Generate business value with speech and text analytics
- Genesys Cloud CX 3 Digital or Genesys Cloud EX license
Genesys Cloud CX 3,Genesys Cloud CX 1 WEM Add-on II or Genesys Cloud CX 2 WEM Add-on I license
This step-by-step walkthrough demonstrates how to successfully configure and run your speech and text analytics solution.
Define initial KPIs
The foundation of your speech and text analytics system must begin with a specific key performance indicator (KPI). For this reason, you must ensure that your topics support analysis of the targeted KPI.
Common KPIs include the following examples:
- Optimize agent performance
- Monitor agent language: greeting, closing, compliance
- Measure agent skills: build rapport, express empathy, ask for sale, create urgency
- Process adherence: contact capture, identification, check for resolution
- Improve customer experience
- Improve first call resolution: contacted previously, never heard back
- Improve customer satisfaction: complaint, lawyer mention, profanity
- Reduce handle times: communication issues, system slowness
As part of the initial speech and text analytics rollout, answer these key questions:
- What are the company’s primary business objectives?
- What KPIs does the company use to measure its success?
- How are these KPIs currently measured?
Identify achievable goals
The KPI you target must be clearly defined and measurable. Make sure you define goals that are realistic and ask yourself what your organization currently does to meet those goals.
As part of the initial speech and text analytics rollout, answer these key questions to identify goals that will be targeted by the speech and text analytics solution:
- What are the company’s primary business objectives?
- What is your realistic goal for <insert KPI> this year?
- Have you been meeting this goal? If not, how far off are you and have you ever met this goal in the past?
- What are you currently doing to meet the goal?
- What external influences affect meeting the goal?
Common concrete and achievable goals a speech and text analytics solution can help an organization attain include the following examples:
- Reduce customer care call volume by 2% over the next 6 months
- Reduce the AHT (Average Handle Time) for activation calls by 15 seconds over the next 3 months
- Raise CSAT (customer satisfaction) survey scores from 65% satisfied to 70% satisfied over the next 3 months
Estimate the KPI impact
The KPI should have a financial impact and/or be recognized as important to the business strategy.
To help you determine the business strategy impact speech and text analytics will help you achieve, answer these key questions:
- If we meet these goals, what will be the effect on the bottom line?
- If we meet these goals, how much will we be gaining as extra revenue?
- If we meet these goals, will money we be we be saving?
- How do these goals impact the organization’s business strategy?
The following is a good example of a goal’s bottom-line assessment:
- Reducing AHT by 15 seconds on activation calls will shave off the equivalent of 2 FTEs (full time employees) each year, which translates to $80,000.
Define a baseline and determine how results should be measured
Understand how the KPIs you will target with speech and text analytics are measured today, and determine the baseline from where you will start measuring results.
To determine how to define a baseline and how to measure your speech and text analytics goals, answer these key questions:
- What is the current baseline, and how is it measured?
- What can we measure directly in the speech and text analytics system to approximate or predict external metrics?
The following is a good example of how a speech and text analytics system is tailored and used to set a baseline to measure a specific agent skill:
Upsell attempts are currently self-reported by agents, but they may not be accurate reflections on the true upsell attempt percentages. To measure the percentage of upsell attempts made by each agent during their calls, the Upsell topic will be created to include all the common upsell phrases and it will be run against agent calls. The initial results will be sanity-checked against the current self-reported numbers. When necessary the topic will be altered to include more phrases.
Turn on voice transcription
Voice transcription transcribes contact center voice interactions (that is, audio) into text that is stored as speaker separated conversational language. Voice transcripts generated by the system are shown in the Transcript tab as part of the interaction detail. For more information about how to enable voice transcription, see Configure voice transcription.
Create a default program and out-of-the-box topics
Programs are packages of topics that instruct speech and text analytics which business level intents to look for in recorded conversations between interaction participants. Programs are mapped to specific queues or flows and can contain topics of varying languages and dialects.
When voice transcription is enabled a default program is created. The Out-of-the-box topics will be deployed against the newly created program with the topic language determined by the Default Country Code setting (Admin > Account Settings > Organization Settings > Settings). All the interactions will be transcribed and analyzed against this program and its mapped topics.
To create and edit a program, see Work with a program.
Should you create more than one program?
Programs are a package of topics that instruct the speech and text analytics solution about which business level intents or topics to look for in recorded conversations or digital transcripts between participants based on the flows or queues mapped to the program.
Your organization should be setup with a single default program that contains the required topics to be analyzed. A single program can also contain topics from many different languages/dialects. The system will select the correct set of topics from default program based on the detected or provided language.
When you have multiple lines of business, distinguished by flows or queues, a separate set of intents (that is, topics) can be detected for each business. To do this a separate program is created for each line of business. The program should contain topics unique to the line of business, but it can also contain topics used in other programs.
Create a custom program
If it is determined that multiple programs are required in your organizations, custom programs can be created on an ad-hoc basis.
