Home » Uncategorized

Measuring the Contact Series Bias


Covid-19 has helped to propel the work-from-home movement. Many companies have announced plans to maintain a remote workforce. For some workers, this might represent a permanent change in scenery. Workers are now being hired specifically to work from home. For others, it could be on a rotating part-time or periodic basis such as myself. Thanks to a laptop I was supplied in 2020, I can now stay home and just keep on working regardless of illness. (It is fairly difficult to call in sick these days.) Due to how operations have been reconfigured since the pandemic, I even find it a bit challenging using up vacation time – more so that in pre-pandemic years. Perhaps for many companies, all sorts of concepts about work are being reconstructed to adapt to emerging circumstances.

Work-from-home proponents have focused on the savings to companies in terms of real-estate: e.g. fewer desks occupying less office space. While the work-from-home movement seems likely to continue, employee performance should remain a concern. Performance should be a consideration not just for companies but employees too, if they would like the arrangements to continue. Workers have to make working from home competitively advantageous for employers. In a sense, almost any work that once occurred entirely within a cubicle can be performed remotely: it is possible to locate this cubicle anywhere in the world. It makes perfect sense – given that call centres have cubicle-type operations that can be set up globally – for them to diversify their operations by going virtual. In this blog, I will be exploring some key concepts for measuring the performance of call centre agents working remotely. But my main focus will be to introduce what I call the Contact Series Bias to assess the performance of customer service workers.

1) Work Related to Sales: Not all call centre agents work in a sales capacity. But for those that do, the number of units sold is an important metric. The number of units sold per client-call-handled represents a type of success rate: e.g. 5 units sold for 20 calls handled. Through this performance lens, depending on the nature of the product, it might be unnecessary to maintain day-to-day supervisory or oversight capacity: in effect, the sales metric is the oversight. If sales are poor, then the agent is performing poorly. If a team leader sitting with the agent provides instructions or advice resulting in a decline in sales, it would seem that the team lead has merely contributed to a decline in sales. The team lead only ensures conformance – how they feel employees should do the work – which unfortunately does not directly relate to performance.

(In a remote setting, the performance metric can remain effective even without physical oversight. When agents are working from home, effective use of metrics can help make the review of live or recorded conversations more tactical rather than systematic. In fact, beyond the probationary period, the tendency to conduct a systematic review of conversations might be an indication that metrics are not being used effectively to assess the handling of calls.)

2) Work Related to Transactions: Another type of metric that can probably be maintained in relation to a remote workforce is the number of transactions, which I should emphasize is not really the same as the number of sales. Consider the following example to distinguish between a sale and transaction: a sale could be a 24-month contract for streaming video; a transaction could be a payment on the contract. In other words, the agent handling a transaction is not necessarily doing any selling. Arguably, a contract renewal represents a sale. Suffice it to say that an agent hired to handle transactions might have no sales targets. They can be judged based on how effectively they process transactions. To this end, again depending on the nature of the product, there might be little need for supervision. A metric pertaining to processing activity can represent the oversight.

3) Work Related to Service: The third type of metric is probably the most challenging since there appears to be no clear basis except the number of calls (or contact incidents such as emails) handled. These types of calls might go to customer service or technical support. I know there might be some attempt to codify the nature of such calls; but this should not be confused with the measurement of performance. The idea of performance is difficult to pin down since there can be many different needs and expectations on which to premise the metric. There might be no sales or transactions to help guide analysis – but rather an esoteric concept such as client service (satisfaction being the implied outcome of service). I will be spending the balance of this blog discussing the service metric and how it might apply to staff working remotely.

Contact Series Bias as an Aspect of Social Teratogenesis

I would not be surprised to encounter an agent describing a caller as problem client. Certainly there might be clients that are indeed problematic. But given a large number of clients spread over an extended period of time, it would be unusual for some agents to attract all of the problem clients while other agents seem to constantly get the best clients. The idea that an agent can create a problem client conforms to the concept of “social teratogenesis.” In the context of a call centre, social teratogenesis can relate to how problem clients are socially constructed. An agent can transform an ordinary caller into a problem client. Consider the mathematics: if teratogenesis does not exist – and clients of all different types were randomly distributed among agents – then performance metrics however they are expressed should be relatively comparable between agents over time. (There should be no bias or persistent skewing among agents.)

