The MustHave Guide to Accurate Call Center Forecasting - PDF document

1 The Must-Have Guide to Accurate Call Cen
The Must-Have Guide to Accurate Call Center Forecasting The Must-Have Guide to Accurate Call Center Forecasting | 2The importance of measuring variabilityAre you missing any critical factors?Get the data rightThe FoundationsContents The Must-Have Guide to Accurate Call Center Forecasting | 3ForewordCharles Watsoninjixo WFM An � ctiv for cast for an op rations is lik having a h althy h art. Your for cast tak s in data. It th n pumps out insight for your sch duling and s rvic l v l manag m nt functions. Ev rything is pr dicat d on g tting th for cast right. For casting is th most chall nging function in workforc manag m nt. At tim s, accurat data can b a chall ng . Oth r tim s, busin ss information do sn‘t always mak it back into th proc ss. A for cast r is l � to mak judg m nt calls and tak accountability for th m. G tting ducat d on th principl s of � ctiv for casting is a gr at way to s t yours lf up for succ ss. Th r ar ways to for cast, r for cast, and addr ss volatility in a call c nt r op rations that hav

2 b n us d in th b st op rations for y
b n us d in th b st op rations for y ars. This book will h lp for cast rs of all l v ls lock down th basics of call c nt r for casting, and inj ct n w id as into your op rations. The Must-Have Guide to Accurate Call Center Forecasting | 4 The foundationFor casting is th corn rston of WFM. If you g t th for cast wrong, th whol proc ss is unstabl . It th r for mak s s ns that for casting tak s c nt r stag in m asuring WFM t am p rformanc . The Must-Have Guide to Accurate Call Center Forecasting | 5Are you taking care of all the elements ofthe forecast?It‘s not just about th volum of work (numb r of calls, mails or chats). Don‘t forg t th av rag amount of tim required to complete each unit like the average handle tim (AHT) and th total workload (volum multipli d by AHT). And mak sur you ar using th most appropriat m trics – . g. o� r d calls not answ r d calls. While it is common for centers to analyze the call volume for cast, it is l ss common to s analysis of th AHT. Call volume forecast and AHT are equal partners in

3 the workload calculation and should be
the workload calculation and should be analyzed separately as w ll as in combination. Combin d, th y h lp to id ntify opportuniti s for accuracy improv m nt. You should question each metric and understand underlying caus s. For xampl , in som c nt rs, AHT is r lativ ly constant across all tim fram s. In oth r contact c nt rs, th r is larg variation .g. AHT during th night shi� is much long r than during th day. Your goal will b to und rstand why AHT is inconsist nt. Is it a function of l ss sup rvision during night shi�s? Ar mor n w hir s on unattractiv shi�s? Ar custom rs calling with mor di�icult probl ms wh n ag nts hav tim to talk?Digging into th s typ s of qu stions can improv th accuracy of forecasting the actual workload in each time p riod of th day. Th y can also h lp to id ntify opportuniti s for r ducing AHT through oth r m ans.Did you select the correct timeframesfor analysis? The period over which you analyze forecast accuracy is important. Analyzing accuracy at th monthly or w kly l v

4 l s rv s as a r asonabl scor card. How
l s rv s as a r asonabl scor card. How v r, it do s littl to h lp you to discover where the forecast may be consistently over or und r th actual d mand.It is similarly misl ading to m asur s rvic l v l or av rag sp d of answ r (ASA) ov r long p riods. Th r can b dramatic �uctuations within th w k or month that o�s t ach oth r, making th ov rall av rag look good. The Must-Have Guide to Accurate Call Center Forecasting | 6 You n d to do th analysis at th int rval l v l ( .g. 30 minutes) to focus attention on those elements of the for cast that can b improv d. D aling with wid swings at the daily or half-hourly interval level puts an unrealistic d mand on th Op rations t am. Th goal is a consist ntly high l v l of accuracy, not an high av rag l v l ov r tim .Are you using the proper methodology for analyzing accuracy?P rc nt varianc , standard d viation and corr lation co �ici nts can all b us d to id ntify patt rn anomali s and m asur accuracy. Th most commonly used analysis of forecasting accuracy is the f

