GB2522711A - Contact Processing - Google Patents

Abstract

A call processing apparatus operable, in respect of incoming calls which are routed to respective agents selected from a cohort of two or more agents, to detect, as a parameter of the agent to whom a first call was routed, whether an incoming call represents a second or subsequent call by the same caller within a predetermined time period, comprises an agent performance detector operable to detect, for each agent, a respective agent performance indicator dependent upon the proportion of calls to that agent which are followed by a second or subsequent call by the same caller. An agent impact detector calculates an agent performance indicator, referred to as agent impact, using various techniques. The agent impact detector therefore provides an example of an agent performance detector operable to detect, for each agent, a respective agent performance indicator dependent upon the proportion of calls to that agent which are followed by a second or subsequent call by the same caller e.g. using Automatic Number Identification, ANI.

Description

CONTACT PROCESSING

The present invention relates to contact processing, for example in relation to a contact centre.

The use of contact centres as a means of interfacing between businesses and customers is now well established. An example of a contact centre is a call centre in which multiple operators or agents acting on behalf of a business speak in turn to customers, either as outgoing calls for marketing purposes or the like or in response to incoming calls from the consumers, for example to service enquiries or purchases made by the customers.

In general terms, customer satisfaction in relation to the customer's experience with the contact centre is very important to the businesses concerned. Not only is there a danger that a dissatisfied customer may move his or her custom elsewhere, but also the dissatisfied customer may relate his or her negative experiences to other potential customers. On the other hand, a highly satisfied customer may recommend the business to others.

However, amongst all the other performance metrics which may be used to assess the operational performance of contact centre routing software and the like, and the work performance of contact centre agents, customer satisfaction is in fact difficult to measure. It is possible to ask the customer, for example by means of a customer survey, whether the customer is satisfied with his or her experience with the contact centre. However, there is a danger that many customers would ignore such a survey or even resent the time it would take.

One possible proxy for customer satisfaction which can be measured is to detect so-called "first call resolution". In other words, was the customer's enquiry or issue fully resolved by a single contact to the contact centre? If so, it can at least be inferred that the customer was satisfied with the experience. On the other hand, if the customer had to make repeated contacts on the same issue, it can be inferred that the customer may have been dissatisfied with the experience. Of course, such inferences may be inaccurate on an individual contact basis, but taken over a large population of contacts they can provide an adequate indication of levels of customer satisfaction with the contact centre experience.

It is a constant aim to improve the technology and operation of contact centres.

The invention is defined by claim 1.

Further respective aspects and features of the invention are defined by the appended claims.

Embodiments of the invention will now be described with reference to the accompanying drawings in which: Figure 1 is a schematic diagram of contact handling apparatus associated with a contact centre; Figure 2 schematically illustrates a telephone call made via an Internet telephone provider; Figure 3 schematically illustrates a repeat call detector; Figure 4 is a schematic flow chart illustrating features of the operation of the detector of Figure 3; Figure 5 is a schematic flow chart illustrating a process for populating a contact blacklist; Figure 6 is a schematic flow chart illustrating an overview of a process for changing contact routing according to a detection of repeat contacts; Figure 7 schematically illustrates the functionality of Figure 3 applied to different types of contact; Figure 8 schematically illustrates an IVR and a queue controller; Figure 9 schematically illustrates a queue associated with a single agent; Figure 10 schematically illustrates a queue associated with multiple agents; Figure 11 schematically illustrates multiple queues partitioned according to contact or call type; Figure 12 schematically illustrates an ordered hierarchy of queues; Figure 13 is a schematic flowchart illustrating the generation of an agent impact as an agent performance indicator; Figure 14 schematically illustrates a process for assigning a financial value to an agent impact; Figure 15 schematically illustrates a process for routing contacts according to agent impact; Figures 16 and 17 schematically illustrate routing processes according to agent impact; Figures 18 and 19 schematically illustrate queue allocation processes according to agent impact; Figure 20 schematically illustrates agent impact across a cohort of agents; Figure 21 schematically illustrates a recording and/or monitoring process triggered by agent impact below a threshold impact value; Figure 22 schematically illustrates the detection of a distribution of agent impacts; Figure 23 schematically illustrates example repeat call distributions; Figure 24 schematically illustrates the detection and analysis of repeat calls and transferred calls; and Figure 25 schematically illustrates example results of the analysis of Figure 24.

Referring now to the drawings, Figure 1 is a schematic diagram of contact handling apparatus associated with a contact centre.

In the discussion which follows, the term "call", referring to a voice call, may be used as an example of a more generic "contact", which can include not only voice calls but also Internet chats, e-mails and the like. Accordingly, unless the technical context specifically requires it, any references to "calls", "chats", ore-mails" should be treated merely as examples of contacts.