It is important to ensure that each program is mapped to the intended queues or flows that will handle the type of interactions for which the program was created.
To create and edit a program, see Work with a program.
Enable transcript content search
Perform the following steps to enable transcript content search.
- Select Admin > Quality > Speech and Text Analytics.
- Turn on the Transcript content search toggle option.
Turn off topic detection in the IVR (interactive voice response) part of the voice interaction
To disable topic spotting in the IVR, remove the default program from the speech and text analytics settings, and map the queues to the desired topic spotting.
For detailed steps, see Topic spotting – Can I turn off topics detection in the IVR leg of the voice call?.
Setup speech and text digital interaction
To enable sentiment analysis and topic spotting for digital interactions (email, chats, and messages), you must set an expected dialect (language) in the speech and text analytics settings page.
For more information, see Speech and text analytics.
Setup out-of-the-box topics
A topic is made up of phrases that represent a specific intent (for example, cancellation). Each program consists of one or more topics that outline the topics of interest to be detected in the flows or queues that are mapped to the program.
When a topic is included in a program, the system searches for all the phrases included in the topic’s definition, in all the interactions associated with the program, based on the flow and queue mapping.
When one of the phrases is found, it is identified as an event and the topic registers as found at a specific time during the interaction.
To review out-of-the-box topics and decide which ones you want to deploy, see Out-of-the-box topics.
Review and edit out-of-the-box topics
Not every out-of-the-box topic will be applicable to your organization. Genesys recommends that you review the out-of-the-box topics list, consider the underlying phrases shortly after deploying them, and decide which topics and/or phrases are worth keeping, require editing or should be removed altogether.
When reviewing these topics, consider both your organization’s vertical and the KPI(s) your speech and text analytics solution will target.
For more information, see Out-of-the-box topics.
Create custom topics
To support the identified KPI(s) you must create topics that are tailored around your specific industry, business and use case. Identify these topics and ensure that they support a business goal.
Follow these steps when creating custom topics:
- Define the topic – Each topic should have a goal in mind and be aligned with the agreed KPI(s).
- Assign a name to the topic – Ensure that the topic has a name that is self-explanatory so users will understand the purpose of the topic (for example, Customer Escalation, Mini Miranda Rights, and so on.)
- Populate the topic with phrases.
- Speculate possible phrasing
- Collect phrases while reviewing to actual customer/agent interactions
- Add possible variations of phrases
For more information, see Work with a topic.
Add words to dictionary for improved transcription
You can add new or custom words to the dictionary from the transcription engine. To do this add a phrase or a set of phrases with the new word(s) into a topic.
Topics help boost the recognition of specific words and phrases during voice transcription. They adapt the underlying language model to look for organization-specific language in conversations.
Use the content search to find meaningful information
The Content Search view displays interactions that contain a transcript of the conversation between external (customer) and internal (IVR, ACD – Automatic Call Distribution, agent, conference, or voicemail) participants, based on filter criteria.
Content search – transcription search
- Search for interactions that contain specific words in a voice transcript.
- Enter the word or words you want to find in the Filter by transcript content field.
- Filter for specific words that are an Exact match, Similar to, or Not similar to.
- Filter for multiple words by entering additional words and searching again.
Using the transcription search you can find very specific areas of opportunity that you might not be covering within your set of topics. For example, if you are interested in finding and quantifying how many customers are threatening to pursue legal action against your organization, you can search for interactions where the external participant (customer) stated “I will call my lawyer” or a similar phrase. Based on your query results you could decide to build a topic around this theme and more closely track this type of interaction.
Content search – sentiment score search
Filter interactions according to the customer’s overall sentiment from -100 to +100.
This score weighs all positive and negative markers at the end of the interaction to provide an indication of how the customer experienced their interaction with the contact center.
By searching for a low sentiment score interactions (between -50 to -100), you can locate interactions where the customer left the conversation feeling angry and/or frustrated. On the other hand, you can search for interactions with a high sentiment score (between 80 to 100), where the customer specifically indicated their satisfaction when finishing the conversation.
Once the interactions are located, they can be analyzed to find the root cause of the customer sentiment, plan your next steps (for example, contact the customer, agent coaching), or reward an agent for his/her work.
Content search – sentiment trend search
Filter interactions according to the customer’s sentiment trend.
The customer’s sentiment trend is determined by comparing the sentiment in the first half or more of the interaction, to the sentiment in the last few phrases of the interaction.
By searching for low score trend interactions (between -50 to -100) you can locate interactions where the customer had a steep sentiment decline between the first half of the conversation and the second half of the conversation. Also, you can search for interactions with high sentiment trend score (between 80 to 100) interactions where the customer expressed negative sentiment in the first half of the conversation, and had their experience changed to a positive one in the second part of the conversation. Once the interactions are located, they can be analyzed to find out the root cause of the customer sentiment, plan next steps (for example, contact customer, agent coaching), or reward an agent for his/her work.