5 ollowing: You calculat th p rc ntag b
ollowing: You calculat th p rc ntag by which th actual call volumes or AHT that you record throughout the day vary from the previously forecasted volumes and AHT. Th formula for this is: (Forecasted value - actual value) / forecasted value. At th int rval l v l, it is also a pr tty accurat pictur . Som m thods for presenting the analysis of percent variation is shown in th charts. Forecast Interval Accuracy May 5% Interval 10% Interval 15% Interval 20% Interval +25% Interval June July Aug Sept Oct Weekly Interval Accuracy eeks The Must-Have Guide to Accurate Call Center Forecasting | 7 Wh r th r is a w alth of data to analyz , it is h lpful to hav an asi r way to put a �ng r on th puls of th accuracy ov r a long p riod of tim . This is b st found by calculating the standard deviation of the variation p rc ntag s. A small d viation is d sir d rath r than wid swings in variation. Th standard d viation calculation will be revealing even if the average variation seems quite Another tool to analyze variations over time is correlation co &

6 #x00660066;ici nts, which analyz patt r
#x00660066;ici nts, which analyz patt rns from on p riod to anoth r. Th corr lation co �ici nt analysis can b applied to the variation percentages but is most useful when applied to the arrival patterns of work volume and th chang s in AHT ov r th int rvals. It compar s two p riods to s if th patt rns ar a match or not. For xampl , th typical Monday might adh r to a r lativ ly consist nt patt rn, but th corr lation analysis may reveal that one particular Monday varies in pattern even if the total volume of workload is within normal boundari s. This would sugg st that furth r und standing of what happ n d on that Monday is us ful. The level of detail is also critical to determining which historical data is “normal” and which is not. This h lp you when deciding to allow the data to factor into the history k pt for for casting. Data which is outside of an acceptable range Consid r d for adjustm ntStor d s parat ly as a sampl of a particular repeatable eventDiscard d as an anomaly unlik ly to r cur (such as a pow r outtag ) The Must-Ha

7 ve Guide to Accurate Call Center Forecas
ve Guide to Accurate Call Center Forecasting | 8Are you collaborating with colleagueswhos actions can in�u nc workload? P opl outsid of th WFM t am hav in�u nc on for cast accuracy. Any anomali s in th actual workload should b id nti� d and tagg d with a r ason. Th s r asons includ : mark ting campaigns, mailings, billing cycl s, and oth rs. By tagging th m, futur for casts ar b tt r abl to accommodat and pr dict th s driv rs.A good WFM team will make sure that it communicates � ctiv ly with oth r d partm nts. In most cas s, full und rstanding of what mak s custom rs and sta� b hav di� r ntly from usual is ss ntial to improving th accuracy of th for cast. Is this som thing that your t am does on a regular basis? The Must-Have Guide to Accurate Call Center Forecasting | 9 Get the data rightManaging th data �ow to th WFM syst m is on of th important tasks of th WFM analyst. Th Must-Hav Guid to Accurat Call C nt r For casting | 10Managing th �ow of d

8 ata into a WFM syst m is truly a cas ga
ata into a WFM syst m is truly a cas garbag in, garbag out. The whole process starts with a for cast that is bas d on history. If that history is inaccurat , th whol proc ss falls apart. Check the accuracy of your for casts. For example: If you ar typically mor than 5% o� for th day, a r vi w of th historical data and the process of storing it may be a worthwhil proj ct. Th r is no s ns in �n -tuning sch dul s to m t a proj ction that is not accurat to start with. Inst ad, g tting a better match of plan-to-actual will minimize the daily reactionary processes. Th s r actionary proc ss s incr as str ss and impact custom r s rvic . On of th b n �ts of an automat d workforc management system is the data link between the Workforc Manag m nt (WFM) tool and th Automatic Call Distributor (ACD)r porting syst m.This data link, or int gration, automat s th loading of data containing contact volum , handl tim s, ag nts logg d in and mor . With this int gration, th WFM t am do s not hav to k y in th data manually. This