Incoming calls are initially handled by an interactive voice response (IVR) unit 10 which, in an automated manner, obtains information from the caller about the nature of the call by using one or more menus 20 and accepting user input in response to those menus. For example, the user input may be in the form of spoken words which are recognised by speech recognition modules within the IVR 10 or by keystrokes on a telephone keypad for example. A further source of user input relates to the telephone number which the user dialled to access the contact centre. For example, a contact centre may have several telephone numbers all of which generate calls which are passed to the IVR 10 for handling. One telephone number may relate to general enquiries whereas another telephone number may relate to emergency situations such as a lost or stolen credit card. By detecting which number was originally dialled, the IVR 10 can make an initial assessment of the probable nature of the incoming call. To illustrate this, Figure 1 schematically shows two dialled numbers, dialled number A and dialled number B, being initially routed to two different instances of the menus 20. Generally speaking, however, if the user has called using the wrong dialled number, the menu systems are established to allow the user to re-enter the correct part of the IVR menu system.

In the case of an internet chat or email contact, the IVR may be augmented or replaced by a process that parses the text (and possibly, in the case of an internet chat, responds to the answers which the user has given to a set of predetermined questions) to ascertain the nature of the contact, which can trigger an automatic response or enable the email or chat to be queued for the appropriate group of agents.

Once the nature of a contact or call has been detected by the IVR 10, the call is passed to a routing engine 30 which routes the call to a human operator or agent 40. There may be an intervening stage of queuing during which the call is held in a queue awaiting the attention of an agent 40 if there are insufficient agents free at that time to handle the call straightaway.

Ultimately, the call is (or at least should be) passed to an individual agent 40 for handling. If necessary, the agent can transfer the call to another agent, for example if an enquiry arises for which the initial agent does not have the necessary skill or knowledge.

A call processor 50 receives call data 60 from the IVR and from the routing engine and detects aspects of the handling of each call or contact using techniques to be discussed below.

In some embodiments, the call processor provides a routing control signal 70 to control or influence the operation of the routing engine 30 so that acting together, the call processor 50 and the routing engine 30 may be considered as an example of a contact (or call) routing apparatus implementing a contact (or call) routing method. Associated with the call processor is a data analyser 80, the operation of which will be discussed further below.

A final feature of Figure 1 is a voice recorder 90 associated with the IVR. The voice recorder 90 is operable to record user calls, either for every call as received or on a selective basis for a subset of calls. Of course, if the contact type is other than a call, for example an Internet chat, the functionality of the voice recorder 90 can be replaced or augmented by a log of the textual content of the Internet chat.

It will be appreciated that aspects of the apparatus shown in Figure 1, and of apparatus described elsewhere in the present description, may be implemented at least in part by software-controlled data processing apparatus. Such software, and a storage or other medium by which such software is provided, such as a machine-readable non-transitory storage medium (for example, a flash memory or a magnetic or optical disk) are also considered as embodiments of the present invention.

Figure 2 schematically illustrates a telephone call made via an Internet telephone provider as background to some of the discussion below. An example of such an Internet telephone provider is Skype (TM). When a user makes an outgoing telephone call over the public switched telephone network (PSTN), the user's terminal (for example a personal computer or PC) 100 actually connects using an Internet connection to a point of presence (POP) 110 associated with the Internet telephone provider. A connection over the PSTN is then made from the POP to the contact centre. A feature of this arrangement which is relevant to the contact centre but not to the caller is that the automatic number identification (ANI) associated with the call is not that of the caller but instead is an ANI associated with the FOP of the Internet telephone provider. The reasons why this is relevant will be discussed below. Similar false ANI data can be associated with the call by some mobile telephone networks and by some telephone terminals within a private branch exchange (PBX) telephone system such as a company telephone system.

Figure 3 schematically illustrates a repeat call detector.

As mentioned above, so-called first call resolution or first contact resolution can provide a proxy for the successful operation of a contact centre and for customer satisfaction. If a customers enquiry is dealt with by a single contact, it can at least be inferred that the customer may have a positive view of the interaction with the contact centre -because otherwise, the customer would have made a further contact either to complain or to continue to try to resolve the enquiry. So, it is considered useful to be able to detect second or subsequent contacts relating to the same matter.

However, just as detecting customer satisfaction itself is difficult as an automated level, so is detecting whether a contact relates to a previously handled matter. A proxy measure is again used, in that if a second or subsequent call is made by the same caller within a certain time period, it is inferred that the second or subsequent call is part of the same overall enquiry and relates to the same subject matter as the previous call or contact within that time period.

