The terrier "Nipper" listening to his master's voice (1898). Speech analytics technology might be new, but the idea isn't.

It also gives the quality manager the chance to liberate himself from the stereotypical quality management process where his quality monitors would listen to one or more calls per agent per week, give them feedback and then wait until the following week to see how well they had acted on their feedback.

Speech Analytics can be used in quality improvement projects to measure and address specific issues and take focused, effective action.

Here’s how a quality manager might address Average Handling Time using Lean Six Sigma methodology and Speech Analytics.

“D” for “Define”

The first step is to define the project. The problem to be addressed needs to be defined concisely but also specifically. It may be useful to calculate the Cost of Poor Quality (COPQ) and so define the financial impact of excessive Average Handling Time. This is when decisions are made about the scope of the project, which include which operational processes will be covered, where the process in questions begins and ends and who needs to be on the project team.

“M” for “Measure”

The “Measure” phase is likely to take longer than the “Define” phase, because the first part involves establishing what needs to be measured and how, while the second involves data collection itself. A good place to start is by making a detailed map of the process. In contact centers, it is very tempting to rely on the process maps the IT department will have of the call flow. It is a good idea, however, to involve agents to get their perception of the process and ensure that any unofficial workarounds are also documented.

It is also important to conduct a Voice of the Customer exercise to get the views of the end customer of the process. When companies decide to speed up processes at the expense of customer satisfaction, the result is often a loss of brand value and customer loyalty. Customer opinions should then be transformed into the expression of customer needs and further quantified as Critical To Quality (CTQ) indicators which can be measured.

Once this has been done, the next stage is root cause analysis, where the most likely root causes related to each CTQ are selected to be validated by the collection of data.

Data collection is the point where Quality Management and Speech Analytics applications can be used to measure these root causes in terms of whether the agent is following the scripted elements of the call, whether s/he is giving the customer the correct information and meeting the customer’s requirements.

“A” for “Analyze”

Once the results are in, they need to be analyzed to validate the root causes. Analytical techniques will include control charts to establish the process’s stability and capability.

In this example, if call lengths prove to be widely variable and include data points which are more than 3 standard deviations either side of the average, then the process will be considered to be unstable. This makes it very difficult to predict what is going to happen next. The process becomes very difficult to manage and future outputs difficult to estimate.

Where there are customer mandated specifications for handling time, for example, it is not possible to complete the call flow in less than 4 minutes, then any calls which are shorter or longer than these limits are considered to be “defects”. Process capability is a measure of how many defective outputs a process produces. It also makes sense to analyze the correlation between the results for each root cause and the CTQ in question using either scatter plots or, if necessary, correlation and regression calculations.

“I” for “Improve”

The results of the analysis will be used as the input for the “Improve” stage. Possible solutions to improve the CTQ indicators will be brainstormed and the results whittled down using various prioritization techniques to establish practical, feasible and effective solutions.

To minimize the risk, it is a good idea at this point to run a pilot project, perhaps using a small team of agents to compare their results against the larger “control” group. Assuming the results of the pilot project are as expected, then full implementation can take place.

“C” for “Control”

Once the new solution is running, the same data collection plan that was developed in the “measure” phase will be used here to check if the solution is working as expected.

The pilot project may also have highlighted the need for further measurements to properly manage the new process. Once again, this is where Quality Management and Speech Analytics applications can be used to monitor the process and ensure that the agents are progressing through the workflow in the most efficient and effective manner.

The use of Speech Analytics as a measurement tool enables the user to take much larger and more significant samples of data. This means that root causes can be verified quickly and accurately and then addressed with laser like precision.

Quality management projects in contact centers have not always been especially effective in the past. One of the main reasons for this is that it is not possible for human quality monitors to listen to every call. Speech analytics technology can get us closer to the ideal of an automated quality monitoring process.