In an earlier blog, I described how the percentage of unique callers can be used to evaluate the effectiveness of an agent. (This is because an ineffective agent will tend to cause clients to call back.) The percentage of unique callers could be regarded as a type of service metric. The better the service, the less likely it is for clients to call back: i.e. the higher the percentage of unique callers. Keep in mind that at least conceptually, the service agent is not handling a sale or transaction. The client might already have a contract. Perhaps they also paid for several months of service. The client now expects to receive service. This is not a situation where the client immediately threatens to go to the competition if they are dissatisfied – although this might happen at some point later in time. The agent responsible for providing service interacts with the client before the contract is lost.

In a sense, the service agent is part of the contract-fulfillment process. The product contains a kind of service contract – whether expressed, implied, or expected – that exists in varying degrees depending on the type of product. I recall recently opening an account for online transactions and discovering a logistical problem: I cannot put any funds into this account. Although I do not recall signing a contract saying so, I certainly have the expectation to be able to move funds into and out of the account. So of course I contacted customer service. Interestingly, the last time I opened a similar account with the same company, I experienced the exact same problem. Evidently, they do not have a learning process in place. These are the sorts of problems I actually enjoy fixing. Since I am a client rather than employee of the company, all I can do is leave the account idle until the problem is resolved.

I doubt that many companies have the ability to measure the percentage of unique callers by agent or queue. Perhaps even fewer can do what I am about the describe: companies can maintain a trace-history of repetitive calls – what I have termed the Contact Series for repetitive calls. The Contact Series represents the order of contacts for each unique caller. This means that when a caller becomes repetitive, a check of the Contact Series for the unique telephone number would show the series of contacts in chronological order. Although this is a useful feature for many purposes, the network or call distribution system might not carry or compile the call log data in the desired manner. It is possible to process the downloaded data outside the network to obtain the series. My response whenever I find a compilation requirement that is not supported by the network is to write a script in Basic using Excel. It is not my language of choice; it just happens to be the most accessible.

Assume a client calls six times. The Contact Series might appear as follows: Ruth – Sam – Ted – Ruth – Wendy – Linda. (There are six agents – one for each time the client called.) In other words, Ruth is the first agent to speak with the client while Linda is the last. If social teratogenesis were not a factor, the probability of Ruth appearing on the left-hand side (LHS) is about the same as on the right-hand side (RHS). I don’t know if this reasoning leaps out as correct to all readers. Regardless of how many calls are handled by an agent – accepting the argument that there are indeed problem clients – the likelihood of the agent being the first and last point of contact should be about the same. Conceptually, it should be unlikely for bias or skewing to persist over time.

On the other hand, again taking fault as intrinsic – this is to say, an aspect of the client brought by the client to the conversation with the agent – the random distribution of calls should generally lead to a LHS/RHS of 1 or LHS/(RHS + LHS) of 0.50. Again, whether the distribution of fault occurs prior to clients calling the call centre – i.e. clients carry the fault occuring naturally – or the fault is simply distributed randomly by the network to the individual agents, the end-effect should be similar.

However, I would argue that social teratogenesis is the reality. Random distribution is unlikely to occur. This means that some agents will tend to be implicated in starting repetitive calling (on the LHS). Other agents will more likely be involved in the termination or ending of repetitive calling (on the RHS). If the skewing persists over a period of time, it becomes quite easy to separate the good and bad performers. It also becomes difficult to argue that good and bad performance is unrelated to the skills, abilities, and behaviour of the agent. In other words, the problem client is the product of social construction. The true source of fault is external contact: in my example, it is the agent. Consider the Contact Series below to assess teratology by examining the bias.

