Not So Young Frankenstein

Let's say you were a modern day Dr. Frankenstein and you decided you wanted to bring an organizational creature to life.  Let's further say you were a nefarious Dr. Frankenstein and the organizational creature you wanted to bring to life was one that continuously produced errors...lots of them...day in and day out.  What would go into the design of an error-producing organizational machine?

First, you would have hundreds of detailed processes that the people in the organization would be expected to perform.  Forcing people to execute lots of different process with lots of details increases the probability of errors.

Second, you would not define a standard of what constituted the correct way to do each of those processes.  Some processes would be defined. Others would not.  Some processes would ostensibly be defined, but people in the organization would disagree about what really was the correct disposition.  When "correct" is not clear, errors can bloom.

Third, you would have those processes change frequently.  Big changes and little changes...just constant process changes.  This keeps the workers off-balance and increases the chance of errors.

Fourth, you would have poor change control mechanisms.  Processes would change but the changes would not always be effectively communicated.  There would be no master register of changes and no way to effectively determine if all of the updated processes were being correctly followed.  The lack of effective process performance feedback is great for error making.

Finally, on the process side, you would minimize the amount of  automation available to help workers get the details correct.  If something had to be drilled they would have to do it by hand.  No machine to help employees get the exact size hole in the exact spot.  If there were a lot of details to remember, people would just have to suck it up and remember them.  Further, the job aides you did give them would not work well together.  Having old legacy systems that require a lot of cutting and pasting increases the chances for errors.

Onto the people in this error-producing organization.  A seventh characteristic of a well-designed error making organization would be a low hiring standard.  No college degree required, no employment tests to assess baseline competency.  The lower end of the labor pool is best for higher errors.

After hiring low quality workers, give them poor training.  The training around the processes would be incomplete and the acceptable performance standard for completing the training would be low.  An adequate demonstration of basic competence (70% passing score...which of course is 30% errors) would be enough.  Having poorly trained and mediocre workers makes for really high errors.

Next signal to them that they are low quality by keeping them on a tight leash...make it known that you know when they are a minute late coming back from breaks and lunches...ask them what is wrong if they are in the bathroom too much.  Further send this "you're completely replaceable" message by telling them that there is a large pool of applicants waiting for their job.

It also helps if you can make the work itself super boring and repetitive.  There is nothing that drives errors better than a boring repetitive job as it leads to steps getting skipped.  Oh and make the environment randomly stressful by, say, workers getting yelled at by customers. High fatigue and stress are great for errors.

Finally, don't pay the workers that well.  Low wages and no variable compensation for higher performance.  When everyone gets the same no matter how they do and pay is based on tenure, which is another way of saying how long they last, it really helps people not care which feeds the error machine.

And speaking of tenure...since the jobs are boring and stressful, and the workers don't feel valued, they quit...frequently.  We don't care about them quitting, because as we told them, "there is more where you came from."  We want high turnover because it means lots of inexperienced employees which is like oil for the error-machine.

What do you think of our design?  How could an organization designed like this not be a world class error-machine?

This is not a dystopian vision of some "Breaking Bad" science fiction future.  No, this error-making organization exists today.  In fact there are lots of them and they have been around for years.  It's known as a call center.

I am not trying to be Debbie Downer here.  I am trying to shed light on the fact that the 40-year old paradigm for the design and management of call centers almost guarantees a high error rate.  A call center leader who doesn't believe her/his organization is an error machine is just plain struthious.

It, of course, doesn't have to be this way.  The fixes are right in front of us...just do the opposite of what you would do in the design of the error-machine.

The problem is all of those activities drive up short-term costs and many of today's call center leaders can't see past that.  It is analogous to when American managers visited Japanese plants in the 80's where they couldn't believe that employees had the power and were encouraged to "stop the line" to fix quality problems.  They couldn't see that the long term cost reductions due to the increase in quality far outweighed the short-term cost increase of interrupting the production run.

One fix with immediate ROI is allowing the agents to use automation.  I go into detail in this post Can a Focus on Getting Calls right Have the Far-reaching Benefits Just-in-Time Had? listing all the sources of financial return that come with defining "correct" and using automation to ensure the agents handle each call correctly.

There is a lot of talk these days about customer service and getting the front-line agents to turn customers into zealots.  I think this is the right direction.  But before we ask our agents to love our customers to death, isn't it important that our agents know how to correctly resolve the customers concern?   Can you really fall in love with a car brand that delivers a great service experience if your car is constantly in the shop?

