FRIDAY, MAY 25, 2018
Font Size
Featured Drive Improvements in Outbound Telesales with Lean Six Sigma: Part 2 of 2

Drive Improvements in Outbound Telesales with Lean Six Sigma: Part 2 of 2

Outbound telesales is a process full of waste, poor design and variation that is ripe for the application of improvement tools. Not only can process quality be improved, but also revenue can be increased. Part One of this two-part article addresses the types of waste that may exist in an outbound telesales operation and how to begin resolving those wastes. Part Two of the article looks at agent-assisted automation and paths to higher conversion rates.

The way to address non-value added (NVA) activities is to 1) eliminate them, 2) reduce the steps/time it takes to complete them or 3) find a way to get them done less expensively.

Use Agent-assisted Automation

The key technology for accomplishing all of these things is agent-assisted automation. As implied by the name, this is a technological solution that automates the work of an agent in terms of 1) specific conversation tools and 2) computer assistance.

In telesales, agent-assisted automation can address two huge sources of extra motion waste: wrong numbers and answering machines. Pre-recorded audio and pre-programmed system actions allow agents to politely end calls, update their system and signal to the dialer they are ready for their next call – all with a couple of key strokes. Agent-assisted automation can do the same with an answering machine – leave a message (instead of simply hanging up), update the internal system and case notes to plan the next call to that prospect, and signal the dialer the agent is ready for the next call, again all in parallel. Shaving 15 to 30+ seconds on every call adds up to more time spent on value-added activities, which is an immediate productivity lift.

The next two big areas of muda (or waste) are process defects and waiting. Telesales calls are fairly scripted, meaning they do not branch in dozens of different directions. There is some kind of greeting, perhaps some kind of verification, a pitch of some kind, a few ways to handle “no’s,” a close to the pitch and a way to end the call. That whole process, including updating systems and case notes (but with the exception of closing an interested qualified customer) can be completely executed with pre-recorded audio.

With the agents executing outbound calls with pre-recorded audio that has been provided, the process defects are almost completely eliminated. There is some training and coaching needed at the beginning to make sure the agents make the right choices, but once the correct choice is made, the process is defect free. Also eliminated here is the bulk of the between- and within-agent variation. (Madrigal, 2013)

These steps straight out of a Process Improvement 101 class (eliminating and streamlining non value-added work and driving out agent variation) would already make an outbound group 10 percent to 15 percent more productive.

And we have barely scratched the surface.

Extending Productivity Gains

As with the reduction of inventory in a factory, the elimination and streamlining of the NVA has drained the swamp and by so doing revealed that there are primarily two processes involved with telesales work. There is a lot of pounding through numbers and administrative work and a basic pitch to identify qualified and interested buyers, and there is “the closing.” The former requires virtually no real skill, while the latter requires the agent to be enthusiastic, well spoken and accent-free. This suggests that there is a natural and potentially productive division of labor; in fact, there is.

There is a British expression, “horses for courses,” which means that different people are suited for different things and it is important to get people doing work that is aligned with their skill sets. By splitting the agent role into a Tier I admin group that is pounding through phone numbers to get prospects on the line and a Tier II group of closers, there are now two different hiring profiles.

The first group, using the pre-recorded audio provided by the agent-assisted automation, do all the dialing, waiting, leaving messages, dealing with wrong numbers, etc. When a prospect expresses some interest, the Tier I agent transfers the customer to a Tier II agent with a few button presses. Tier I agents almost never even speak with a customer. Some agents even turn their microphones off.

The Tier II team is the closer group. This is a job with a higher skillset requirement. The agents need to be enthusiastic, have the ability to connect with people, be well spoken, know how to deal with rejection, etc. They are harder to find and must be paid greater compensation. The good news is that a company does not need as many of them as they are only taking the qualified calls being handed off by Tier I agents.

The news gets better. Though the Tier I work has been streamlined, there is still a lot of waiting time (dialer delays, phones ringing, etc.). As a result, most Tier I agents find they can easily handle two calls at once and some can handle up to three! As the calls are in different phases, the basic interchange with the customer is often very straightforward. Even though it is a cacophony of voices in their headsets, the agents do not seem to have any trouble handling it. In fact, if a company is deploying a kind of piece-rate compensation system, the agents are motivated and grateful to be able to handle multiple calls.

