What should be able to cool call center for IT-parts and what options are there at all

Five or six years ago, authentication by voice in the call center seemed fantastic. No less fantastic was speech recognition in search of keywords, in order to understand what happens before and after pauses longer than 3 seconds. Today it is a working solution.

Ten years ago, the automatic scheduling system was a wonder. And today, banks are saving just on installing such a system and a lot of money on the fact that they are encouraging the best operators not with bonuses, but with convenient schedules.
')
Three years ago, few people considered paper calls (loan applications, for example), letters, regular calls, messages on Twitter and posts on Facebook as identical entities that simply come to the operator’s workplace to respond.

Basic call center starts from about 10 jobs. At first, everything is done intuitively, and without special software. Then, up to 50 operators need automation only in a number of bottlenecks, everything else is solved, again, by the head or senior shift. That's when the number of operators passes for 50, it becomes more interesting. We need software or even several software solutions that fully cover the contact center infrastructure and solve many problems.

Situations

There is no call center yet. For example, the Russian representative office of the western company opens. From my practice, as a rule, this is either a new customer, or a customer to whom we have already deployed a data center, switched, he already has corporate telephony, postal services. The customer needs a single point of service for customer requests. The situation is relatively simple, because building from scratch is more often easier than refactoring.

There is KC, it needs to be expanded or problems solved. Often, you need to upgrade the auto-information functionality, give supervisors more control functions, set up a shift distribution system on a schedule, compare indicators of different operators. Operators need to quickly identify the client, see its call history, forward the identified client further along the chain.

Automatic telephone station

Before turning to solutions, I want to note that all the components discussed can be virtualized, that is, solutions are software that is easily installed on the customer’s server.

The Automatic Telephone Station component is the core of the call center. This is the basic functionality of call distribution, familiar to you from Asterisk class systems. Depending on the business requirements change. For example, banks require several levels of fault tolerance, i.e. N systems (usually 3-5) should fail to make a telephone connection fail. Scalability is important - there is no need to change the solution architecture when expanding a business, you do not need to throw away old modules.

A call center is an add-on to a PBX and implies a change in the operation of the PBX logic under the logic of the CC, namely: the notion of a group of operators or operator skill emerges - the union of operators into a logical group to which the call is distributed; a telephone set that serves as a means of receiving and making calls is assigned the role of an “operator”, an entity behind which there is a living employee, by which you can watch statistics, assign groups or skills. For business, this means that all the basic functionality is added, which allows you to turn a group of employees with corporate phones into a full-fledged CC or call center.

1. Multichannel and AWP

Today the question that a person addresses, and not the channel of circulation, comes to the first place. The client no longer cares how he will receive an answer to his request, he can write to Facebook, send mail, sms and so on. All of these instances come in a single operator interface. For example, he can hang up after the call, and then not pick up the phone again, but answer the SMS in his program, then the client will send him a tweet with a question and so on. Staying in one application, without leaving the working screen, the operator answers all calls. Such systems are used in large retail, banks, operators, insurance around the world. You can try to tweet about the problem of your mobile operator - the whole “Grossen triplen” is constantly monitored. One of the shift operators will try to help you.

Everything is done automatically, including the registration of the reasons for the calls, work with all types of calls (phone call, web chat, email), receiving messages in the form of a scrolling text. The screenshot below shows the operator’s automated workplace, which processes the incoming call - the client’s card automatically opens, the phone number from which the call was received, for further work the employee goes to the caller’s personal account, where all information on operations is displayed.


Call card






SMS sending example

2. Workforce Management

WFM solutions are used to create a work schedule. Up to 50 people, everything is done in Excel, but for a hundred people, even a few compilers cannot cope. More precisely, how - cope, but far from the most optimal way. By law, the work schedule must be prepared and provided to employees in advance.

Systems of this class allow you to set a number of rules, take into account the level of qualification, additional skills, for example, proficiency in foreign languages, vacations, including study, as well as specific details, in particular, an individual schedule, and create a schedule automatically. You can make predictions - how many operators will be needed if the organization gives advertising, and the flow of requests will increase by 20-30%. WFM is applicable not only for operators who handle phone calls, but also for those who work with any kind of customer calls: through the website, email, mobile application, coming to the office personally and submitting a paper credit application.

3. Outbound call system

Imagine that you need to ring around 50 customers who left their orders in the online store at night. It is possible to dial in numbers and occupy the operator in turn, but this is not very rational. If the call center is large, it is enough to make sure that you have at least one free operator in reserve (that is, do not give him calls yet) and start the call.