Call Center Coaching Remains Labor-in-Vain

I write all these deliberately provocative pieces and send them out into the ether.  They are mostly (and some would add "rightly"!) ignored. But finally someone disagreed...in writing...to my Futility of Call Center Coaching piece.  I was as happy as the scientists at SETI (Search for Extra Terrestrial Intelligence) finding a patterned response amidst the intergalactic noise!

You can read my original article here:  http://www.isixsigma.com/operations/call-centers/futility-call-center-coaching/ The article has a lot of math in it, but I summarize it in English below.
 
 And you can read the response here:  http://www.elkindgroup.com/coaching/is-coaching-a-waste-of-time/

NICE sells a type of call recording equipment that is the cornerstone for most monitoring and coaching efforts.  So it doesn’t surprise me that they wouldn’t take lying down the pillorying I gave to the go-to solution they enable.

And I love that someone did respond, not because I think their response is correct, but because I love the chance for dialogue about this important topic.  Given that coaching is expensive and given that it is the go-to method most call centers are relying on to improve agent metrics, isn’t it important that we can prove we are getting a return on our investment?

I actually agree with a couple of the points they make, but overall, unfortunately, Andy, Rob and Corey have provided no evidence in their response that proves there is an ROI from coaching in call centers.  Many of their claims are, in fact, completely wrong; others I’m afraid reflect a poor understanding of effective approaches to operations improvement. 

 

Here is a paraphrase of their main point in the first three paragraphs of their response and really their main point overall:  “we have experience and data that shows continuous improvement from coaching, but it is inappropriate to share client information.”  

As an aside, they actually say “a comprehensive program of performance management supported by behavioral coaching.” I am not exactly sure what that means but am guessing it has to include letting go the bottom x% based on performance.  This adds in a completely different dimension which I did not argue against or model in my original article.  I am going to ignore the PM part and keep the focus here on their defense of coaching’s ROI, which is what my paper was about.

Speaking of what my Futility of Call Center Coaching paper was about, let's review that.  First, and most important of all, I never said coaching couldn't help an individual agent improve.  It would be easy to produce example after example about individual agents helped by coaching.  In fact, this country and probably the world have a love affair with the power of a coach to improve performance and change lives.  And you know what?  Rightfully so.

My original paper was just a simulation...a mathematical model of what happens when we "add" a little bit to the performance (I called that add "coaching") of each "employee" each "month" and then some of those "employees" who we just "helped get a little better" "leave" (at various turnover rates typically found in US and overseas call centers) and we replace them with new and by definition lower performing employees.  


There were no real employees.  There was no real coaching.  No one really quit.  It was a simulation.

But what happens in a system with parameters like that?  The answer:  the turnover eats the monthly gains in individual agent performance and center wide performance quickly plateaus, just like it does in the call centers the KomBea management team has run and are currently working with.  If performance in a call center is not continuously improving, then the "coaching" (or whatever is "adding" to the performance of individual agents that are constantly leaving) cannot have an ROI because while there is a lot of "I" there is no "R."

Back to their main critique.  First, I understand the need to protect client information.  While they don’t need to share information on the clients, it sure would be good to see the data itself, because, as I mentioned, the principals in my company, KomBea, have the opposite experience.  We have worked with hundreds of call centers, including running our own, and we have never seen continuously improving anything when it comes to agent performance.  

Now defining terms in any argument is important.  By “continuously improving,” I mean over many months and years, just as you would expect in a manufacturing environment for measures like defects or throughput or changeover cycle time or scrap or WIP.  Go into a world class manufacturing environment and those measures are all continuously improving…over years and even decades.  People lose their jobs in Manufacturing if those measures aren't improving.  This is what we should be solving for in call centers.  If by continuous improvement, the authors mean over a few months or for a “season,” heck that is just as likely learning curve improvements that would have happened without coaching.

Please guys, show us the continuously improving data you have…masking it in any way you need to protect your clients.  And more important, 1) show us that the continuous improvement is over years, not a few months, 2) demonstrate the improvement is due to coaching and not other system wide changes or learning curve improvements that would have happened without coaching, and finally show us 3) that the benefits of that improved performance were worth the investment in terms of recording software licenses, monitoring personnel, and off-phone coaching time (which has to be paid for with extra agents on the phone covering for the agents being coached to maintain service levels) that is the real cost of getting any improvement.  I'll eat my hat if they can.