Ruth – Sam – Ted – Ruth – Wendy – Linda
Ted – Ted – Ted – Ruth – Linda
Ruth – Ted – Sam – Sam
Ruth – Ted – Ted – Wendy – Linda
Ruth – Wendy – Sam – Ted – Ted – Sam
Ted – Sam – Ruth – Wendy – Sam
Ruth – Ted – Linda – Sam

Ruth: LHS = 5 (out of 7), RHS = 0 (out of 7)
Ted: LHS = 2 (out of 7), RHS = 0 (out of 7)
Linda: LHS = 0 (out of 7), RHS = 3 (out of 7)
Sam: LHS = 0 (out of 7), RHS = 4 (out of 7)
Wendy: LHS = 0 (out of 7), RHS = 0 (out of 7)

Based on the above, Ruth and Ted seem suspiciously associated with the initiation of repetitive series. Ruth in particular triggered repetitive series 5 out of 7 times (71 percent of the time). However, whenever Ruth was involved in a series by being either on the left or right, she was on the left 100 percent of the time. In contrast, Linda and Sam seem to contribute to the end of series. If this distribution holds over many days, it seems highly improbable for the distribution to be random. It also seems unlikely for the distribution to be the result of intrinsic aspects of the clients since all of them are different. If the distribution is neither random nor internal, it likely extends from external determinants.

All things being equal – if the percentage of unique callers were about the same and the proportion of LHS to LHS were about equal for Contact Series, at this point it might be reasonable to focus on the number of calls handled. Given no difference in quality, the differences in quantity can then be compared.

Origins of Social Teratogenesis

Social teratogenesis is directly related to social disablement. The big idea behind social disablement is that disability – rather than being inherent or part of a person – is something that is socially constructed. For example, for individuals that are deaf, there is nothing unusual about their condition. Society labels them as abnormal and makes them feel out of place by making this condition a liability. If the world were not designed to accommodate mostly hearing people, being deaf would not be an issue. It is not unusual for people with disabilities to feel like criminals just for expecting the same rights and opportunities as everyone else.

As fate would have it, one day I wondered about these dynamics if a company literally has a significant percentage of its clientele as actual criminals. I believe the criminalization or objectification of clients is sometimes used to justify poor service. Clients become instrumental or mere pawns in a company’s business model. If the clients were indeed criminals, how exactly should this change the quality of service? The answer is that from the standpoint of a service provider, it should not matter at all. Clients might be made out to be monsters or problem clients (teratogenesis = monster making); but what if in fact most of the clients are normal? What if the cause of their questionable behaviour is not an inherent aspect of the client but related to how they are treated? In order for this assertion to be valid, the skewed response to environmental conditions needs to persist in relation to different agents that might be triggering a teratogenic response.

I am certain that many people have been exposed to the following excuse during Covid-19: as a result of the pandemic, call centres are experiencing unsually high levels of volume. What if as a result of the pandemic, customers are experiencing an unusually poor level of service? Customers are being forced to call back repeatedly because service is just that poor. They are declared to have unusual or inappropriate service expectations. Their needs are discounted or dismissed.

Teratogenesis might be particularly evident in the case of monopolies or oligopolies such as big box stores and financial institutions. I would argue that the objectification and diminishment of customers might be at a high point because the pandemic provides a convenient excuse; but more than this, it provides the opportunity for teratogenic environments to develop (work-from-home operations) and behaviours to gain a foothold (apathetic or incapable agents).

Now as remote workforces become more common, it is important for this preconception of a captive audience to be addressed by organizations. It is necessary to identify and deal with teratogenic determinants whether these might be in the form of agents or certain problematic aspects of operations. It is at a teratogenic level – i.e. at the service rather than sales or transactional level – where customers are lost and might decide to move on.

Implications to Stats

An agent that triggers repetitive calling benefits from the fact that clients that call back might be randomly distributed to other agents. Just by bad luck, these other agents might get caught up in the Contact Series through no fault of their own. Using the percentage of unique callers is therefore an indirect and imprecise measurement – problematic by its lack of granularity. (The percentage of unique callers is however useful for adjusting market metrics in order to remove operationally-induced repetitive calling entrained in the market stats.) In relation to individual agents, the Contact Series Bias provides much more direct guidance. In situations where service is the focus rather than sales or transactions, metrics pertaining to the Contact Series Bais can be maintained.

Again, although it would be possible to have somebody actively listening to the calls and making suggestions, the advice would appear to be moot if it focuses on agents with superior RHS metrics or if it fails to improve the LHS metrics for problematic agents. Metrics for the Contact Series Bias can serve as the oversight. Correction or mitigation by a team leader or supervisor can be based on the guidance provided by these metrics in order to minimize the misdirection of scarce resources. The metrics would help organizations fix what is specifically known to be failing. It is therefore possible to quantify among customer service agents the extent to which particular employees might be burdening their coworkers or conversely helping to make their life easier for the team.