The productivity gains from this Tier I to Tier II change are incredibly multifaceted.

  1. One agent handling three conversations is a big productivity gain (Figure 1).
  2. Tier I agents are easy to source. It is a low-cost, entry-level job; since the agents’ microphones are often muted, it can be offshored for additional labor savings.
  3. Training is also reduced since most of what needs to be done is baked directly into the automation.
  4. The Tier I monitoring costs and off-phone coaching time are dramatically reduced (or can be redirected to Tier II) because the agents are not speaking, the process does not vary that much and the automation is always correct.

Figure 1: Revenue and Cost Improvements Using Agent-assisted Automation

Figure 1: Revenue and Cost Improvements Using Agent-assisted Automation

Finally, though it is too soon to have collected data on this, the redesign of the process and use of automation shows signs of addressing the muri (unreasonable demands) associated with telesales work. It thus has the potential to reduce turnover and all the costs and corrosive effects on performance from that turnover. This requires a bit more explanation.

Outbound telesales agents love agent-assisted automation. The Tier I agents using pre-recorded audio for qualifying the prospect do not feel bad when the hang-ups and curses inevitably come. Agents report that it does not feel personal; the customer is not saying “no” to the agent, the customer is saying “no” to the software. Further, the agents do not get yelled at because of their accents, they do not get as fatigued because they do not have to talk at all and they make more money because of the piece-rate incentives. Agent satisfaction measures soar.

Also, higher-priced Tier II closers are experiencing fewer “no’s” or customer belligerence because the agents are only being handed calls in which the potential customer has shown some interest. NVA activities have been virtually eliminated. Tier II closers are selling on every call and their conversion rates go up (and their pay goes up, too, if compensation is variable). Job satisfaction increases in this group as well.

It is unlikely that anyone will ever make a career out of making outbound telesales, but even small extensions in agent tenure can improve center-wide performance metrics and reduce the human resources costs associated with turnover.

The Path to Higher Conversion Rates: Tier I

There are two conversion rates of interest: the conversion of someone who answers the phone and is led to be an interested prospect (the Tier I agent pitch and counters to the customers’ initial hesitance), and the conversion of a prospect to a customer (the Tier II pitch and counters). Analyzing and managing existing variation and doing controlled experiments are the key Six Sigma tools for improving conversion rates.

Generally speaking, the first step for improving either conversion rate begins with graphing the agents on a p-chart (see Figure 2 below). At this point, the goal is to identify agents that are statistically better at successful pitches and counter practices.

For the Tier 1 agents who are using a predefined process and pre-recorded audio, there are fewer statistical differences among the agents. There are agents that are converting less due to burning through too many numbers in an attempt to raise their piece rate, but this is rare.

Limited-to-no between-agent Tier I differences (due to the automation) is the bad news. The good news is the fact that the Tier 1 pitch is completely automated – experimenting with different pitches is a snap. It is possible to run as many different pitch variations 1) as there are hypotheses about variables that affect the success of the pitch and 2) as there are agents to run sound experimental designs. Not only is the pitch something to experiment with, but the recorded voice is a variable that can also be tested. Some voices convert better than others. One company found that a women’s voice with a slight Hispanic accent converted better than all the others; they went with that for all agents.

Moreover, each agent running multiple machines enables an improvement team to run better experimental designs with more power. Rather than run a randomized block design with half the agents running one pitch and half running the other, a completely randomized design can be run with agents running the two pitches on their machines. It is a better design with an extra degree of freedom (no blocking factor), enabling the detection of smaller differences.

The Path to Higher Conversion Rates: Tier II

As for improving the Tier II conversion rate, since these agents are not using automation, graphing them on a p-chart will likely yield agents that are statistically better. Although as the Tier II group is smaller, there may not be enough agents to show such a difference. Why is Sally converting at a higher rate than Jane? Is it Sally’s voice? Her empathy? Is there something in how she pitches and rebuts the initial “no” or is it her tenacity? Be sure to use statistics to identify real differences and not just eyeball reasons why Sally outperforms Jane. Treating noise as a signal is a surefire way to develop superstitious (unproven) behavior and take “improvement” actions that actually make things worse.