The authors go on to make three specific points to attempt to refute my claim that coaching has no ROI.  The first part of their first point is that coaching and process improvement are separate activities that both add value. 

Here we are in complete agreement...they are separate activities.  I know process improvement adds positive ROI and I will show an example of that below.  My point however is that I don’t think the way most call centers are coaching agents has positive ROI.

Moreover, though PI and coaching are distinct solutions, real problems can arise when we apply the wrong improvement approach to a particular problem.  Because the go-to method is coaching, center management often fruitlessly throws the coaching solution at problems that would be more effectively solved with process improvement.  When all you have is a hammer, you treat everything like a nail and just whack away.  Whacking away is generally not a great strategy unless you happen to find yourself hiking in a jungle.

Their second paragraph under point number 1 is where I really take issue.  Here is their claim:  “Monitoring is the best way to ensure that front-line agents are adhering to business processes and executing them in the manner that was intended.”  This statement could not be more false and reflects the worn-out mental model that the whole call center improvement industry rests on…that the best way to improve the overall performance of my center is by trying to improve the performance of each individual agent one-at-a-time.  This mental model is wrong.  (See:  The Two Widely Held and Deeply Flawed Mental Models Perpetuating Call Center Mediocrity)

Before I share an example, consider this:  do they run manufacturing operations by having an army of inspectors videotaping each worker and then coaching them one at a time?  In manufacturing they got rid of all the inspectors decades ago and moved to baking quality and process adherence into the process itself through the use of automation, error-proofing, Andon lights (to signal process problems), etc.  If videotaping is not the best way to achieve process adherence and efficiency in a complex operation like manufacturing, why is it the best way to do it in a complex operation like call centers?  That's my point.  It’s not. 

Let’s get into a real life example.  Many calls in the call center industry have a disclosure component…something that the agents must tell the customers every single time they get one of those types of calls.  This is often a key component of call quality metrics.  The authors claim that monitoring is the best way to ensure the process is adhered to and the correct disclosure is provided. I disagree. The best approach is to know the disclosure is done correctly without any checking of any kind being necessary.

To this end, disclosure compliance is best accomplished with agent-assisted automation,  pre-recorded audio and error-proofing.  This approach results in a process where the call cannot advance or be completed until the proper statements are read to the customer by the system and the customer acknowledges having heard and understood them, which is what you want to happen.  When that step is completed, then the agent can process the order and complete the call.  Done in this fashion, that portion of the call is always 100% correct and no monitoring or inspection or coaching is needed.

We are currently working with one of the largest financial institutions in the world.  Disclosures are critical-to-quality in the financial services industry and failure to do so is punishable by stiff fines and seriously bad PR.  In many industries...financial services, healthcare, etc...less than 100% disclosure compliance will soon be unacceptable.

Our client told us that for years and years the best they have ever been able to achieve is 92% compliance and that level of performance was rare.  Most of the time, compliance was in the high 80%s and was most certainly not continuously improving.  This despite the typical huge investment in recording software seat licenses, an army of monitors, and regular off-phone coaching time for agents. 

Are Andy, Rob and Corey really suggesting that our client should spend more on recording software, hire more monitors, and take the agents off the phone for longer to try to get past 92%?  Are Andy et al. arguing that our client just isn’t doing it right and if they let more low performers go, the next replacement crop of agents will be able to perform better?  Are they arguing that all the coaching investment over the last several years which could never produce results greater than 92% one time was worth it? Are they really arguing that there is a positive ROI in there...somewhere...and that if they just keep at it they will eventually find it?

From the first day, from the first hour, in fact from the first call this financial services giant took using our ExactCall solution, the disclosure compliance rate was 100% and it has never dropped below 100%.  Between-agent variation on disclosure compliance has been completely eliminated.  Read that sentence again.  This was achieved without monitoring a single call.  In fact, the use of our technology has cut monitoring costs in half because the monitoring team never listens to the disclosures anymore…a significant part of every call…because they are always exactly correct.

I keep blathering on about ROI so here is how our client calculated the ROI on this process improvement investment:

1) the value of fines avoided (they know what they have paid and what they would pay with out the improvement...going forward, they would be able to see the fines other financial institutions are paying who don’t use the technology), plus 

2) the savings from the reduction in monitors, plus 

3) the savings from the reduction in extra agents needed to meet service levels due to less off phone coaching time, plus 

4) a reduction in training time (since you don’t have to teach the agents how to do the disclosures), plus 

5) a reduction in AHT (pre-recorded audio is generally faster and is not challenged as much by callers), plus 

6) a reduction in turnover because our solution makes the call center agent job significantly more enjoyable and less fatiguing.