A difficulty then arises in that to use these proxy measures of customer satisfaction and calls relating to the same matter, there is a requirement that calls or contacts from the same customer need to be identified. If the IVR provides an output which indicates the identity of the caller then such identity data can be used. However, generally an IVR does not provide such information, partly because a business's customer database is considered confidential to the business, and partly because of data protection considerations. Therefore, in the case of dialled calls, the AN I is used. If two calls relates to the same ANI within the time period, they are considered to be calls from the same customer on the same matter and the second of the calls is considered to be a repeat call.

The aspects discussed above with respect to Figure 2 then become important, in that the ANI is not always unique to the particular customer and indeed is not always correct. In the case of the ANI inserted by a PBX system, it may be that the ANI is correct (in that a call back on that number would ultimately reach the customer) but does not identify a single customer amongst several. In the case of the ANI inserted by an Internet telephone provider's POP, the number may bear no relation to the identity of the customer. Furthermore, it may be that two consecutive calls received bearing a single ANI of this type actually relate to entirely different customers, for example two unrelated individuals who happen to work for the same company or possibly two unrelated individuals who happen to be users of the same Internet telephone provider. It would be incorrect to infer that such unrelated calls in fact represent repeat calls on the same matter.

Note that although some customers might have a unique or quasi-unique (that is, unique amongst a group of customers) customer number or other identification, the system cannot rely on the customer number being provided, for example because: * the customer may not have such a number (for example not yet having closed their first purchase), or * the business does not want to burden the caller by requiring the number (for example, if a business wants the customer to buy something, why force them through an identification and verification process at an early stage), or * it is perhaps unlikely that the customer would remember their number or identification or have it to hand.

Referring to Figure 3, a system is schematically illustrated which resolves or at least alleviates this problem.

A so-called blacklist 120 is maintained, such that the blacklist 120 includes those telephone numbers (ANI values) which correspond to false or ambiguous ANI5. An ANI detector detects the ANI of each incoming call but excludes or ignores any ANI value contained in the blacklist 120.

The blacklist 120 can be populated by an operator, for example using data obtained from Internet telephone providers or the like. However, as an alternative or in addition, the blacklist 120 can be automatically populated. This is achieved in the present embodiment by a filter 140 which also receives each incoming ANI value in parallel with the detector 130. The filter 140 detects the number of instances of each ANI in a predetermined period such as 24 hours. If the number of detected instances exceeds a threshold number then the filter 140 provides the corresponding ANI to the blacklist 120 for insertion into the blacklist.

So, in the case of false or ambiguous ANI5, the likelihood is that they will be re-used by multiple users within the predetermined period and will be detected as ANI values to be added to the blacklist 120.

Advantageously, however, even if the ANI is a correct ANI, relating unambiguously to a single user, if that user makes more than the threshold number of contacts in the predetermined period then that user can be inferred to be a statistical outlier not representative of a "normal" user who is simply struggling to obtain resolution of an enquiry, so it can also be appropriate to exclude such an ANI.

The filter 140 can act to reinstate previously blacklisted ANI5 by deleting them from the blacklist 120. For example, if an ANI on the blacklist 120 is not detected as an incoming ANI, or is detected fewer than a second threshold number of instances, over a second predetermined period such as one week, the filter 140 can instruct the blacklist 122 delete that ANI from the blacklist 120.

Accordingly, as mentioned above, the detector 130 acts to detect incoming ANI values which are not present in the blacklist 120.

Optionally, an identification associator 150 is provided and, in association with a database 160 and optionally further data 170 from the IVR indicating, for example, a call type (discussed below), subdivides the detected ANI data further, for example by call type. In this way, a first call relating to a first call type would not be associated with a second call of a second call type for the purposes of detecting repeat calls. The use of the database 160 also allows data relating to other contacts such as Internet chat contacts or e-mail contacts to be linked with the ANI data by the ID associator 150.

A correlator 180 then detects instances of repeat contacts or repeat calls within one or more predetermined periods such as two hours, 24 hours or one week. Such repeat contacts or repeat calls may be broken down by call type or may be aggregated overall call types. The identity of the agent who took the first of the calls (and who, it can then be inferred, did not manage to achieve resolution with the customer) forms part of the output data of the correlator 180. The identity of that agent can be obtained from the call data 60 provided from the IVR and the routing engine as an input to the correlator 180. In the present embodiments, the detection of the repeat call is applied as a parameter of the agent who took the first call. In alternative embodiments, in the case of a chain of three or more calls linked by this technique as repeat calls, it could also (or instead) be applied as a parameter of the agent who took the immediately preceding call.

An agent impact detector 190 calculates an agent performance indicator, referred to as agent "impact", using the techniques to be discussed below.

Accordingly, Figure 3 schematically illustrates an example of a call processing apparatus operable, in respect of incoming calls which are routed to respective agents selected from a cohort of two or more agents, to detect whether an incoming call represents a second or subsequent call by the same caller within a predetermined time period, the apparatus comprising a caller identification detector operable to detect identification data associated with an apparatus from which each call is being made; and a comparator operable to compare the detected identification data so as to detect whether a second or subsequent call is made with the same identification data within the predetermined time period; in which the comparator is operable to ignore detected identification data in an identification data blacklist.