The p-charts of agent performance will also enable company managers to get better at hiring and training as well as performance management. What is it about Sally’s approach? Can that attribute be hired for? Can other agents be trained and coached to follow Sally’s approach? Jane’s data looks worse, but is her performance statistically different than the rest? It is desirable to manage out those whose conversion rates are statistically worse, as shown in Figure 2.

Figure 2: P-chart of Sales Conversions

Figure 2: P-chart of Sales Conversions


Becoming better at 1) identifying the best attributes to achieve the desire performance and selecting new hires based on those differentiating attributes, 2) training and cross-pollinating best practices, and 3) managing out those whose performance is statistically worse are great approaches for continuously lifting center-wide conversion rates.

Companies know how to cut costs, but all companies continue to look for ways to drive revenue growth. Pick Six Sigma projects targeted at increasing revenue to garner management’s undivided attention.


  • Adsit, D. & Bobrow, W. (2007) “Take the Guesswork and Gamble Out of Hiring Call Center Employees.” Call Center Magazine.
  • Deming, W. E. (1982) Out of the Crisis. The MIT Press.
  • Madrigal, A. (2013). “Almost Human: The Surreal Cyborg Future of Telemarketing.” The Atlantic.

Part One of the article addresses the types of waste that may exist in an outbound telesales operation and how to begin resolving those wastes.

Register Now

  • Stop this in-your-face notice
  • Reserve your username
  • Follow people you like, learn from
  • Extend your profile
  • Gain reputation for your contributions
  • No annoying captchas across site
And much more! C'mon, register now.

Leave a Comment


Chris Seider

1. You used a p-chart inappropriately. P-charts like all control charts are a tool meant to see if a process is changing over time–not categories. A more appropriate analysis tool is a chi-squared analysis for looking for differences among conversion of sales by your Tier 2 agents.
2. You mentioned agents need to be accent free. Really? I take that comment as either racist or you’ve never moved outside of the MidWest USA which is known to have the flattest “accent” in the states. Most everyone has some type of accent and I found the comment inappropriate.
3. Telesales must be a challenge but I wonder if there’s real data showing a higher conversion by folks who hear an automated or highly nuanced automated system when they answer versus a human. Fascinating article and perspective.

Dennis Adsit

Hi Chris,

Thanks for weighing in, I guess.

1. The use of control charts to compare individual people is frowned on per se because Control Charts are supposed to be used to compare long-term variation (between subgroups) to control limits based on short-term variation (i.e., within subgroups.) For continuous data this would mean that the control limits would be dependent on the order in which you arranged the individuals. For p-charts, however, the control limits are based entirely on p-bar (binomial distribution,) so the order doesn’t matter. In other words, there is nothing wrong with using p-charts as long as you’re analyzing proportions or counts. Moreover, one powerful advantage of this approach is how the graph drives home the notion of agents that are statistically different and agents that are part of the system variation…something that just doesn’t come through with a chi-square output that is a page of numbers. By the way, the chi-square flags the same agents as statistically different as the p-chart does. But you probably already knew that.

2. I can see that it might have perhaps been helpful to insert the word “generally” or “relatively” in front of “accent-free.” Thanks for pointing that out. However, and speaking of inappropriate comments, rather than call people you don’t know racist, a curious label anyway just for saying that someone from one culture calling into another has an accent, you might want to open your own outbound call center and have, say, [Scottish] people who were trained to speak [German], call into households in [Germany] to try to sell them auto warranty services, for example (feel free to insert any two countries you would like in the brackets). Then get back to me and tell me that the outbound agents’ accents relative to the prospects’ accents don’t matter. Good luck with that.

