !ajor prob$e! o 'usto!er 'hurn e are usin) SAS M 43 to dee$op a !ode$ or predi'tin) 'usto!er 'hurn at %e$$2%e$$
Business bje'ties 1 #o dee$op a 'hurn !ana)e!ent pro)ra! to redu'e the 'usto!er 'hurn b" deisin) innoatie in'entie p$ans 2 &!proe pro6tabi$it"
ata Set #he )ien data set 'onsists o 71,047 ro*s . 'ontainin) a tota$ o 78 ariab$es /in'$udin) a ariab$e na!ed 9%:+;<=, si)ni"in) *hether the 'usto!er had $et the 'o!pan" t*o !onths ater obseration ne o the ariab$es na!ed 9%A>&B;A#= *as used to di?erentiate the a$idation dataset ro! trainin) dataset #rainin) dataset 'ontained data o 40,000 'usto!ers and a$idation dataset 'ontained 31,047 'usto!ers
ata Mode$in) #ota$ o ( di?erent !ode$s *ere used to predi't the 'hurn o 'usto!ers #hese !ode$s *ere@ e'ision #ree /binar" e'ision #ree /three *a" tree
Mode$ in SaS nterprise Miner 131
%o!parin) re)ression *ith and *ithout transor!ed ariab$es %u!u$atie $it 'hart
a$uation o !u$tip$e !ode$s %u!u$atie $it and 'u!u$atie per'ent response a$ues at 50 per'enti$e
As 'an be seen ro! tab$e, the peror!an'e o 9;e)ression *ith transor!ed ariab$es= is best a!on) the di?erent te'hniCues used
%u!u$atie $it a$ue at 10 per'enti$es or re)ression *ith transor!ed ariab$es is 23478
Pro6tabi$it" o a Proa'tie ;etention P$an +sin) re)ression !ode$ *ith transor!ed ariab$es as inputs, the o$$o*in) a$ues are 'a$'u$ated Assu!ption@ Subs'riber in the 1 st de'i$es is tar)eted D E Base $ine 'hurn rateE 1( F E >it E 23478 H E Su''ess rate E -1E 13478 >%E >ieti!e a$ue o 'usto!er %E %ost o in'entie Pro6t E
Probabi$it" o %hurnI Su''ess ;ateI >%-%
E DI I HI >% % >% E Month$" ;eenues I /1JrK/1Jr-;etention ;ate rE dis'ount rateE 10F #ota$ aera)e !onth$" 'hurn rateE 2F ;etention ;ate /annua$ E /1-/04I12E 07( Aera)e Month$" reenues per 'usto!er E 588528 Aera)e >% per 'usto!er E 588528I/1J1K/1J1-7( E 1040(
#hus %e$$2%e$$ 'an spend a !aLi!u! o 1040( on a 'usto!er #ota$ nu!ber o 'usto!ers E 10,000,000 subs'ribers +sin) 1 st de'i$es, *ho hae hi)hest probabi$it" o 'hurnin)@ Aera)e in're!enta$ reenue HI/Aera)e >%
E
E /10F o 10 !i$$ionI1(I23478I13478I1040( E 1181 bi$$ion #hus %e$$2%e$$ 'an a?ord a !aLi!u! in'entie 'ost o #.#$# !illion.
Possib$e &n'enties ?ered Based on aboe deried i!portant ariab$es, the o$$o*in) in'enties p$an 'an be o?ered to the 'usto!ers to redu'e the possib$e 'hurn •
•
•
•
ro! the !ode$ *e )ot NPAOS as one o the pri!ar" a'tors or 'hurn predi'tion &t !akes business sense, as a 'usto!er *ho 'han)es his o$d 'e$$ phone is $ike$" to 'hurn, be'ause !an" !obi$e seri'e proiders )ie ne* 'e$$ 'onne'tion bund$ed *ith 'e$$ phone ro! de'ision tree *e 'an see that ater 304 da"s 'usto!ers 'han)e their handset So the 'o!pan" 'an 'o!e up *ith a p$an o o?erin) 'usto!ers ne* o 'e$$ phone /at a pri'e s$i)ht$" hi)her than the 'ost pri'e *ithout 'han)in) their 'onne'tion #his o?ered ater !onths #his o?er *i$$ a$$o* the 'o!pan" to retain their 'usto!ers /ensurin) uture reenue *ithout an" in're!enta$ 'ost ro! the de'ision tree it is isib$e that the probabi$it" o 'hurn in'reases i NPAOS is !ore than 304 and %:A% 'usto!ers, the 'o!pan" shou$d o?er $itt$e !ore in'enties to su'h 'usto!ers #his is be'ause een i in the short ter! the 'o!pan" in'urs !ore 'ost, but retainin) su'h 'usto!ers *i$$ in'rease 'ash o* to the 'o!pan" in the $on) run