Figure 4 is a schematic flow chart illustrating features of the operation of the detector of Figure 3. At a step 200, identification data such as ANI data is detected, for example by the detector 130. At a step 210, blacklisted data is ignored as discussed above. Then, at a step 220, repeat (second or subsequent) contacts are detected by the correlator 180.

Figure 5 is a schematic flow chart illustrating a process for populating a contact blacklist.

As mentioned above, the contact blacklist could be populated manually but this process describes an automated technique. At a step 230, if there are greater than a threshold number of contacts from the same customer identification (such as ANI) in a predetermined period then at a step 240 that customer identification is added to the blacklist 120. This is therefore an example of a caller identification detector operable to detect whether more than a threshold number of instances of detection of the same identification data are made in a second predetermined time period and, if so, to add the detected identification data to the identification data blacklist.

For example, the calls may be voice calls, and the identification data may indicate a telephone number associated with that voice call.

Figure 6 is a schematic flow chart illustrating an overview of a process for changing contact routing according to a detection of repeat contacts. This is just a general overview; an example of such a routing technique is to route a repeat call to an agent selected from a group of specialist agents who are chosen and/or trained to provide resolution of potential problem situations. But specific techniques will be discussed further below. At a step 250, repeat (second or subsequent) contacts are detected, and at a step 260 the routing of those or other contacts is altered in response to the detection. These operations when carried out by the call processor 50 in cooperation with the routing engine 30 therefore represent an example of a call router operable to route incoming calls to respective agents selected from the cohort of two or more agents; in which the call router is operable to select an agent to which an incoming call is to be routed according to the detection of whether the call represents a second or subsequent call by the same caller within the predetermined time period.

Figure 7 schematically illustrates the functionality of Figure 3 applied to different types of contact. In particular, as mentioned above, the ID associator 150 can operate in conjunction with the database 160 to associate different types of contact with one another so as to detect repeat contacts even in the context of a first contact of the first type followed by a second contact of a second type. Referring to Figure 7, in the case of a voice call then at a step 300 the telephone number or ANI is detected. In the case of an Internet contact such as an Internet chat or a webpage request, then at a step 310 the IP address of the machine on which the contact was made or a cookie previously stored on that machine is detected. In the case of an e-mail contact, then at a step 320 the e-mail address is detected. These different sets of identification data for identifying the user or customer making a contact are associated with one another by the database 160. The database 160 can be responsive to user identification data input by the user as part of the contact in order to associate the different contact techniques together.

Operating in this way, the system of Figure 1 and Figure 3 provides an example of a contact handling apparatus operable, in respect of incoming contacts which are routed to respective agents selected from a cohort of two or more agents, to detect whether an incoming contact represents a second or subsequent contact by the same user within a predetermined time period, the contacts being of at least two contact types selected from a group of contact types, the apparatus comprising: a user identification detector operable to detect apparatus identification data associated with an apparatus from which each call is being made; and a comparator operable to compare the detected apparatus identification data with other instances of apparatus identification data or user identification data relating to a contact of a different contact type so as to detect whether a second or subsequent contact is made with the same identification data within the predetermined time period.

Figure 8 schematically illustrates an IVR and a queue controller 350 forming part of a routing engine.

In response to the data acquired by the IVR 10 relating to the type of call being made, the call was passed to the queue controller 350 which allocates the call to an appropriate one of a set of queues 360. When a call reaches the head of its respective queue, the call is passed to an agent 40 for appropriate handling. Various examples of these queues will now be discussed.

Figure 9 schematically illustrates a queue associated with a single respective agent, so that calls joining the queue have to wait until each of the calls ahead in the queue has been dealt with by the particular agent to whom the queue is dedicated.

Figure 10 schematically illustrates a queue associated with multiple agents, such that calls held in the queue are dealt with in a queue order by the next available agent 40 as soon as that agent finishes a previous call.

Figure 11 schematically illustrates multiple queues partitioned according to contact or call type. Here, each one of multiple queues Q1, 02, 03... is associated with a respective call type (call type 1, call type 2, call type 3...) And with a respective group of agents (41, 42, 43...).

The routing engine is responsible for allocating calls to the queues according to the call type detected by the IVR. But a queue allocator 370 forming part (for example) of the call processor has at least some control over which agents are allocated to the agent groups 41, 42, 43...

Techniques by which agents are allocated to agent groups by the queue allocator 370, for example in response to repeat call data and/or agent impact will be discussed below.