Add up all those tangible financial benefits and subtract the cost of our solution.  The payback period is less than six months.  Everything after that is gravy.  I am still waiting for someone to layout the ROI case for one-on-one agent coaching that clearly.

Later in their first point, they go on to quote the late quality guru, W. Edwards Deming.  They say Deming’s 6^th of 14 points was:  “Institute modern methods of training on the job for all, including management, to make better use of every employee.  New skills are required to keep up with the changes in material, methods, product and services, design, machinery, techniques and services.”  They offer this as a defense of coaching.

Before I respond, I just have to say the fact that they quoted Deming delights me.  First, I am a huge Deming fan.  I am such a fan, I recently wrote an article entitled Deming not DiMaggio.  Second, I don’t know if there is a required reading list for call center leaders, but if there is, Deming’s Out of the Crisis is not on it because the way most centers are run leave Dr. Deming spinning in his grave.  So to see that there are some call center experts who are even aware of Deming’s 14 points gives me hope for the future of this industry.

Onto their point:  I never said not to train employees.  They obviously have to know what to do.  But also Deming’s point explicitly says new skills are required to keep up with a host of changes on a variety of fronts. 

Here is the difference between what my idea of training is and what I think their idea is:  I want to train the agents how to use software that ensures the call is completed correctly.  

It sounds to me like Andy et al. would like to train the agents and hope they remember what they are supposed to do and hope they do it correctly at the appropriate time, but then, knowing people aren’t perfect, they would like to buy a bunch of monitoring software, hire a bunch of “inspectors” because that is what monitors are, take the agents off the phone, show them what they are doing wrong, and then put them back on the phone and hope they do it right, and then when they get tired of repeating this process and they have lost hope they will apply their “comprehensive program of performance management” (read as “fire”) to the employees, then they will recruit, hire and train new ones and hope they work out better. 

The last paragraph describes the life of call center management teams today.  It is also very similar to the Myth of Sisyphus...the mortal condemned for annoying some god to rolling a stone up a hill and then watching it roll back down...for all eternity.  Moreover, there is a lot of “hope” in their improvement approach.  It's cliche, but hope is not a strategy.  Nor is the word hope to be found anywhere in books, articles or descriptions of the Toyota Production System, one of the most powerful improvement methodologies ever devised for complex operations.

On the way to looking up Deming’s point #6, which I feel Andy et al. have misinterpreted, I am curious as to why they skipped over Deming’s point #3:  Stop depending on inspections.

• Inspections are costly and unreliable – and they don't improve quality, they merely find a lack of quality.
• Build quality into the process from start to finish.
• Don't just find what you did wrong – eliminate the "wrongs" altogether.
• Use statistical control methods – not physical inspections alone – to prove that the process is working

Andy et al.'s second of three points is that coaching reduces turnover.  Here again, I actually agree with them.  Coaching probably does reduce turnover.  Knowing your manager cares about you and wants you to get better should result in wanting to stay longer.  

I argued in Why Your Turnover Reduction Efforts are Not Working that because call center agent jobs were so repetitive, tiring and stressful that efforts to make the job easier (process improvement) were more likely to reduce turnover than efforts to make the environment nicer (coaching is one example, but there are a lot of things call centers try to do to "spruce" the place up...some are actually quite comical, not as activities in an of themselves, but comical in the sense that someone would actually think they might effect turnover).  


But, that said, let me give them the benefit of the doubt for a second.  In fact, let me extend Andy et al.’s argument:  it is actually quite possible that coaching reduces turnover so much that the ROI from coaching all comes from turnover reduction and the benefits of any performance increases they are able to get from coaching are all upside.  This is an empirically testable question and it should be investigated.

However, for now, my article mentioned that call center turnover in the US averages 36%.  All or certainly almost all those centers have monitoring and coaching programs for the agents, and still the turnover rate is 36%...every seat turns over in less than three years.  And this should not be surprising:  who would want to make a career out of being a call center agent?  On a good day, it is very repetitive and exhausting talking to people all day long while trying to stay positive.  On a bad day, with frustrated customers literally yelling at you and saying things you probably taught your kids not to say, it is extremely stressful.