9 can b a gr at tim sav r, but it is a
can b a gr at tim sav r, but it is a proc ss that must b manag d to nsur th quality m ts WFM purpos s.Initial SetupDuring th initial impl m ntation of a n w WFM systemyour IT or t l coms coll agu s will work with th WFM v ndor to s t up th data link. This includ s th physical connection between the systems and the format of the data that will b transmitt d.The physical connection will be within your corporate n twork if your ACD and your WFM syst m ar install d on your pr mis s. If th ACD is on-pr mis s and WFM in th cloud, th v ndor will provid middl war to s cur ly transmit data to th cloud. If both ar in th cloud th n a s cur , direct connectionMost of th major v ndors hav work d out th int rfac proc ss s. Th standard m thodologi s for impl m ntation ar w ll known to th WFM and ACD v ndors, so The Must-Have Guide to Accurate Call Center Forecasting | 11there should be no need for this to be a bottleneck in d ploying WFM.If a r al-tim adh r nc modul is o� r d by your WFM syst m, th data link for th r al-tim data fro

10 m th ACD will probably r quir a s para
m th ACD will probably r quir a s parat link to b stablish d. This second link provides half or quarter hourly contact data. This may r sult in additional charg s from th ACD v ndor.Nowadays, phon calls ar just on of th chann ls that contact c nt rs hav to handl . Th good n ws is that when it comes to connecting to systems that route mails, tick ts, w b chats and social m dia, th principl s ar th sam as for calls. It is normally possibl to coll ct volum and handl tim data.Care must be taken in setting up automatic feeds since the handle time of an email or chat session may not be as easy to identify as the handle time for a phone call. All tim s, handl tim data is not availabl and it is n c ssary to us a constant Av rag Handling Tim (AHT) to calculat sta�ing.In this cas , you should do som sampling of actual handling tim s to stablish a starting point. Th n, you can calibrate the constant AHT value by monitoring the ability of the agents to complete the non-call work within th r quir d tim scal . For xampl , if insu

11 60066;ici nt ag nts ar sch dul d, you s
60066;ici nt ag nts ar sch dul d, you should incr as th AHT valu . Ensure that the vendors work out all the details and that you understand the limitations of the information your syst m will r c iv . Also, g t an und rstanding of th rol s you must train your ag nts to p rform. Using the right work states for non-call contacts will be as important as using th a� r-call work stat for phon calls. Without accurat r cording, your av rag handl tim will b o�.You can also link th WFM syst m to oth r call c nt r syst ms that n d sch dul data. th s may includ a payroll syst m, tim clock, quality monitoring syst m, or -l arning syst m. Th s syst ms typically do not f d information into th WFM syst m. Inst ad, th y gath r information from it (although th data may �ow in both dir ctions). The Must-Have Guide to Accurate Call Center Forecasting | 12The format of the transmitted data is a critical part of the s tup. Each ACD r porting packag has its own format of data and r port layouts. Th data coming into th WFM syst

12 m must b mapp d to th corr ct ٠
m must b mapp d to th corr ct � lds for call volum , ach l m nt of handl tim , and oth rs, for ach work typ and p riod. If th ACD is quipp d for skill-bas d routing, or th r is a third-party routing d vic involv d, this mapping can b chall nging. It is vitally important for the WFM system to know what kindof call each is originally.For example, l t‘s imagin a sal s call ov r�ows to a service teammate for handling due to a long queue within Sal s. It is important for th WFM syst m to count th call as a sal s call.If it is r port d as a s rvic call b caus it was handl d by a s rvic ag nt, th n th count for th s rvic typ is arti�cially in�at d. Also, sal s calls ar und r-r port d, r sulting in a for cast for f w r sal s calls than is accurat .The forecasts made with this accurate data will be used to stablish sch dul s. As a r sult. f w r sal s ag nts will b sch dul d. This will forc mor calls to ov r�ow, making this a vicious circl .For most c nt rs, th b st solution to this probl m