Figure 12 schematically illustrates an ordered hierarchy of queues. In many respects Figure 12 is very similar to Figure 11, except that instead of each of the queues (01, 02, 03...) being associated with a call type, the queues are instead formed as a hierarchy. This system is sometimes referred to as so-called precision queuing and operates so that a call of a particular type is passed to a hierarchy of queues which are arranged so that agents in a first group 44 associated with the highest queue in the hierarchy (01) are considered to be the most suitable agents for dealing with the particular call. However, if the agents in the group 44 are not available to take the call after the call has waited a predetermined time such as 20 seconds, the call is cascaded to the second level queue 02 which has another associated group 45 of agents. If those agents are not available within a further predetermined time such as 20 seconds, then the call passes to a queue 03 with a further associated group 46 and so on. A queue allocator 380 operates in a similar manner to the queue allocator 370 discussed in connection with Figure 7, in that it is operable to promote an agent to a higher one of the queues in the hierarchy of queues in response to that agent's performance or repeat call statistics and to demote an agent to a lower one of the queues in the hierarchy of queues in response to that agent's performance or repeat call statistics. Again, techniques for doing this will be discussed below.

Figure 13 is a schematic flowchart illustrating the generation of an agent impact as an agent performance indicator. These operations are carried out, for example, by the agent impact detector 190 of Figure 3.

At a step 400, the agent impact detector 190 detects the proportion of each agent's contacts which lead to a second or subsequent contact within the predetermined period. This proportion will be referred to as PA.

At a step 410, the agent impact detector 190 detects the proportion of all agents' contacts which leads to a second or subsequent contact within the predetermined period. In other words, this proportion is detected across the cohort of agents. This value will be referred to as Pc.

Finally at a step 420, the agent impact I for a particular agent is calculated as follows: = -V ( PA -Pc) where V = the call volume or number of calls handled by that agent over a test period.

Therefore, the agent impact value I will be positive for agents whose repeat call statistics are better (lower) than the average, and negative for agents whose repeat call statistics are worse (higher) than the average. The value I is also weighted by the call volume V so that agents who exhibit a high call volume will see a larger magnitude agent impact I than agents who exhibit a lower volume.

The agent impact I can be detected as a single value across all call types but, in an embodiment of the invention, it is advantageously detected call-type-by-call-type. Empirically, it has been noted that agents are not necessarily uniformly good or uniformly bad at handling calls, and in any event there can be other factors which contribute to the agent impact in respect of certain call types. Examples will be discussed below.

The agent impact detector therefore provides an example of an agent performance detector operable to detect, for each agent, a respective agent performance indicator dependent upon the proportion of calls to that agent which are followed by a second or subsequent call by the same caller. The agent performance indicator may be dependent upon a difference between the proportion of calls to that agent which are followed by a second or subsequent call by the same caller and the corresponding proportion applicable across the cohort of agents. The agent performance indicator may be dependent upon a number of calls handled by that agent in a period under test. In some embodiments, the agent performance detector is operable to detect a call parameter indicative of an incomplete call and to disregard calls for which such a parameter has been detected.

Figure 14 schematically illustrates a process for assigning a financial value to an agent impact.

At a step 430, the agent impact is detected as discussed above. Then, at a step 440, a financial value is attributed to the agent impact. There is techniques are available for applying a financial value to the agent impact and one example, according to an embodiment of the present invention, will be discussed here. In this example: = l x (k-i) x AHT x wage I shrinkage where 1caIIs is the agent impact value discussed above and detected at the step 420, ln is a financial value attributed to that agent impact, AHT is the average handling time of each call by an agent, wage is the amount that the agent is paid, shrinkage is a measure of how much effective time is usable in a working day (so that if every minute of the working day can be used to answer calls, shrinkage = 1) and the factor k is given by the following: k = average total work if a call is repeated / AHT Here, the average total work if a call is repeated represents the sum of times taken to handle the original and the repeated call(s). The factor k may vary by call type, and the multiplier wage will vary from agent to agent.

This therefore represents an example of the agent performance detector being operable to detect a monetary value or cost associated with each agent's agent performance indicator.

Figure 15 schematically illustrates a process for routing contacts according to agent impact. Again, this represents the process at a high level and further details will be given below.

At a step 450, the agent impact is detected and at a step 460, contacts are routed according to the detected agent impact. This therefore represents an example of a call router operable to classify incoming calls by call type selected from a set of two or more call types; a call processing apparatus operable to detect an agent performance indicator for an agent in respect of at least some of the set of call types; and in which the call router is operable to route a current call of a current call type to an agent selected in dependence upon the agent performance indicator for that agent in respect of calls of the current call type.