In popular overseas locations for call centers like the Philippines and India, it is even worse.  Call center turnover offshore is routinely north of 100%.  In fact, it is not uncommon to find turnover approaching 150-200%.  (I am not disparaging the Indian and Philippine call centers...I talked to a guy running a 600 seat call center in Florida recently who had 350% annualized turnover).  Now of course, they monitor and coach their agents offshore as well, but still their centers can completely turnover twice a year.  Are Andy, Rob and Corey arguing that the coaching these domestic and offshore centers are doing is reducing the turnover from the even more stratospheric levels it would be if they didn’t coach?  Hmmm…OK.

The effects of coaching on turnover and the potential ROI of that is a different article.  My article was meant to highlight the corrosive effects of turnover on center-wide agent performance.  Should you pour resources into coaching and trying to improve your center one-agent-at-a-time or into process improvement efforts that lift the performance of all the agents?  


When you pour improvement dollars into a resource that is likely to walk out the door in less than a year (outside the US) or in less than three years (in US-based call centers), isn’t it going to be really tough for that investment to show an ROI?  All that experience, training and coaching (read as “better performance”) quits and you start over with inexperienced people (read as “lower performance”).  When you pour resources into process improvement the improvements remain even when (not if) the agents leave.  More on this investment tradeoff at the end.

At the beginning of their final critique of my article, they argued that long-tenured employees can and should continue to improve.  They didn’t like that I built a model with an assumption for a declining effect of “coaching” for longer-tenured “employees.” 

I won’t argue with them on this.  I explicitly stated the terms that went into my model so that researchers could build their own models with their own assumptions to see under what conditions one-agent-at-a-time improvements in high turnover environments pay off.  To those researchers that want to replicate it:  good luck coming up with model parameters that 1) show an ROI and 2) even remotely resemble call center environments here on Planet Earth.

But I do take issue with their second and fourth paragraphs under their third point.  In their second paragraph they say, “These call centers…must continually raise the bar of performance expectations at every level of their organizations…In an organization that is continually raising the bar, a solid performer who doesn’t continue to improve becomes a poor performer who doesn’t have a job.”


This is tough talk.  Let’s go back to our client, who desperately needed perfect disclosure compliance, yet averaged in the high 80’s.  How would they go about raising the bar of performance expectations here and how would that help?
  

I hope what they have in mind would not look like this obviously ridiculous example:  no one is running a 3 min 30 sec mile, so we are going to raise the bar to 3 mins 15 secs.  And to show how serious we are about our standards and what is acceptable, we are going  to keep the few guys who are under 4 mins and cut from the team any one running the mile above 4 mins and go recruit some new guys who we hope will be better.  Does anyone really think that would have a positive effect on the team's times in the mile?


When I used to lead Six Sigma improvement efforts, we used to say "people are three sigma at best" (low 90s% process adherence)…they can never be six sigma, because, well, they are human.  You can raise the expectations bar all day...you can publicly pillory the poor performers in the parking lot...right next to the high performers' employee-of-the-month preferred parking places...and it won’t make an ounce of difference.

A lack of understanding of “system” performance…what it means and how to improve it…is reflected in their final paragraph, “Our clients often find that it’s by focusing on their third quartile that they can maximize the overall impact of their coaching efforts.” 

What does the third quartile of agent performance represent?  The performance of the agents between the 50^th and 75^th percentile in terms of performance on some measure.  In our example, it might have been agents who were hitting disclosure compliance between 84% and 88% of the time (the average was 88%).  This is well within the normal distribution of performance of this “system” and would actually be one of the worst places to focus your limited process improvement dollars and efforts.


Deming is known for a lot of things, but one of his classic demonstrations was his red bead experiment.  (I really encourage you to follow that link to see Dr. Deming in action!)  He had a large bag of white and red beads, white being “good,” red being a “defect.”  He would have managers reach into the bag and pull out ten beads and then plot their performance against each other.  He would give out “awards and praise” to those who drew the fewest red beads and would “reprimand” those that had too many red beads.  Those getting scolded would howl in protest saying there was nothing they could do…it was “luck of the draw.”


That was Deming’s teachable moment and, if it is possible for an 80-year old to pounce, that is what he would do.  He would say, you can measure your workers all day, but the fact remains that you have a “system” (the hiring, training, management, tools, knowledge bases, compensation, incentives, measures, dashboards, etc) that is producing a performance average and a distribution of performance around that average.  Deming said the best thing to do was to try to raise the performance of the entire system, not focus on the performance of each worker.


The only place it makes sense to focus on individual worker performance was at the extremes:  those statistically better and those statistically worse.  That is why he recommended the use of p-charts and other control charts and statistical techniques to make sure you were distinguishing signals from noise (see Deming’s Point #3 above).  