13 is to stablish multipl skills. Sal s
is to stablish multipl skills. Sal s calls might qu u up to t ams with th Sal s skill and Sal s backup skill at th look like this:Sal s skill – for ag nts who ar primary on sales callsSal s backup skill – for backup ag nts who are secondary on sales callsS rvic skill – for ag nts who ar primary on service callsS rvic backup skill – for backup agents who are secondary on service calls The Must-Have Guide to Accurate Call Center Forecasting | 13sam tim . Anoth r sc nario is that th r may b a d lay or thr shold that must b xc d d b for th backup t am is tapp d. Th r sults is that th calls handl d and abandon d in th Sal s and Sal s backup skills qu u s are counted as a total for the sales skill as well as the two-s rvic skill.The call center can see how many calls each group of ag nts is handling to h lp out in a s condary skill .Th y can also see what the variation is in handling time b tw n primary and s condary ag nt t ams. S tting up the system with a primary and backup directory number for ach skill may not b attractiv to th I

14 T/T l com t am, but it is ss ntial to t
T/T l com t am, but it is ss ntial to the accuracy of the data for Ongoing ManagementOnc th impl m ntation is compl t d corr ctly, it might s m lik all th hard work is don . Unfortunat ly, that is just not th cas . This is wh r som us rs b com dissatis� d with syst m p rformanc ov r tim . Wh n th for casts ar inaccurat , it is common for th caus to b poorly manag d data from th ACDs, rath r than som probl m in th for casting algorithm of th WFM syst m.The way that many WFM systems use historical data is to cr at av rag s ov r a p riod of tim , smoothing out th minor variations. Th p rc ntag of a w k’s call volum that has or is likely to happen on Monday versus Tuesday is stor d in th WFM syst m. This allows th us r of th WFM system to supply any amount of call volume for the upcoming w k. As a us r, you can choos th appropriat w ks for th av raging. Th syst m will th n know how it will distribut among th days.In addition, th WFM syst m stor s a p rc ntag of an ntir day’s load that happ ns during ach half- or

15 quart r-hour. It us s this distribution
quart r-hour. It us s this distribution patt rn in a similar way. If bad data corrupts thos p rc ntag s, th r sults will b incr asing l v ls of inaccuracy. Similarly, th systems store the AHT average over time in each period so that this can b us d as w ll.Th r ar hardwar /so�war failur s in th data link that caus th data r port d to b lost or corrupt d. For The Must-Have Guide to Accurate Call Center Forecasting | 14 xampl , th WFM syst m might hav nothing but z ros for thr half-hour p riods du to such a failur . If th corr ct data is still stor d in th ACD r porting syst m, it might b possibl to r construct th missing data, or it may b totally lost.Th WFM analyst must r cogniz that th z ros ar th r . Th n, th analyst may adjust th numb rs to stimat th n d d information bas d on similar patt rns. Anoth r option: the analyst may choose to discard the entire day of data as worthl ss and misl ading.For example, y st rday might hav b n an unusual day with a major mark ting campaign or an IVR failur s nding massiv numb

16 rs of calls to th c nt r. Th r might
rs of calls to th c nt r. Th r might have been a large class of new hires on the phones driving up AHT for a coupl of hours. Th comput rs might hav b n slow. All of th s situations r pr s nt abnormal events that should not be averaged into the normal on s for a n w for casting patt rn. Th analyst must review the data and identify if there is anything out of th ordinary and d t rmin th caus . If th caus is on that is lik ly to r p at its lf, th n th data may b sav d in a �l for futur us as a good pr dictor of that kind of situation (such as a mark ting campaign).If th caus is a on -tim v nt, th n th data can b adjust d (normaliz d) prior to storing it, or it may b b st to discard it. But allowing th ab rrant data to b stor d and averaged into the forecast will destroy forecasting accuracy ov r tim . Another issue to be aware of is any change that might tak plac in ith r th ACD or th WFM syst m. For xampl , if an upgrad is impl m nt d for ith r syst m, the data mapping may need to be redone since report formats t nd to