Most speech analytics systems work by identifying key phrases which are input into the application. They can either be phrases that the agent must say or must not say, or they can be phrases that we would like to hear the customer’s say. These include “buying signals” during a sales conversation, or phrases that we hope the customer’s won’t say, which indicate that the conversation is not proceeding as planned.

Most calls have prescribed opening and closing phrases. These ensure that the company is identified to the customer clearly and consistently. At the end of the call, set phrases also help to ensure that the customer is left with a polite last impression of the company in his mind.

The Script Adherence

Script adherence is related to Average Handling Time (AHT). When agents go “off script” for large parts of the call, their AHT usually increases. Speech analytics can identify the calls where the agents do go “off script” and the lengths of these calls can be analyzed to see just how much it affects AHT.

Script adherence is also important for legal and regulatory compliance. The set phrases, such as security questions, warnings that calls will be monitored and recorded and the verbatim recital of terms and conditions can all be input into the speech analytics application. In this way, agents’ adherence can be monitored quite closely.

The “yes or no problem”

Another area which is ripe for automation is the verification of verbal consent. Not all speech recognition systems can recognize individual words such as “yes” or “no”. In addition, without some kind of context, it is difficult to be sure that they would recognize the specific “yes” or “no” you are looking for in the call. What is likely to work better is to include the agent’s question to which the “yes” or “no” is the required answer in the phrase that is the search term with a conditional “yes” or “no” added to the end. In the early stages of implementation, a certain degree of trial and error will be needed to identify and input the variants of “yes” or “no” which customers may use. UK customers, for example, might say “Yeah”, “Go on then”, “Why not?” or “Aye” instead of “Yes”.

Certain customer phrases may also be worth analyzing. Listening to successful sales calls will enable the sales manager to identify commonly used “buying signals” which show the customer’s interest in the product. These might include: “How much does it cost then?”, “When’s it going to be delivered?”, “Have you got it in green?” or “When do I have to pay by?”. Identifying the calls these appear in and comparing them with the calls’ outcome or wrap-up codes may help the sales manager quantify opportunities that are being lost and focus on where his sales agents need additional training and coaching.

In the support environment, customers use specific phrases that relate to certain issues, products or services. By entering these phrases into the speech analytics application, the system will produce an analysis of how many calls are made which relate to certain issues without depending on codes input by agents. It may even be a worthwhile exercise to analyze how closely the agents’ issue codes match the results of the speech analytics, and then look at the correlation of the 2 sets of results against First Call Resolution (FCR) figures.

Spotting unhappy customer

Customers also express their dissatisfaction. Where they use commonly occurring phrases such as “This just isn’t good enough” or “I’m not at all happy about this”, these can also be entered into the application and reported on. Once again, these can be compared with outcome codes for specific calls which may give us further insights into how we can improve FCR.

As with many other aspects of technology, speech analytics, if used properly, will enable us to gain further insights into how are agents are performing on a global scale and help us to drill down, identify and solve the issues.

Learn more about ZOOM Solution for Speech Analytics.

The challenge for contact centers is that their supervisors are unlikely to be linguists themselves, even if they are native speakers of the language, so they have little or no expertise in language testing or evaluation.

How can supervisors be given at least a framework to evaluate an agent’s language quality?

The first stage is to define levels, the highest level will be that of native speaker. This does not necessarily mean no mistakes. How many native speakers do you hear who actually make no grammatical mistakes at all when speaking their own language? The lowest level will be that of a complete inability to convey any meaning at all.

“Define the levels – from native speakers to non speakers.”

It’s essential to put one or two levels in between the two extremes to enable supervisors to allocate their non native speaking agents to realistic levels. The upper level should be based on agent’s being able to be understood without effort by a native speaker without necessarily being at a native speaker’s level of accuracy. The lower level should be based on the agent being able to be understood by the native speaking customer, but only with some effort.

In addition to defining levels, we can complete the matrix by defining evaluation criteria within the language framework. The most important will be an overall assessment of comprehensibility. The reason why our agent speaks or writes is to be understood by the customer. If our agent cannot achieve this, then having a richer vocabulary than Shakespeare or Dickens won’t be much use to our customer.