Figures 16 and 17 schematically illustrate example routing processes according to agent impact. In Figure 16, if an agent n has a positive agent impact for a particular call type X, then at a step 500 call type X is preferentially routed to that agent. Examples of how this can be done are to use the queue allocator 380 to move the agent n into a higher level queue in the hierarchy for that call type, for the queue allocator 370 to move that particular agent into a queue for that call type and/or for the call processor 50 to instruct the routing engine 30 that if there is a set of available calls which could currently be routed to the agent n, one of the set of calls which is of call type X should be so routed. In Figure 17, the converse applies, so that if an agent m has a negative impact value for a particular call type Y then at a step 510 calls of the type Y are preferentially routed to other agents. For example, the queue allocator 380 to demote the agent m to a lower queue in the hierarchy of queues for that call type, the queue allocator 370 could remove the agent m from a queue relating to that call type and/or the call processor 50 could instruct the routing engine 30 not to route calls of that type to the agent m if there are any other calls which could instead be routed to that agent. Accordingly, in some embodiments, the call router is operable to select an agent from the cohort of agents according to an agent selection algorithm based at least in part upon the agent performance indicator for each agent in respect of calls of the current call type, so that the current call type is preferentially routed to an agent having an agent performance indicator which indicates, in respect of the current call type, a high level of agent performance.

Figures 18 and 19 schematically illustrate queue allocation processes according to agent impact. These drawings represent the process mentioned above, in which the queue allocator 380 raises an agent in the hierarchy of queues if the agent has a positive agent impact for a particular call type or across all call types (at a step 520, Figure 18) but demotes or lowers an agent in the hierarchy of queues (at a step 530, Figure 19) if the agent has a negative agent impact for a particular call type or across all call types. Accordingly, this is an example of the call router maintaining an ordered list of call queues each associated with one or more agents and having a queue priority order, the call router being operable to associate agents with the call queues according to their respective agent performance indicators.

Figure 20 schematically illustrates agent impact across a cohort of agents. In this representation, each agent is represented by a respective horizontal position along a horizontal axis, and the agent impact value I is represented along a vertical axis so that positive values of I, representing agent performance better than the average amongst the cohort are indicated by the upper region of the vertical axis. Each agent's performance is therefore represented by a vertical bar which either descends from the horizontal axis in the case of a negative I value or stands upwards from the horizontal axis in the case of a positive I value. The agents are ordered along the horizontal axis according to their impact value. A curve 540 indicates the general trend which is that there are some agents who are particularly negative in terms of their agent impact value and some agents who are particularly positive in terms of their agent impact value, and a majority of agents in between, either not very positive or not very negative compared to the average across the cohort.

This representation can prove very useful in determining which agents would benefit most from retraining. A group 550 of the worst-performing agents, as determined by this measure, may prove the best value for a retraining budget. Indeed, using the equation discussed above linking agent in fact to financial value, the possible benefit of improving the performance of the group 550 of agents can be assessed and compared with the cost of retraining.

Similarly, a group 560 of highest-performing agents may be promotion candidates or candidates for a higher pay or bonus award.

Interestingly, empirical tests show that agent performance is not necessarily uniform across all call types. Some agents are particularly good at a certain type of call and not good at another type. This can depend upon what the agent is interested in or what the agent has been trained to do. This could result in an agent having a positive agent impact in respect of one call type and a negative agent impact in respect of another call type. In such situations, in the absence of retraining or other measures, an appropriate response by the queue allocators or the call processor is to preferentially route calls to that agent of the type in which the agent has a positive agent impact, and preferentially route calls away from that agent of the type in which the agent has a negative agent impact.

Figure 21 schematically illustrates a recording and/or monitoring process triggered by agent impact below a threshold impact value. At a step 600, the agent impact for a particular agent in respect of a current call type is detected to be below a threshold (for example, a mildly negative threshold indicative of worse than average performance but not dramatically bad performance). At a step 610, this triggers one or both of a recording of the agent's interaction during that call and the active monitoring of the call by a supervisor listening into the call.

A technique will now be discussed relating to a detection of the statistical distribution of repeat call values. Figure 22 schematically illustrates the detection of a distribution of the proportion of repeat calls by agent in which, at a step 620, the proportion of repeat calls is detected for each of the cohort of agents. At a step 630, the statistical distribution across the cohort of agents is detected by call type, and that a step 640 the statistical distribution is analysed using one or more of the techniques to be discussed below.

Note that in connection with the examples given in Figures 22 and 23, the previous definition of agent impact, namely: = -V ( PA -Pc) which is a relative measure comparing an agent to the average performance of the cohort of agents, is modified slightly to provide an indication of the absolute agent impact lABS: 1ABS = V. PA Using the absolute agent impact measure, all values are positive, and smaller values (representing lower repeat call proportions) are indicative of better agent performance, and higher values are indicative of worse agent performance. The weighting V is still retained, but in some other embodiments V might not be used so that: 1A58 = PA To illustrate these techniques, Figure 23 schematically illustrates example absolute agent impact distributions. Here, call type is represented along a vertical axis so that different vertical positions correspond to different call types. The distribution of absolute impact values across the cohort of agents is represented along a horizontal axis using a notation such that the horizontal centre of a box 650 indicates an average (for example, the median value) for that call type, the left and right extremes of the box 650 indicate a variance (for example, the upper and lower quartile values) and the extremes of a line 660 passing through the box 650 indicate extreme upper and lower values. Such a representation is sometimes referred to as a "box and whisker" plot.