17 chang . In addition, if any n w call ty
chang . In addition, if any n w call typ s or routing patt rns ar impl m nt d, th WFM syst m will n d to r c iv th data for these as well and map them to appropriate agent In fact, the WFM analyst should review the data that is being transmitted on a daily basis to ensure that it is reasonable and valid before it is stored in the WFM system as a forecasting statistic. The Must-Have Guide to Accurate Call Center Forecasting | 15groups in th sch duling. It is all too common for chang s to b mad in th ACD but n v r link d prop rly to th WFM syst m. This r sults in signi�cant disconn cts in th proc ss s, sch dul s, and s rvic l v l r sults.Changes in hours of operation will also require an adjustm nt to th patt rns. Most WFM syst ms allow th WFM analyst to compl t ly r plac an xisting distribution patt rn if n d d, and this is on of th tim s wh n that may b th b st approach. It is also appropriate when queues are consolidated or there is a need to cr at n w work typ s.Such chang s ss ntially r nd r th historical data us l ss and th

18 e WFM analyst may have to enter a reason
e WFM analyst may have to enter a reasonable estimate of the new percentages for the initial distributions, compl t ly ov rriding prior history. Ov r th w ks that follow, th syst m will gath r th n w data and adjust th p rc ntag s bas d on that data.SummaryTh idiom “garbag in, garbag out” is p rhaps ov r-us d, but it is obvious that if th fram work of the data is not correct then the resulting analysis will b fundam ntally �aw d. This has cons qu nc s for th r lationships among th stak hold rs, th v ndor and ultimat ly, th custom r until th inaccuraci s ar r cti� d. Th basic r quir m nt is to mak sur that this do s not happ n in th �rst instanc . You will see the WFM vendors will be happy to work with your team to better understand the WFM interface setup in your center and to maintain accuracy of th data.Give them a call. You’ll both be happier in the end, and so will your agents and customers. The Must-Have Guide to Accurate Call Center Forecasting | 16 Are you missing any critical factors?by Charl

19 es Watson The Must-Have Guide to Accurat
es Watson The Must-Have Guide to Accurate Call Center Forecasting | 17What do you think about when you forecast? For most p opl , for casting is about th data. Mor sp ci�cally, it‘s about taking historical data and applying algorithms. Th n, it‘s publishing a forward-looking s t of proj ctions bas d on that. Many for cast rs sp nd 90% or more of their time on the “historical” part of for casting. How v r, what happ n d in th past is only part of the equation when looking forward. This is just th start. The Must-Have Guide to Accurate Call Center Forecasting | 18Business intelligence is a critical component to getting the forecast right. What is business intelligence? It’s taking in th int rnal and xt rnal op rating in�u nc s that can driv volum or productivity. This part of the process requires close relationships and regular touchpoints with s v ral d partm nts, including cli nt manag m nt, op rations, mark ting, �nanc , sal s and IT. O� n, Workforc Manag m nt (WFM) has to tak a v ry activ rol

20 in ngaging with th s d partm nts. Yo
in ngaging with th s d partm nts. You can’t wait for th m to com to you b caus o� n th y don’t know or und rstand th impact. WFM do s.Changes to ProductivityG n rally, productivity is r pr s nt d by av rag handl tim (AHT) in a contact c nt r. Th long r a call tak s, th f w r calls an ag nt can handl in a day. Wh n av rag handl tim incr as s, productivity d cr as s and th h adcount r quir d go s up. This conc pt is pr tty straightforward to WFM prof ssionals. Internal - Operations ProcessesProcess changesPerformance management changesEmployee TenureEmployee Turnover Internal - Company ChangesProduct changesPricing changesPolicy changes Ext rnal - In�u nc sRegulationsCustomer demographicsCustomer preferencesEconomy The Must-Have Guide to Accurate Call Center Forecasting | 19So, how can we get an even more accurate forecast for productivity?First, talk to th op rations l ad rship t am. Is th r anything inherently that will make handle time higher or low r than what it’s s n in th past? Som xampl s h r may b , a poli