Below this level would come grammar. Again, the native speaker level does not necessarily mean that the agent’s language is error free, but it must have at least no more errors than that of the native speaker. The two levels below are more difficult to quantify and may depend on how long an average phone call is likely to be. One method might be to count the number of errors per minute and decide what numbers you can use for each of the two lower levels. For written communication, you could set a limit based on the number of error divided by the total number of words in the e-mail. This does not mean that your evaluators will count the word length of every e-mail, but if a rough sample of 20 e-mails shows an average length of 100 words per e-mail, then you can make decisions on how many mistakes are acceptable limits for each level.

A similar rationale can be adopted for vocabulary. Here, the test is whether the word is appropriate. Aproppriacy is not always an easy thing to judge, and for this reason, regular and frequent calibration meetings will be needed to ensure that all evaluators are on the same page.

“Sum up those mistakes and calculate the agent’s score.”

For pronunciation, the test to define the levels will reflect the test for “comprehension”. How much effort is it for a native speaker to understand what the person is saying? This does not always mean that a native speaker’s accent is comprehensible, in the United Kingdom, there are many people who find it hard to understand what people from Scotland or the North East of England are saying, but they are all native speakers of English.

In the written language, pronunciation can be replaced by spelling. Here, a numerical approach similar to that for grammar and vocabulary can be adopted. Evaluators can afford to be harder on agents too, since many aspects of e-mails can be handled with templated answers or parts of answers which should therefore reduce the scope for spelling errors.

This framework, while greatly simplified, is based on the same principles that language examiners, such as the University of Cambridge Local Examinations Syndicate, use when evaluating how well examination candidate do when writing and speaking English and other languages. They use their frameworks for the same reason, do define levels to measure their candidates against and also as a common yardstick to ensure that they are all judging according to the same standards.

Could you compare any of your supervisors to a Roman centurion?

Holding agents to account for their performance is a specific skill, how to present an agent’s poor performance back to him without insulting or alienating him and then get agreement on what improvements need to be made and how to put them into practice. Newly promoted supervisors have their own supervisors as a model for their behavior. They were also untrained, so the cycle of poor leadership and poor results continues. The cost of this poor leadership is increased recruitment and training costs as agent attrition remains higher than the minimum and reduced operational performance because agents are not motivated to produce the maximum possible performance.

There are many techniques to hold agents accountable for their performance but they all have certain things in common.

Most methods start with the supervisor presenting the facts about the agent’s poor performance to the agent. In one respect, this should not be difficult, since most contact centers monitor their agents’ performance on a second by second basis. The difficulty comes with the language the supervisor uses and attitude he displays. The supervisor needs to present the information in an objective, unemotional manner. It’s better to concentrate on the facts and avoid using words like “you always”, “you should have” and other phrases which contain either assumptions or opinions.

In most methods, the next step is to give the agent time to respond. Many new supervisors feel uncomfortable about this because either they feel that they are right and don’t want to have to justify their decisions or because they do not know how the agent is going to react and they want to avoid a confrontation. The benefit of giving the agent this space is that the supervisor will learn a lot about the agent’s attitude. If the agent is apologetic and takes ownership of the problem right from the start, it will make solving it a lot easier. If the agent decides to adopt a confrontational tone, then it gets this out in the open where the reasons for this can be addressed immediately.

Once this has been done, the final step is to make the agent realize how he is responsible for the performance issue in question. The supervisor can kick off this part of the discussion by asking “How do you think we should address this?” If the agent’s attitude is less co-operative, perhaps the appropriate question might be: “If you were in my shoes, how would you deal with this situation?”

Once again, it’s a good idea to give the agent space to answer these questions. If the agent takes responsibility for the situation and proposes a workable solution where he will make most, if not all of the effort, to solve the problem, then all you need to do as a supervisor is to set an appropriate target and agree when to review progress. If the agent either fails to take responsibility or does not have enough experience or knowledge to propose a workable solution, then the supervisor will have to propose a solution of his own, discuss it with the agent, set the appropriate target and agree when to review progress in a follow up meeting. Unless there are very strong reasons, the responsibility to put the solution into practice should always be given unambiguously back to the agent.