Those statistically better than the rest of the floor need to be studied.  What are they doing?  How are they able to be so much more effective?  Have they identified a best practice that we can steal and propagate across the floor?  Those statistically worse…not worse by some arbitrary line someone with no understanding of how to improve a system pulls out of his/her butt to say “above this good, below this bad”…need immediate focus and help.  And if they can’t improve they need to be let go because, statistically, the “system,” however good or bad it currently is, should be able to attract and produce someone who performs better.  As for those in the middle, in Deming’s view, leave them alone and focus your limited resources on improving the system.

To conclude, I am glad Andy, Rob and Corey responded.  They made a couple great points and I agree with those points.  However, I don’t think they offered any evidence that coaching has a positive ROI.  I am sure they disagree with this rejoinder and I am actually hoping they will share their data and come out swingin' against my position here.  

But the real reason I am glad they responded is because this dialogue is absolutely essential to the improvement of the call center industry.  This industry is terrible!  Agents don’t like their jobs (that is why the turnover is so high!), customers don’t like to have to call, and the costs are still way too high (I argued in this paper Call Centers:  The Deep and Still Largely Untapped Vein of Operational Profits that no matter what companies have done to lower their call center costs...outsourcing, self-service, support communities...that there was an additional 40% profitability upside.  I stand by that claim).  And agent performance is not continuously improving.  If call center leaders were held to the same standard that manufacturing leaders are held to in terms of performance and quality, every call center leader in the world would be fired.  It is bleak.  It is not getting better.  How could more of the same be the way out of this wilderness?

You say, I agree…performance is bad and we have to invest in both process improvement and coaching to get better.  


I say we have to be smarter than that.  Who has all that money to throw around willy-nilly and hope?  We can't let our admiration for the power of teaching and coaching and our love affair with great teachers and coaches blind us to when that solution is cost effective and when it is not.  It is absolutely not the solution for improving agent performance in high turnover environments like call centers.

So if we want to improve agent metrics and the improvement dollars are limited (and they always are), the bulk of our investment should be in process improvement.  We need new tools and solutions to get the caller's problem diagnosed correctly every time.  We need new tools and solutions to get the disclosures right every time.  We need new tools and solutions to correctly handle the warranty claim and to process the Return Materials Authorization every time.  We need new tools and solutions to make the right cross/upsell every time.  


Not some of the time.  Not most of the time.  Every time.  It's 2013.  We have been putting large numbers of people in very large rooms and routing calls to them for 40 years.  How can we still think 85% process adherence is OK?

Agent-assisted automation...with its pre-programmed system actions, pre-recorded audio, and error-proofing...holds the promise of being one of those tools and solutions.  And we know approaches like this have a positive ROI, no matter how high the turnover is.  Only after we feel we have exhausted the process improvement opportunities should we turn to methods with less clear investment cases.  

And if call centers are anything like manufacturing, the process improvement opportunities might never be exhausted.  

The Only Problem You Need to Solve to be a World Class Call Center

Here is the good news: if a call center wants to be recognized as one of the best call centers in the entire world, there is really only one problem it needs to solve: between-agent variation. That’s right. Fix that problem and you will have established a new standard of excellence. Of course there are other operational challenges that need to be solved, but between-agent variation is such a big problem with such huge costs associated with it, problems two through five or ten on the list are almost irrelevant.

Now for the bad news: 1) between-agent variation is not new…it has been an nagging issue for the entire four decade history of the industry, 2) billions have been spent on the problem, yet it really is not improving, and 3) though the problem can be drastically reduced with three simple steps, few centers have the ability and willingness to do what it takes to fix it.

Between agent variation is not a geeky, statistical, issue. It is a real visceral problem we all can relate to. Because of between-agent variation, when each and every one of you calls into a call center today, you have no idea what kind of experience you are going to have. Is my agent going to be empathetic, or am I going to get one of the downright surly ones? Am I going to get an experienced agent who knows how to solve the problem, or a brand new who is not fully prepared and who has to put me on hold a lot to get help? Am I going to get to speak to an agent I can understand or am I going to get someone with a barely intelligible accent? Is my agent going to be focused on listening to my problems or is my agent going to be texting their friends during the call? Is the agent I get going to be conscientious and do all the required steps and read me the required consumer protection information or is my agent going to be bored and burned out and skip steps to get off the call?