21 cy chang that r quir s th ag nt to r a
cy chang that r quir s th ag nt to r ad a long r script or v rify a custom r’s information. Quit o� n th s chang s ar mad ind p nd ntly of WFM, b caus th y’r s n as “small chang s” to AHT and focus d on l gal or custom r s rvic r quir m nts. It may b d t rmin d that th incr as in AHT is o�s t by som oth r factor. Unl ss this is t st d through WFM though, it can m ss up th for cast.Let’s look at the impact of a smallchange to AHTIn this xampl , a call c nt r tak s on av rag 95K calls p r month with a 5-minut handl tim and an 80/30 s rvic -l v l goal. Th di� r nc b tw n AHT b ing up or down 5% to the forecast can result in a 6 full-time quival nt (FTE) swing in r quir m nts. To put that into cont xt, onc you load shrinkag on, your 6 FTE b com s 10 h ads. Th impact to th �nancials is v n gr at r b caus you’ll hav additional salari s and b n �ts costs for that additional labor. S this xampl on th n xt pag . Th Must-Hav Guid to Accurat Call C nt r For casting |

22 20 FTE Requirements FebAugS ptOctNovFTE
20 FTE Requirements FebAugS ptOctNovFTE R quir d (+15 s c AHT)FTE R quir d (p r th for cast)FTE R quir d (-15 s c AHT) The Must-Have Guide to Accurate Call Center Forecasting | 21 Here are some examples toinclude in that checklist:Changes to policies or procedures that impact the call lengthChanges to contact routing( . g. IVR or m nu chang s)Any chang s to th products or s rvic s b ing o� r dAny marketing or billing materials sent to customers or prospective customersChanges in the customer tenure (a high r % of n w r custom rs can m an long r calls)Ext rnal factors in th conomy or industryS rvic disruptions that can g n rat long complaint callsExp ct d long qu u tim s (if you know you’r going to b und rsta� d, xp ct AHT to go up)So what are some of the areas to look at forvolume impacts?The following drivers will changeby industry and contact typ . Y t they are common in many contactc nt rs and can hav a signi�cantimpact on the actual contactvolum you r c iv .The takeaway here is that WFM has to be actively engaged

23 with the operations team to really unde
with the operations team to really understand how AHT will b di� r nt in th futur than th past. It’s worthwhil cr ating a ch cklist, so you don’t hav to r m mb r v ry impact factor. The Must-Have Guide to Accurate Call Center Forecasting | 22How should you divide your forecasting time up between data analysis and business intelligence?In my xp ri nc , b st-practic is to sp nd at l ast 1/3 of your tim on th busin ss int llig nc sid . That m ans if you took a 40-hour work w k to build a for cast, you should spend about 13 of those hours on discussions with busin ss partn rs. Th s discussions ar to und rstand how chang s in th busin ss will impact th for cast. You can also sp nd this tim �guring out how b st to captur thos chang s into your for cast. Of cours , this rul -of-thumb will vary. As you matur th r lationships and r gular touch points, you’ll b abl to get the business intelligence factored in with less time inv stm nt. Additionally, you may hav a busin ss wh r th data analysis is buttoned up or you may