With practice and support from senior management, the newly promoted supervisor can use these techniques to ensure that all issues are addressed quickly and effectively and that his team will get the attention that they need.

The control chart was originally used to present how a particular statistic varied over time. This is a very important measurement for production managers in the automotive and other engineering industries where machines produce parts that must fit together with a high level of precision. The chart shows the stability of a series of measurements carried out over a period of time. The more stable the measurement is, the more predictable the process is.

It goes without saying that contact center agents are not machines. Call quality cannot be measured anywhere near as precisely as tolerances in a machine part produced by a computer controlled lathe, so how can control charts be of any use to a contact center manager?

Would you believe this team managers' sales forecasts?

The control chart above shows the daily results of a contact center’s sales team for a month. The blue line represents the actual sales results for each day of calling, while the green represents the average, and the yellow and red represent 1 or 2 standard deviations above and below the average. The width of the band represented on the chart by the standard deviation lines is a pretty good reflection of how stable the sales process is and therefore how reliable the sales managers’ promises are to head office.

A quality manager might be very interested in the result where sales went more than 2 standard deviations above the average. They might reflect a rare moment of inspiration in the sales team, or they might reflect a new and convincing lie that a sales agent has started telling the customer. Needless to say, the days where the sales results fall below the average will also need to be investigated.

Is this quality process working?

Quality management and feedback is also a process which can be shown on a control chart. The illustration above shows a control chart for a contact center providing customer support. The blue line represents the actual quality scores over the month, the black line represents the overall trend, the green line represents the average, while the yellow and red lines represent 1 and 2 standard deviations below the average respectively.

What is very noticeable is how far 1 and 2 standard deviations are from the average. This shows that the quality process is not particularly stable. If each data point represents a quality evaluation, it is not hard to see which evaluations are pulling the score down and to investigate what happened and find out what corrective measures can be taken to follow up.

These charts can also be used for educational purposes to show agents and operations team leads the effect poor results have on the team’s overall performance. The visual impact of the chart is likely to be more meaningful than a dry recital of figures and will enable everyone in the contact center to see, understand and act on the “big picture” better.

Not a good approach to customer relations

The Cambridge English Dictionary defines empathy as “the ability to share another person’s feelings or experiences by imagining what it would be like to be in their situation”. I would suggest that agents are not required to feel empathy for the customer, but they have to be able to demonstrate empathy.

First of all, this means recognizing how the customer feels. This is sometimes easier to understand when the customer is on the phone than when the customer is communicating through e-mail or chat. The agents then have to respond appropriately to the customer’s needs and the feelings they have expressed.

To measure how empathetic an agent is, the first question on an evaluation form would be to ask how upset the customer appeared to be. The evaluator will consider the customer’s language including the tone that the customer adopts. In written communications, this can be understood by obvious signs, such as the use of capitals and explanation marks (“!”) or strong language. There are also less obvious signs that might be lost on agents who are not native speakers of the customer’s language. These include the use of adverbs or sarcasm. (“I am just asking you to fix my product” – in this example, the word “just” speaks volumes about how the customer feels.)

The second question is whether the agent’s response was appropriate to the customer’s feelings. For example, if a customer is angry about his product not working, the agent might firstly respond that he sees why the customer is angry and then reassure the customer that he is working on his problem. Many times, when a person shows that he recognizes why the customer is angry, this will reduce the customer’s level of aggression. The customer feels his grievance has been recognized, even if the agent does not necessarily admit responsibility on the company’s behalf. On the other hand, telling a customer to “calm down” will rarely if ever solve anything. The agent who says this is showing that he does not understand why the customer is upset and is, in effect, dismissing his feelings as irrational behavior.