This representation can be analysed at the step 640 to detect various aspects of problems with the contact handling system.

Firstly, if the (average, such as the median) absolute agent impact for a call type is less than a threshold value, an example of which is indicated schematically in Figure 23 by a vertical broken line 710, then it may be inferred at the step 640 that an issue or problem relating to the automated routing or other non-human handling of the calls is not present in respect of that call type.

Secondly, if the variance of the absolute agent impact (for example, the inter-quartile separation, as indicated by the lateral extent of the boxes in Figure 23) is less than a threshold variance 720 (as in an example 690 in Figure 23), it may be inferred by the step 640 that there is not an issue or problem relating to agent handling of the calls or contacts.

This leaves various possibilities: (a) a high median (average) absolute agent impact (above the threshold amount) but a low variance across the cohort of agents (below the threshold variance), as in an example 670 in Figure 23. This is inferred by the step 640 to indicate a potential problem with the automated aspects of call handling and routing.

(b) a low median (average) absolute agent impact (below the threshold amount) but a high variance (above the threshold variance), as in an example 700 of Figure 23. This is inferred by the step 640 to indicate a potential problem or issue with the human aspects of call handling, or in other words to indicate an issue or potential issue with some of the agents.

(c) a high median (average) absolute agent impact (above the threshold amount) and a high variance (above the threshold variance), as in an example 680 of Figure 23. This is inferred by the step 640 to indicate a potential problem in both human and automated aspects of call handling.

The call processor may provide a notification of the detection of a problem, for example by an indicator display.

Accordingly, these operations as performed by the call processor can provide an example of a call processing apparatus operable, in respect of incoming calls which are routed to respective agents selected from a cohort of two or more agents, to detect, as a parameter of the agent to whom a first call was routed, whether an incoming call represents a second or subsequent call by the same caller within a predetermined time period, the apparatus comprising: an agent performance detector operable to detect the distribution across the cohort of agents of respective agent performance indicators dependent upon the proportion of calls to that agent which are followed by a second or subsequent call by the same caller; and a routing fault detector operable to detect a potential fault in the routing of incoming calls if an average of the agent performance indicators exceeds a threshold value indicative of a threshold proportion of second or subsequent calls. The operations as performed by the call processor can also provide an example of an agent fault detector operable to detect a potential fault in the handling of calls by some of the cohort of agents if the variance of the agent performance indicators exceeds a threshold variance. In some embodiments, an agent fault detector (as part of the is operable to detect a potential fault in the handling of calls by some of the cohort of agents if the median (average) absolute agent impact exceeds a threshold and/or the variance of the agent performance indicators exceeds the threshold variance.

Figure 24 schematically illustrates the detection and analysis of repeat calls and transferred calls in which, at a step 710 Repeat calls have been discussed above. Transferred calls represent calls transferred form one agent to another at the instigation of the transferring agent. This can be indicated by data provided from the routing engine to the call processor.

The correlation of repeat calls and transferred calls, on a call by call basis can be indicative of various aspects of the operation of the contact handling system to be described below with reference to Figure 25.

Note that both transfers and repeat calls end up with a second agent reworking the call (assuming the transfer is not for a particular purpose such as upsell). Being similar effects they are interesting to correlate because they have different causes: transfers are conscious decisions of an agent, whereas repeat calls are caused by the original agent not realizing the customer was dissatisfied.

The call processor, operating in accordance with these techniques, provides an example of a transferred call detector operable to detect calls transferred from one agent to another agent of the cohort of agents; and a call routing analyser operable to analyse the routing and handling of calls in response to the detection of transferred calls and the detection of second or subsequent calls.

Figure 25 schematically illustrates example results of the analysis of Figure 24. The relationship between transferred and repeat calls is plotted (for the purposes of this discussion) on a two-dimensional scatter diagram. Various features are illustrated in relation to the occurrence of a predominance of samples at a particular position on the scatter diagram.

These are described with respect to the following numbered regions: A region 750 representing a high repeat call rate but a low transferred call rate can indicate so-called "answer shopper" or callers who make repeat calls for inappropriate or mischievous reasons.