24 have a lot of automation in th r gr ss
have a lot of automation in th r gr ssion analysis. If that‘s th cas , you’ll hav mor tim to r inv st into th quality of th busin ss int llig nc . Customer utilization (E.g. calls/transaction)B2B (Business to Business) impactB2C (Business to Customer) consideration Call obviation (E.g. calls/transaction)Migration to web-/mobile, self servicemmprovement in �rst call resolution Call drivers(E.g. calls/transaction)Process changesBilling/acounting errors or changesPrice changes The Must-Have Guide to Accurate Call Center Forecasting | 23 Time Spent Building a Forecast Business IntelligenceData AnalysisTh �rst st p is to build in tim for busin ss int llig nc and mak sur you hav a ch cklist of topics to inv stigat . With a checklist, you don’t l av it to chanc that you’ll g t th information you n d. You’ll not only g t mor accurat for casts, but you’ll hav strong r r lationships with critical busin ss partn rs. The Must-Have Guide to Accurate Call Center Forecasting | 24 The importance of measuringvariability by M

25 aggi Kl nk The Must-Have Guide to Accu
aggi Kl nk The Must-Have Guide to Accurate Call Center Forecasting | 25Wh n v r th conc pt of standard d viation is m ntion d, I s a lot of y s rolling. But th good n ws is that this us ful calculation is r ally asy, sp cially wh n don in a spr adsh t program.What is it and why would you want to use it?By d �nition, it is a m asur that is us d to quantify th amount of variation or disp rsion in a s t of data valu s. R ally? What do s that m an? If you hav a bunch of numb rs that you want to analyz , it is on of th tools that will help you to understand how widely the data The Must-Have Guide to Accurate Call Center Forecasting | 26points spr ad from th top to th bottom. It also t lls you how tightly th y clust r around th av rag . Th distribution can be visualized on a graph that has the appearanc of a b ll. It looks a lot lik th distribution that t ach rs us in school to giv most of th stud nts a C, a small r numb r a B or D and v ry f w an A or F.The calculation provides a single number that id nti� s th distribu

26 tion patt rn in th following manner: Ap
tion patt rn in th following manner: Approximat ly 68% of all of th data points will b within 1 standard d viation ith r abov or b low th m an (or av rag ). Th s would b th stud nts who would arn a C grad – including C+ and C-. Approximat ly 95% of all th data points will b within 2 standard d viations from th m an. Th 13.6% shown in the sections above would be the students receiving a B or D. N arly all th data points will b within 3 standard d viations from th m an. This is r pr s nt d in th graph by th 2.1% on ith r nd that would b th stud nts who arn an A or F.Standard deviation using the example of duration:In a contact c nt r, you would want to us this typ of analysis to study a singl population at a tim . For xampl , you could study on call typ , on lin of busin ss, just n w hir ag nts v rsus all combin d. L t’s say you hav th AHT statistics for v ry ag nt in your c nt r that tak s a sp ci�c typ of call. You can calculat th av rag handling tim (AHT) and that t lls you part of what you want to know, sp

27 cially for workforc planning calculati
cially for workforc planning calculations. But it is also us ful to know how wid th spr ad is. You could just look for th low st numb r and th high st. But that won’t h lp you to d t rmin wh th r the distribution of the AHTs is strongly clustered around the average or more evenly spread out from lowest to high st. That is wh r standard d viation com s into play. It is singl numb r that provid s a cl ar pictur of that distribution. The Must-Have Guide to Accurate Call Center Forecasting | 27Assum your AHT data provid s an av rag of 359 s conds p r call for all of th ag nts. Th low st numb r in th group is 213 s conds whil th high st is 590 s conds. You s alr ady that th spr ad is pr tty wid . But what w don’t know y t is wh th r th s high/low numb rs ar r al outli rs or wh th r a lot of your ag nts hav AHTs that ar v ry high or low.Calculating the standard deviation for this data s t r sults in a numb r of 38. That m ans that 68% of our agents have AHTs that are only 38 seconds high r or low r than th av rag . In this cas , that is 32