To drive the desired agent behavior, you will need to give agents high scores for correctly identifying when a customer is angry and further high scores for giving responses that show an understanding of how the customer feels. Low scores on these questions will lead to conversations about how to recognize customers’ feelings or how to respond correctly to specific situations where a customer is angry or upset. These conversations will focus agents on how they can handle customers’ feelings as skillfully as they handle the customer’s issue with the product.

Change management back in the days of the French Revolution

For the supervisors, trying to implement the quality monitoring scheme in the face of such complaints becomes an uphill task. Supervisors find themselves unable to face the prospect of hostile feedback sessions with agents who were, in many cases, their co-workers and colleagues until they were promoted. Without the supervisors doing the vital work of evaluating and giving feedback to the team, the quality process will grind to a halt and call content will not improve.

This is the classic change management scenario, but transferred into the contact center setting. The answer to this issue is also the classic way to manage change, don’t order it, impose it or decree it. Sell it!

When you start a sales conversation with a customer, you start with a short introduction of what your conversation is about. It’s a good idea to have the change announced by the most senior member of the company that is available. This is a very visible way to show everyone how seriously the company takes the project.

The next step is to give the agents the chance to react. Their reactions will probably not be positive, but they will give the agents the chance to express their views immediately and get them out into the open. They will give you, the person “selling” the new program, an insight into their “objections” which you will need to overcome. You may find that agents start asking questions. Questions are a “buying signal”, they show that the agents are engaged with the idea. They are starting to imagine what it will be like working in the contact center after the change has been made.

It is a very good idea to analyze their reactions and tailor elements of your presentation to answer agents’ objections and questions. Where possible, emphasize the benefits of the new way of working to them. If calls are going to be recorded, tell agents that this will be for their protection, since if any customer accuses an agent of lying, they can now check the call and conclusively prove to the customer that this was not the case. There may be some questions you cannot answer, either because you have to trial the program to see what will happen, or alternatively because you are waiting for decisions from other parts of the organization. To retain the trust of your “customers”, the agents, be as transparent about this as possible. Good decisions do not need to be supported by lies. At this point, it also helps to get the agents involved in the transition process by asking for their feedback. There is nothing wrong with promising that adjustments will be made as a result of their feedback, but never give your staff the idea that their feedback can lead to the cancellation of the project.

Once the “go live” date has been reached, don’t stop engaging with your staff. Ask them for feedback once it has started and involve them in review meetings. Follow up is vital, not only for the agents, but also the supervisors, whose workload has probably been increased and who now find their roles have become more confrontational as they hold their agents to account more often. In many cases, this will be the first time they have found themselves in such a situation and they may either not raise the issue at all or act aggressively to their agents unless they are supported properly. At this point, supervisors will need additional support as they learn how to really lead their agents and drive performance, in many cases, for the first time in their lives.

A lot of effort goes into getting the technology right when introducing quality programs, but often getting the people right will have an equivalent, if not greater impact.

Your evaluation forms, are they vintage or vinegar?

Identify what the key drivers of your targeted KPI are. For example, we know that average handle time can be effected by product knowledge, script adherence, system use and call handling skills. These drivers may change as your agents become efficient in meeting them and require updating.

Group structure and group related questions. Key drivers should be broken down into groups containing related questions and given a weight of importance depending on how importantly that group effects performance or meeting your KPI goals. For example, “Product Knowledge” will be the title of one of your question groups for addressing AHT. Additional groups and questions may be added as important areas are discovered. Group weightings may also shift as agents become efficient in specific areas. When agents become very efficient in System Use, shift the weighting of importance to Call Handling Skills.

Questions should have more than two possible answers that provide scoring depth and allow for richer analytics. Questions should directly assess a targeted area. For example “Did the agent understand the customer’s needs by restating the customer’s issue?” Using answers such as Yes -100% No- 0% leaves very little area for analytics. If the answers had various scoring such as Yes-100%, Partially 50%, No-0% this will indicate that the agent can easily be coached to correct behavior instead of needing modular training which saves time for everyone. Taking it a step further by providing four answers will eliminate the option for the evaluator of just taking a middle ground stance. This will push them to provide an answer that can truly identify root causes.