3. With respect to your third comment, isn’t that obvious? Companies are dying for revenue and few companies would invest in an approach that delivered a lower conversion rate. Those that decide to accept lower revenue only do so because the lower conversion rate is made up for in shareholder return by creating a better environment for agents (good in and of itself but a huge return in the form of lower attrition costs), and reducing training and monitoring time. Our experience is that the conversion rate is initially the same or just slightly lower using the automation, but it increases over time as the improvement team gets better at running experiments, something that is very tough to do with agents using their live voices.

Chris Seider

Good detailed response.

1. I don’t like to encourage p-charts for the use as you’ve indicated.
2. I didn’t call YOU a racist but suspect we could agree on nuances and word choice better in person over a drink rather than an impersonal blog. No harm was intended but kudos for you to acknowledge the raised concern.
3. Maybe your comment explains why more and more US companies are bringing back call centers from India and the Philippines. However, those in TX or MO do have different “accents” than many in the US so it’s not like this more diverse workplace is as easy to handle as the less diverse countries in Europe (although they are quickly changing with immigration) so we won’t ever be able to eliminate accents for US based customers. It will always be a source of “special cause” I suspect.

Thanks again for the article…the 2nd part was more juicy with details.

Dennis Adsit

Hi Chris,

1. I can appreciate that you don’t like to advocate the use of them. Your call. However, the graph is a powerful “picture is worth 1,000 words” influence tool. It, of course, does not work for very large groups as it becomes unreadable, but you might want to try it to see how it goes over with your management teams and change efforts. You also might want to refer to Deming’s Out of the Crisis, and in particular examples 1, 2, and 3 and Figure 7 in Chapter 3. I’m casting my lot with Uncle Ed.

2. So my *remark* was racist or parochial and, in your view, inappropriate, but *I* am not? Speaking of drinks, here’s a toast to nuance.

3. I think you are exactly right about the movement of call centers serving a particular country back to that country. This is in part due to the fact that the accents and level of cultural understanding are so different, the customer perception is that the agents are less effective at resolving issues.

“A 2008 study by the CFI Group concluded that “when customer service representatives are perceived to speak clearly, they also resolve customer issues 88 percent of the time.” But when they’re not perceived as speaking clearly, “they resolve customer issues only 45 percent of the time.” Although the study concluded that “an in-depth understanding of products and services is as important as language skills,” they were more-or-less inextricable from each other. The survey also found, at least in 2008, that “callers still think they are being served by a contact center located in the United States.” Of the six industries surveyed, half of the customers thought that personal-computer customer-support call centers were based overseas. The PC industry also had the lowest customer satisfaction among the industries surveyed.” From:

Do the people in Texas and Brooklyn and Mankato have distinct accents? Of course they do. Do the distinct accents matter? In part, the answer boils down to a matter of degree…but in general, the less of a perceived difference, the better the agent-customer “connection” and the higher the probability of a sale or successful resolution.

Your comment about not being able to eliminate accents is not true or to avoid a reaction, there is more opportunity to influence it than you can imagine. One of the additional and in fact incredible advantages of using pre-recorded audio is that you can record multiple regional voices and when the outbound dialer calling into (or the ANI (automatic number identification) on an inbound call indicates) a Mississippi area code, you can use a voice with a Deep South accent. Calling into the Boston area? Use a pre-recorded voice with a New England patois. This effectively eliminates accent as a form of special cause variation. That’s the statistics of it. But the more important issue is the removal of one more barrier between a company and its customers, between the customer and agent connecting, between a frustrating experience and a satisfying, effective resolution.

4. In my view, the juicy details of the solution in Part 2 are not as satisfying without soaking in the incredible amounts of waste and inefficiency in call centers outlined in Part 1. Lean has barely touched call centers, But when it does, the savings and improved outcomes for agents, customers and shareholders will drop your teeth. If interested you can read more here:


5S and Lean eBooks

Six Sigma Online Certification: White, Yellow, Green and Black Belt

Six Sigma Statistical and Graphical Analysis with SigmaXL
Six Sigma Online Certification: White, Yellow, Green and Black Belt
Lean and Six Sigma Project Examples
GAGEpack for Quality Assurance
Find the Perfect Six Sigma Job

Login Form