28 1 to 397. If w add anoth r 38 s conds t
1 to 397. If w add anoth r 38 s conds to th s r sults to pick up 95% of th data points, w �nd that v ry f w of our ag nts hav AHTs short r So th 213 low and 590 high ar r al outli rs and v ry f w ag nts hav r sults lik this. The Must-Have Guide to Accurate Call Center Forecasting | 28For AHT in th 5% on ith r nd of th scal , furth r inv stigation is n d d. Ag nts may hav a t chnical probl m. Anoth r mor lik ly sc nario is that th y may need 1:1 coaching to help them handle calls in the proper way. But l t’s look at anoth r s t of AHTs with an av rag of 359 s conds, with a low of 213 and a high of 590. This tim wh n w calculat th standard d viation, it com s out to 70. Now, you add 70 and subtract 70 from th av rag to �nd th bulk of our ag nts rang from 289 to 429 (rath r than 321 – 397). By xpanding with anoth r 70 s conds, w ar now at a rang of 219 to 499. This s cond band includ s 27% of our sta�, and that wid rang sugg sts a mor wid spr ad probl m. Som overall training program in the center

29 might be needed to encourage adoption of
might be needed to encourage adoption of the standard approach to this type of call. This will lik ly hav th � ct of low ring th AHT overall result! The Must-Have Guide to Accurate Call Center Forecasting | 29 Of cours , th r ar many r asons why th s AHTs vary so much and further analysis may help to narrow down th pot ntial caus s. P rhaps, looking only at th AHTs on individual days of the week or certain times of day will r v al som patt rns to xplor . It is common to s high r AHTs on v nings and w k nds, for xampl . This could b caus d by a high r p rc ntag of n w hir s in unattractiv shi�s, f w r sup rvisors during th s p riods, f w r r sourc s to handl compl x r f rral calls. Or it could just b that th callers save their most time-consuming calls for when th y ar o� work. It has v n b n r port d in som centers that calls are shorter during the day because ag nts don’t want th boss to h ar th m on p rsonal calls.Now, that you s how this kind of calculation might b us ful, how is it don ? Th as

30 i st way is with a spreadsheet tool.Ente
i st way is with a spreadsheet tool.Enter all of the numbers to be analyzed in a column or row in th spr adsh t.Using th AVERAGE function, you can calculat th av rag or m an of this data. (This is not r quir d to do standard deviation but will be helpful when under standing th r sults.)Choos any op n c ll and s l ct th ‘standard d viation’ STDEV function.Wh n you s l ct this function, th syst m will ask you to highlight the group of numbers to be used in th calculation. You want to highlight all of th numb rs in your analysis and click ENTER or OK.Th standard d viation will app ar in th op n c ll. Note: Th r is also a calculation STDEVP and this is rarely used in call center calculations as it requires that rath r than a sampl of data, you hav th ntir population of data. Th r sults ar slightly di� r nt from STDEV. If you r ally want to know how to do this manually with a calculator, you can s arch Exc l h lp, Wikip dia or oth r sourc s for th d tails. I won‘t bor you with th m in this articl !Standard d viation is a us ful

31 tool to apply to th pl thora of data th
tool to apply to th pl thora of data that you hav in call c nt rs. Av rag s alon n v r t ll th whol story. It is quit h lpful in analyzing for casting accuracy, sch dul �ici ncy and intraday � ctiv n ss. Th s ar th standard m asur s of workforc manag m nt t amp rformanc . For th s m trics, you will want to do your standard d viation calculation on th p rc ntag of varianc in your statistics rath r than th sp ci�c count of di� r nc s (du to th changing workload).Wh n us d lik this, it r ally do sn’t matt r if it is a busy or slow tim , th d viation patt rn will m rg r gardl ss. Using this singl numb r, it is asi r to track progr ss in managing th s p rformanc statistics.Giv it a try. You will lik ly �nd it is asy to do and a gr at h lp in managing your operation. Plus, you will sound fabulous in meetings! Talk to a WFM expert Thank you for reading!If you lik d this E-book pl as shar it:To l arn mor about how injixo can g t you start d with a highly- � ctiv cloud-bas