Design each group and question not only for analytical scoring structures but also for coaching feedback and maintenance of historical feedback. When evaluations are completed, agents should have access to them and be able to take in valuable constructive feedback, aligning their work with the QA standards. This will enhance any coaching or one to one sessions that may be necessary to correct behavior. Users should see a track record of improved performance across the key drivers of any KPI.

The evaluation form is the back bone of all quality assurance campaigns. Using this as a foundation to address your KPIs will provide a solid starting point for your path of continual improvement. Ensure your evaluation form matures along with the improvement which will drive your goals.

When it comes to motivating an agent to change their behavior, if they come up with the idea to make change on their own they are more likely to do it. I’m specifically talking about agents self-evaluating their own interactions. I, as well as my peers, have found that coaching, training and prescribing techniques or changes to agents is met with open arms but quickly fades as the coaching or training session closes. They can walk out of the meeting with all the best intentions, but as soon as that first call comes through, somehow that previous discussion is a distant memory. This is only to be proven through countless feedback forms going back, in some cases for years, that address the same issues.

I have seen agents come in bright eyed with ideas of their own on how to improve their call handling or sales performance following self-evaluations and make those changes. They were empowered, enlightened and willing because it was their own idea, their own realization! Our jobs are much easier then. I would rather praise an agent every time for making a change they decided to make themselves and coach them to support their goal than to have those discussions that disciplinary action may be required if they don’t correct their behavior.

Some will argue that agents would score themselves intentionally high with hopes of increasing their QM statistics. Some will argue that it is too expensive or time consuming to take the agent off the phones to conduct their own evaluations. I have found, more times than not, that people are fairly self-critical and judge themselves with higher expectations. If there is any question, that is what the calibration process is for. I personally have found that agents can complete self-evaluations during idle time and make changes in their behavior without as much input from coaches or supervisors.

"The unexamined life is not worth living." Was Socrates the father of self evaluation?

When an agent uses a grading form designed by the key stakeholders in QM, they instantly have transparency about what is expected of them. They have the rules of engagement. They will work with the understanding of what is important to the company, to customers and where the important changes are needed and they make them.

Don’t get me wrong, coaching and training maintains its high level of importance and is extremely necessary. Why not supplement it with an additional tool? For those of you whose agents don’t conduct self-evaluations, I highly recommend giving it a shot. It will give them insight into company expectations, empower them to make change, and give them ownership!

Contact centers are service delivery organizations, so are airlines and hotels. Airlines usually buy their aircraft from the same manufacturers. They all fly similar routes and at the same speed. Most hotel rooms contain an en suite bathroom, TV and double bed. Breakfast is usually served as a buffet. The differentiation is in the service. It’s the service and positive customer experience that keeps customers loyal to brands when there are cheaper alternatives.  It’s the service and positive customer experience that turns customers into fans.

Great service that sets an organization apart from the rest is not achieved by good luck, nor is it achieved by magic. Good service is achieved by developing a robust and rigorous set of standards,  making sure that staff are trained to handle all eventualities in accordance with these standards and running frequent checks and feedback sessions to ensure that standards are properly adhered to.

Excellent service: technology, passion and hard work combined

To achieve this in the contact center environment, the contact center needs the right technology to enable them to monitor interactions, evaluate them quickly, impartially and accurately and deliver the feedback to agents.

What the contact center also needs is hard work from the management team itself. Success in contact centers comes from what people do. Someone has to listen to and evaluate the calls, and then give agents the appropriate feedback and follow up. Missing evaluations is like taking your eye off the ball when playing competitive tennis. It will result in failure.

Petra Kvitova, Wimbledon women's champion 2011, keeping her eye on the ball.

Like most matches, the battle to achieve great service is won by a little bit of technology and know how and a lot of determination and persistence.