If you don’t think this is costing your center real money that your CFO can eat, think again. The costs associated with agent variation are not easy to track but know this: they are real and they are scary big.

Some of the costs are actually easy to track. If I have wide variation in agent performance, I have more costs in the form management resources and off phone coaching time to manage the under performers. There is also the added staff from a workforce planning standpoint to handle the lack of predictability of performance so that the center remains within service levels (so many calls answered within a targeted amount of time). There are also the added costs in the form of recruitment, hiring and training replacement workers.

But the true cost of poor quality (COPQ) is like an iceberg: some of the costs you can see, but the biggest costs are below the waterline. For example, there are huge, somewhat hidden costs associated with not adhering to the defined call handling process. Let’s say an agent forgets or decides to skip over parts of the correct procedure on checking if a product is still under warranty and then incorrectly authorizes a product return. A large technology client of ours had thousands of outsourced agents. It would be an understatement to say that the adherence rate across their outsourcers to the Return Materials Authorization call handling procedure was below the Six Sigma level…it wasn’t even close. The result was a huge hidden factory of costs associated with processing incorrect product returns. My guess is that a third of this company’s costs associated with handling RMAs for this product was due to agents not following the correct procedure during the calls. The costs here are only somewhat hidden here because they fall outside the call center’s budget.

Another category of invisible costs is lost opportunity costs. I taught an Operations class in the Stanford MBA program. There were about 120 MBAs in the class and we had them go buy a prepaid cell phone and call the provider to activate it. I didn’t know what the documented call handling procedure was for this company, but since one of our clients was a Telco, I was fairly certain that the company wanted the agents to tell the new clients how to check minutes, how to add minutes and that they also wanted the agents to cross-sell the new customers additional services. Yet when these MBA students called in to activate their new pre-paid cell phones, the agents did those three things less than 30% of the time. What did this lack of process adherence and between agent variation cost this company in the form of lost revenue?

A final element of these invisible costs lies in the effects of variation on the customer. Everyone reading iSixSigma articles knows the mantra, “Customers don’t experience your average. They experience your variation.” And when that experience is bad, you create detractors who can fail to renew or buy again and who speak out about their experience to friends or on social networking sites and even post calls on YouTube.com. Measuring these kinds of costs has always been difficult, but that does not make them less real. The Stanford MBA students were asked to rate their impression of the brand after this one and only phone interaction. Fifty percent walked away with a negative brand impression and would be considered detractors using Fred Reichheld’s Net-Promoter measurement.

Believe it or not, despite the size and seeming intractability of the between agent variation problem, it is, in fact, very fixable. Three simple steps that we have known about forever, done in concert, would drive a stake in the heart of this problem. Sadly, few centers today do any of them let alone all three of them.

First, you have to define the correct process for each and every call. Every call type has to have the exact process for resolving it explicitly defined and kept up-to-date. Specify what the agents have to say to customers and do in their systems to correctly handle every call. If the correct procedure is not defined and understood, how can the agents be expected to execute it correctly?

And yet routinely, this step is skipped. When we go in to begin working with our clients and to documents call handling procedures, it is common for the subject matter experts to not know the correct procedure and, even worse, to disagree on the correct process.

Of course once the correct procedures are defined, they have to be thoroughly taught to all the agents and the quality monitoring team has to be able to rate every call as to whether the call was handled correctly. Too often, the QM team just rates whether the agent was polite, listened, used the customer’s name, etc, as opposed to finishing an evaluation with a clear rating of correctness of what the agents said to the customers and did in their systems against the defined process.

This step is sine qua non, but it is not enough.

Second, if you do step one, you will have less between-agent variation, but you will still have a lot. When you are ready to begin attacking that variation, W. Ed Deming’s Out of the Crisis is the lighthouse to navigate by. He recommended the use of p-charts and other statistical tools to separate out the agent variation that mattered from the performance of “the system.” Those agents that are statistically worse than their peers…not the arbitrary bottom 25-30% that centers use today to judge who is performing acceptably and who isn’t...but the ones that are statistically worse need to be coached. And it’s up or out with that group.

The agents that are statistically indistinguishable from each other should not be coached or uniquely rewarded or punished. The average and distribution of those agents represents the performance of “the system.” The system consists of the hiring process, the training process, the rewards process, the measurement and feedback process, the workforce management process, the policies, the operating mechanisms, the job aids, the systems, etc. And to improve the performance of this statistically indistinguishable group of agents, the performance of the system needs to be improved, not the performance of each agent, one-at-a-time.