Suppose, according to statistics, only 7 out of 10 subscribers answer. The three remaining at this time go to the subway, replenish the balance or speak with someone. So, roughly speaking, we need 7 operators for 10 running calls. Further, one operator fulfills 10 calls in 30 minutes. Of these, about 4 minutes, he listens to beeps. This means that if you throw out the wait, you can put 10 calls in 26 minutes. There are companies for which hit rate is only 20% (2 out of 10 calls), and in such cases the prediction system becomes very useful.

Outgoing dialing systems take these calculations into account and start connections as the operator is released. The result of the system is a finished call, where the operator enters upon the connection. That is, minus the beeps and minus dial-up - optimization.

We implemented such a system in OTP Bank - it was possible to significantly accelerate the morning round of those who have delayed the loan payment, for example. Plus, the same system makes repeated calls during the day, until it gets through (also automatically). She increased the productivity of employees by increasing the number of calls to clients 2.5-3 times during the day.

4. Synthesis of speech

Used for automatic interactive answering machines (IVR). The most common task of synthesis is to tell the caller a certain number: the balance of the cell phone, the balance on the bank account, the amount of the next loan payment. It also tells you the office hours, the nearest ATM, the time for obtaining a visa and much more. In general, IVR can access the database, retrieve information from there according to certain rules and read it to the client.

Speech synthesis is also used for outgoing calls. For example, in retail, when ordering through online stores, the robot sometimes calls back and reads the date and time of delivery, and then asks to press 1 if everything is correct. You know his brother-robot when you call support - he often asks to evaluate the work of the operator.

In general, robots have become damn chatty lately.

In my practice, several very interesting problems were solved using a dynamic IVR. One large telecom company has organized unique special offers (service packages, discounts, gift bonuses for loyal customers, etc.). There are a lot of offers, and there are a lot of different scenarios, to whom which proposal needs to be issued, ten times more. Due to the correct order of services in the voice menu, the statistics for connecting additional services showed that everything was done as it should.

Or an example from the financial industry: a classic task. There are various customer segments (most often they are divided by profitability for the organization and activity in financial transactions) which have completely different needs for the necessary information. In one voice menu, all the needs simply do not fit, if not to consider non-ergonomic monsters. Dynamic IVRs allow you to rely on business processes and business needs, which is often convenient for customers in the end.

5. Voice recognition

From the voice of a person you can make an imprint on the characteristic distortions of sound, due to the structure of the pharynx, nose and just individual features. When calling in similar conditions (from the same phone with no wind noise or an excavator in the background), the recognition accuracy of a 6-second passage is about 90-95%, depending on voice.

This means several things at once:

1. For authentication is not enough, but as a factor in confirming trust, it completely passes. For example, when you ask for the balance of the phone, the operator no longer needs to make a long identification on the passport. You are calling from the same number, the line parameters are similar, the voice coincided by 90%. You can give non-critical information without inconvenience to you.
2. When you try to fake your voice, an attacker will most likely be able to quietly copy the features of intonation and timbre, but not the physical characteristics of speech. In this regard, recognition works very precisely - while the parodists could not deceive the system.
3. You can conduct a reverse search by comparing your voice with other voices of the base. In fact, this means that in the background any call to the call center of the bank is immediately compared with the voices of famous froders.
4. You can try to find your call from another number among all the conversations. Given the accuracy, you will have to listen to more than a dozen records, but with long sections of speech - these will be hundreds, not tens of thousands of options.

More information for paranoids is here about voice prints .

6. Speech analytics

The robot is able to make out words (not everything and not always, but often) and brazenly uses it. For example, in computing capacity, tasks of searching for downtime in banking processes often revolve. The call center loads 40-50 thousand calls, the robot searches for pauses, recognizes everything that was before and after them, builds its opinion about what this process was. And at the output he says: friends, when a client on Monday morning asks if he can take a loan as a pledge of tare, the operator thinks too long. Or there is something like the fact that an insurance client in a rare case (professional athlete, going to an unpopular country, you need a very interesting insurance) can wait a couple of minutes until the operator finds in the interface how to do it. Also looking for interface brakes.

Of course, word recognition helps us to understand which operator incorrectly greets, who uses parasitic words, who uses the client, and so on. And also - what the client thinks about the call: all sorts of "no, thank you" after offering something, negative-positive feedback and much more. Details for paranoid here: habrahabr.ru/company/croc/blog/235565

Plus there is an analyst of the emotional state. While not a very accurate system, but still useful. For example, now many large Russian banks are testing similar solutions. For the first time in Russia, speech analytics systems were implemented in the National Recovery Service, which serves 83 regions of Russia. If a long pause occurs during a dialogue, people start interrupting each other or switch to obscene language, this will certainly become known to the supervisor.