A region 760 indicating a low repeat call rate (and therefore the inference of satisfied customers) can indicate either a fault with the initial routing of the calls, leading to calls being systematically routed to the wrong initial agent, or possibly a so-called "transfer fraud" situation in which agents, remunerated by the number of calls they answer, transfer calls straight away even if there is no valid reason to do so, in order to free themselves to answer another call.

A region 770 showing a high repeat call rate and a high transfer rate can indicate that all or many calls are being initially misrouted.

A region 780 with some calls being transferred and some calls being repeated can indicate a few badly trained or beginner agents.

A region 790 with a low transfer rate and a low repeat call rate can indicate that routing and call handling are generally acceptable.

It will be appreciated that the above embodiments are merely examples of the present techniques and that various modifications may be made to the systems and techniques described above without departing from the scope of the present invention as defined by the appended claims.

Claims (19)

1. CLAIMS1. A call processing apparatus operable, in respect of incoming calls which are routed to respective agents selected from a cohort of two or more agents, to detect, as a parameter of the agent to whom a first call was routed, whether an incoming call represents a second or subsequent call by the same caller within a predetermined time period, the apparatus comprising: an agent performance detector operable to detect, for each agent, a respective agent performance indicator dependent upon the proportion of calls to that agent which are followed by a second or subsequent call by the same caller.
2. 2. Apparatus according to claim 1, in which the agent performance indicator is dependent upon a difference between the proportion of calls to that agent which are followed by a second or subsequent call by the same caller and the corresponding proportion applicable across the cohort of agents.
3. 3. Apparatus according to claim 1 or claim 2, in which the agent performance indicator is dependent upon a number of calls handled by that agent in a period under test.
4. 4. Apparatus according to any one of the preceding claims, comprising: a caller identification detector operable to detect identification data associated with an apparatus from which each call is being made; and a comparator operable to compare the detected identification data so as to detect whether a second or subsequent call is made with the same identification data within the predetermined time period.
5. 5. Apparatus according to claim 4, in which the comparator is operable to ignore detected identification data in an identification data blacklist.
6. 6. Apparatus according to claim 5, in which the caller identification detector is operable to detect whether more than a threshold number of instances of detection of the same identification data are made in a second predetermined time period and, if so. to add the detected identification data to the identification data blacklist.
7. 7. Apparatus according to any one of the preceding claims, in which the calls are voice calls, and the identification data indicates a telephone number associated with that voice call.

   P104303GB
8. 8. Apparatus according to any one of the preceding claims, in which the agent performance detector is operable to detect a monetary value or cost associated with each agent's agent performance indicator.
9. 9. Apparatus according to any one of the preceding claims, in which the agent performance detector is operable to detect a call parameter indicative of an incomplete call and to disregard calls for which such a parameter has been detected.
10. 10. A call routing apparatus comprising: a call processing apparatus according to any one of the preceding claims; and a call router operable to route incoming calls to respective agents selected from the cohort of two or more agents; in which the call router is operable to select an agent to which an incoming call is to be routed according to the agent performance indicator associated with that agent.
11. 11. Apparatus according to claim 10, in which: the call router is operable to classify incoming calls by call type selected from a set of two or more call types; the call processing apparatus is operable to detect an agent performance indicator for an agent in respect of at least some of the set of call types; and the call router is operable to route a current call of a current call type to an agent selected in dependence upon the agent performance indicator for that agent in respect of calls of the current call type.
12. 12. Apparatus according to claim 11, in which the call router is operable to select an agent from the cohort of agents according to an agent selection algorithm based at least in part upon the agent performance indicator for each agent in respect of calls of the current call type, so that the current call type is preferentially routed to an agent having an agent performance indicator which indicates, in respect of the current call type, a high level of agent performance.
13. 13. Apparatus according to claim 12, in which the call router maintains an ordered list of call queues each associated with one or more agents and having a queue priority order, the call router being operable to associate agents with the call queues according to their respective agent performance indicators.
14. 14. Apparatus according to any one of claims lOto 13, the call router being operable to trigger recording or monitoring of a call which is routed to an agent having an agent P104303GB performance indicator representing agent performance below a threshold level of agent performance.
15. 15. A call processing method operable in respect of incoming calls which are routed to respective agents selected from a cohort of two or more agents, the method comprising a data processing apparatus performing the step of: detecting, as a parameter of the agent to whom a first call was routed, whether an incoming call represents a second or subsequent call by the same caller within a predetermined time period; and detecting, for each agent, a respective agent performance indicator dependent upon the proportion of calls to that agent which are followed by a second or subsequent call by the same caller.
16. 16. Computer software which, when executed by a computer, causes the computer to carry out the method of claim 15.
17. 17. A machine-readable storage medium which stores computer software according to claim 16.
18. 18. A call processing apparatus substantially as hereinbefore described with reference to the accompanying drawings.
19. 19. A call processing method substantially as hereinbefore described with reference to the accompanying drawings.