The best place to start improving the performance of the system is to look at the performance of the agents who are statistically better. What are they doing? What process are they following? What best practices have they identified that we can clone or better yet automate, so that all the agents are following that process? This is a key part of Deming’s Scientific Management Approach and practicing it will have a huge impact on reducing agent variation.

The third and final step is to add automation. The call center industry, unfortunately, has a rather fixed, either/or mindset about automation. A call is either automated…that is handled by an IVR (press 1 for account balances, press 2 for billing questions)…or it’s handled by a live agent.

We need to start thinking about automation as a job aid for the agents. Our ProtoCall^sm software is an example. It provides automation for the live agent to leverage where it makes sense. The agent is live on the call listening to the customer and using pre-built paths and pre-recorded voice files as long as it is appropriate. Sometimes the call goes beginning to end without the agent ever having to break in with their live voice. Most customers don’t notice or don’t seem to care because their issue that led them to call in is being resolved.

Usually, however, the call goes off the beaten path a bit or the customer wants more explanation or wants to make small talk. When this happens, the agent simply comes in with his or her live voice and does what they need to do to help the customer. And, before you ask, the customer does not care about hearing two voices. We have had Philippine agents using pre-recorded US voices and the customers never comment. We have handled millions of calls using this system and get few comments, let alone complaints.

Moreover, the agents absolutely love this automation. They know it makes them more accurate and consistent and it also allows them to not have to worry about remembering everything they are supposed to do, which allows them to really listen to the customer. It also means they don’t have to talk as much which greatly reduces their stress and fatigue levels.

Call Centers today face a number of problems that contribute to agent variation. There is a lot of agent turnover. There is a lot of agent movement inside the center. There is a lot to know and a lot that changes on a daily basis. How can the agents possibly be expected to stay on top of everything? But used as a job aid, the software and pre-recorded messages can easily be updated to counteract the effects of turnover and constant changes that increase the probability of errors.

The call center industry has used automation very effectively to eliminate phone calls. There is no doubt about that. But the untapped opportunity is to leverage automation to increases process adherence and reduces unneeded variation. Agent deployed automation is the best of both worlds…the flexibility and sensitivity of a human combined with the accuracy and consistency of a computer.

People who are versed in Six Sigma and read a lot of articles on this site must be scratching their heads at this point. They are saying to themselves, “Let’s see…excessive variation leads to huge hidden costs. You can reduce the variation if you define the correct process, use statistics to separate real agent variation from system variation, and leverage automation…where is the story here?”

From that perspective, there isn’t one. But the back story may lie in the mental models that need to change if we are going to cut the Gordian Knot on agent quality and variation.

First, GMs and CFOs need to be all over call center leaders to measure the true COPQ. Just measuring the costs you can see would be a good start but it would be a mistake to stop there. Here are some questions for the GMs to ask: How much extra staff are we carrying? Where are the rework factories? How much revenue are we leaving on the table? What could the negative word of mouth be costing us? Once GMs and CFOs realize how much this agent variation is costing them, they are going to be as focused as fighter pilots on fixing the problem.

Second, leaders need to start reflecting on the progress that has been made in reducing variation and improving quality in manufacturing operations, further think about the approaches used in manufacturing and then start holding their call centers to the same high quality standards that the leaders of manufacturing are held to. Customers are calling you for a reason. They might be looking for some understanding, but they probably aren’t looking for a friend. They want their issue resolved. Getting every agent to resolve the customers issue the right way every time…not 30% of the time, not even 85% of the time, every time…is the world we have to start living in. Manufacturing lives in that world and call center leaders can learn from them.

For example, you lost your credit card and I lost my credit card and we each call in. Those calls will be somewhat different…a little more explanation on one, a little more small talk on the other…but the bulk of those two calls will be exactly the same. There is a right way and a wrong way to handle this call. You will not get agents to do and say the right thing at the right time every time by trying to coach them one at a time. We have spent billions doing it this way and the problem is still rampant. There is simply too much to learn and it changes too often and there is too much agent turnover to ever be successful with training and coaching as the go-to quality approach.

Our manufacturing colleagues have a storied track-record of improving quality and reducing variation. They didn’t develop this reputation by trying to fix each worker one-at-a-time. Manufacturers define the correct process. They systematically improve the process with automation where it makes sense. And they systematically improve the workers with a scientific management approach.

Call Center leaders need to adopt this manufacturing mindset. It is the only way out of the agent variation wilderness we have been wandering in for the last forty years.