Data analysis – topic trends view – BETA
The topic trends view helps paint a story by displaying data in a way that is easy to understand (for example, highlighting trends and outliers).
This view displays the list of topics and the distribution of interactions based on the user selected filter criteria (eg. Date range, media type, etc).
- Plot specific topics to analyse business trends and customer issues
- Compare the distribution of multiple topics (such as, interaction or cancellation reasons, product popularity, or competitor mentions), across your body of interactions
- Utilize the comparative data obtained from the report to identify the optimal ways to target customer issues and/or agent performance opportunities
For more information, see Topic Trends Summary view.
Data analysis – agent training using speech and text analytics
Improving agent performance is one key area where speech and text analytics can be instrumental to achieve your organization’s goals. To attain this benefit, it is important to follow guidelines for developing agent training programs in which speech analytics is used to research optimum agent behaviour, track improvement in agent skill use, and improve a specific performance output for the company.
For more information, see Agent training using speech and text analytics.
To successfully deploy and use a speech and text analytics solution, your organization must be setup to rapidly act upon receiving actionable data. Your organization must also be able to provide feedback to the platform administrators about new business opportunities. These new business opportunities can be added to the solution in the form of topics that will generate new actionable information.
The following roles are suggestions
Roles and profiles have proven to be key for successful speech and text analytics deployments. Roles may change based on the nature of the project. The roles listed below are necessary for complex projects requiring enterprise wide analytic and process reengineering efforts as well as agent level interventions.
- Executive sponsor: Provides guidance and support to the speech analytics project, ensures people stay focused and accountable, and have the time and resources they need to ensure success
- Project manager: Develops and oversees the speech analytics project, keeps everyone on track, and manages dependencies and risks
- Business analyst(s): Analyzes calls and looks for opportunities to improve performance
- Process engineer(s): Works with analysts to validate opportunities for improvement, develop recommendations, and present recommendations to groups within and outside the call center
- Quality managers: Uses speech analytics to identify agent best practices. In the absence of a Quality Assurance department, this function may be performed by an analyst or training personnel
- Learning and development manager: Converts quality manager and analyst findings into training programs, and tracks training delivery
- Operations managers and supervisors: Manage the agents and ensure they learn and follow the best practices. Deliver coaching sessions and other support activities as needed
- Project managers in other departments: Take findings from speech analytics process engineers, validate, refine, and execute them for their areas of responsibility
With speech and text analytics users within your organization can gain deep insight into all interactions that are occurring in your contact center.
The following represent several use cases that are valuable to your business.
- Quality managers and supervisors can use the speech and text analytics output to measure agent performance of key skills or behaviors. This allows them to identify areas of focus for coaching or recognition in a much more automated and comprehensive manner
- Business analysts can use the results of speech and text analytics to visualize and explore information that describes business performance based on what was said by customers or agents during the interaction. This can be used to identify issues and find opportunities to improve your business
- Risk managers can better protect customers and the business by identifying high risk interactions that may include complaints or inappropriate agent behavior that should be investigated or mitigated
Follow the three steps below to extract the data discussed in the three use cases above.
Turn the audio from the contact center voice interaction into structured data (for example, text). The structured data can then be used for large scale analysis, accomplished with voice transcription. For more information, see Configure voice transcription.
Define key topics of interest within conversations. The topics of interest will depend on the KPIs (Key Performance Indicator) and use cases you are targeting.
For quality managers and supervisors, decide which agent skills or behaviors you want to track. For example:
- Compliance language
- Build rapport
- Express empathy
- Check for resolution
Most objective evaluation criteria and some subjective evaluation criteria can be measured by detecting phrases within transcripts.
For business analysts, understanding why customers contact you is the starting point of any improvement program. The reasons are usually specific to your business and require an internal discussion to gain consensus on what needs to be tracked. For example are:
- Balance inquiry
- Billing issue
- Cancel mention
- Make payment
In addition to defining contact reasons, you should also create topics around specific products or services that your organization provides. This combination is an important part of understanding your customers.
For risk managers, detecting phrases that put the business at risk is most importance. This may range from language that indicates fraud by the customer or agent, severe complaints or legal action threatened by customers, or specific compliance language that must be communicated on calls. Topics can be created to watch for these markers in conversations.
Review the resulting data in analytics views to make conclusions about performance, and act based on what the data reveals.
Quality managers and supervisors should look at the Topics tab in the Agents view to isolate top and bottom performers on measured skills and behaviors, so that they can isolate opportunities for coaching or recognition for each specific agent.
Business analysts should look at the Topics Trend view to see if there are any unusual trends in call reasons, or mentions of products or services as defined by topics. It is often useful to look at this information according to handle time so you can see what types of call reasons resulted in longer calls. These call reasons should be assigned process improvements to speed up these interactions.
Risk managers should periodically view trends for key topics, or perform ad hoc searches to investigate any concerns raised by the business. Identifying high risk interactions where there may be complaints or inappropriate agent behavior that should be investigated or mitigated enables improved customer protection.