7. Recording system

Problem: there was a conflict between the operator and the client. The client is right, but the operator expresses his position to the supervisor, using the knowledge of the specifics of the work. And convinces him of being right. In the absence of a real version, the supervisor believes the operator, as a result, the offended, but right-handed, customer with losses refuses services, and the operator continues to successfully use the experience gained with new customers. Writing solves the problem.

The second problem. The operator is right, but the client pretends that the CC operators did not provide any information to him, sues the customer, wins the case, the customer incurs losses.

The third situation. Operator training. I called a particularly annoying client, he was answered by a mega-polite girl, arranged a “phone sex” session, with the result that the client was satisfied and even bought something. The call was saved as a reference and was used to train three generations of operators.

Also, to monitor not only conversations, but also operator actions, a screen recording system is used.

8. Customer Feedback

Call evaluation menu Customer Feedback allows you to evaluate the work of operators, CC and ask other questions to the client after his conversation with the operator. After that, the supervisor can easily upload sets of evaluations by operators, specific issues at any time.

9. CRM Integration and Reporting

In banks, customer data is stored in one or several ABS (automated banking system), most often CRM is used for this purpose. When a call is received in CC, the operator’s card from CRM opens on the operator’s screen. The operator sees the information that you may need - recent card transactions, current balance, products that can be offered due to the fact that the client has taken out a loan. In online stores you can see past products, plus recommendations on related purchases. Age and birthdays of children, if it is, for example, a toy store. Operating system and IT-literacy, if this is the support of the provider. Data consolidation saves time and increases efficiency. Here are examples: habrahabr.ru/company/croc/blog/140490

The most difficult thing is, of course, to caste-place the regular capabilities of CRM and CC-solutions for current business processes. Therefore, we need integrators who modify the file very, very much.

In parallel, as a rule, the analyst is done. It can be implemented both within the call-center, and separately, at the level of all BI-systems of the customer. This unified system means that there is no need to manually collect and process technological data from various subsystems (call management, recording systems, personnel records, or operator workstations).


Avaya and Siebel CRM integration module

10. Monitoring

There will be problems with hardware and software. When reserving all the components of the KC, with increased and geolocation fault tolerance, in the absence of maintenance work and bona fide technical support staff of the customer, sooner or later there will be a situation when the components of a subsystem of the KC fail. Then, if CC is designed correctly, and fault tolerance is given due attention, CC work will continue due to switching to backup components, but such work is no longer normal, and if the situation worsens, a refusal to provide the service may occur. The reason, which leads to serious irregularities, may be some kind of “insignificant” trifle, which is not paid attention to for a long time, and then the result is extremely sad. In order not to miss such a trifle, react in time, eliminate the drawback and prevent major problems, use monitoring systems. Such systems analyze a huge number of parameters of the KC subsystems, react sensitively to their changes and promptly inform the technical services staff through various available channels so that the Armageddian does not happen unexpectedly. Messages from the monitoring system can be automatically registered in the accounting systems of the customer, sent out according to a certain scheme to employees, management, operating organization. Most often, e-mail is used for informing, but other channels can be used, for example, SMS or voice message by phone.

Vendors

Avaya is a leading global supplier of solutions and equipment for building communication networks and enterprise IP telephony systems. Products and services include voice, data and voice convergence, customer relationship management, multimedia messaging systems, multiservice networks, structured cabling systems. Avaya leads in sales of message processing systems and structured cabling systems, call centers and voice communication systems for enterprises.

The Speech Technology Center is a Russian developer: recording systems, speech analytics, biometric identification, speech recognition and synthesis. More details on speechpro.ru.

Nice is the world leader in voice and electronic recording solutions with a customer (IP, traditional telephony, email), the subsequent extraction of this information and its analysis. Also speech analytics, Quality Monitoring, WFM, Customer Feedback.

Aspect is the same as Avaya, only plus its own WFO line (Voice / Screen Recording, QM, WFM, Customer Feedback).

Verint is a global provider of solutions for collecting, filtering and analyzing sources of information such as voice, video, and unstructured text.

Enghouse Interactive - cloud CC. Deployed to CROC.

Zoom - recording system.

We work with all these solutions for banks, operators, insurance and large retail. If you need details, specific case studies or trial errors, I will always answer by mail DPesotskiy@croc.ru or questions in the comments.