“A speech tool is not a plugand-play technology. Your organization must be prepared to dedicate the proper resources that will ensure its success.”
Speech analytics is seen as a trendy, cool technology because it can provide insights into customer and agent communications in ways that are unmatched by other methods. For example, speech software can help you identify and improve processes for the types of calls that are difficult for agents or customers to handle, or spot the easier call types that should be directed to self-service.
Speech can also point you to the reasons for decreasing sales or departing customers, and its shared findings often ignite improved enterprise collaboration.
Decisions made based on insights gleaned from speech can quickly and directly impact your bottom line. In fact, speech analytics often pays for itself in less than a year, as long as you have clarity about why you’re buying it and what you can expect from it once you implement it.
In my decade in this industry, I’ve witnessed my share of customers experiencing buyer’s remorse. If they had only read the following tips for selecting and deploying the right speech analytics tool, they could have harvested a wealth of gold nuggets from the soil of their voice data, instead of just pushing dirt around.
If you’re considering investing in speech analytics, these eight tips will help you find the speech tool that will best match your needs and maximize the business value of your results.
8 Best Practices for Selecting a Speech Analytics Tool and Vendor
1. Don’t get caught up in prolonged discussions about accuracy. Initial high accuracy rates can be a bonus, but for some companies, accuracy isn’t even a Top 3 criteria for choosing a particular speech software.
Why? It all depends on how accuracy is defined. I’ve seen smoke-and-mirror calculations on the part of vendors that make direct comparisons very difficult. Some vendors will use scientific-sounding calculations based on errors per hour. Some will tout the accuracy of the phrases that were detected while paying no mind to the phrases that went undetected.
Remember that superior speech analytics results don’t come baked into a tool. Ensuring the highest levels of accuracy requires multiple iterations of testing. This painstaking, meticulous work involves auditing search results over and over again, adjusting confidence levels and recommending tweaks until you reach the sweet spot of accuracy that’s thoroughly aligned with your business objectives.
2. Don’t go overboard comparing features. A speech tool’s standard features and functions should be clean and intuitive and satisfy the bulk of any user’s functionality needs. With the right tool, beginning users should be able to generate basic but powerful insights within the first week of deployment.
That said, more sophisticated features can generate additional insights—but only if the user has been trained to operate them properly and interpret the results accurately.
I’ve seen too many buyers line up the features of different vendors’ offerings and base their decision on the number of ballot boxes that can be checked. Don’t let bells and whistles divert your focus from the performance of core functions. Just because a particular feature is technically possible doesn’t make it practical.
3. Consider your 12-month goals and how the software will help you meet those goals through reports. Ask about the availability of standard and custom reports along with workflows that will support your progress towards these goals.
4. Understand the merits and tradeoffs of phonetic versus speech-to-text technologies. A speech-to-text approach, which produces a digital transcript that is then searched using standard text-mining methodologies, typically provides better results for context-sensitive analysis.
In a phonetics application, the speech engine is loaded with phonemes, the smallest units of speech that make words different from other words. This application produces an index file, which is then searched for phonetic equivalents of desired words or phrases.
The phonetic approach typically results in faster processing because it doesn’t try to match spoken language to saved phrases or dictionaries. There are trade-offs to each method and both are here to stay.
5. Take a hard look at real-time speech processing. Real-time speech analytics solutions, which analyze phone calls while they’re in progress, can alert a supervisor to intervene and resolve important issues before they escalate—or prompt an agent to try a different approach, switch to a different script or attempt a specific upsell.
Of course, that assumes that the technology is working perfectly, which is a big assumption. In our view, this technology needs to mature before it performs as well in practice as it does in theory. Yet even if the technology performed as advertised, questions remain. For instance, do you have the internal resources to capitalize on call alerts in progress?
And would you want to depend on results that are only 75% to 80% accurate? False positives can create serious challenges when acted upon in a live environment. Imagine a supervisor reacting to a live call only to discover she was activating on false information.
The time will come when real-time solutions will enhance the quality of agent training, upgrade service levels and boost customer retention rates. Until then, don’t lose sight of the fact that “near real-time” or post-call filtering can generate useful and valuable results.
6. Determine whether you can integrate metadata with your speech data. Do what you can to maximize the value and usability of speech insights by incorporating metadata, which is data that provides information about other data. For example, intelligent call recording systems have the ability to tag valuable data like account identification, product name, dollar value of the sale, etc., from virtually any servicing system (e.g., CRM, ERP, help desk, home grown). You can then use all that additional data in your call- recording searches in your reporting to create high-value business intelligence.
7. Identify what support and training is included. Ask each of the various speech vendors you’re considering for a detailed overview of their post-sale support. That should include a dedicated end-user representative, a technical support person, support hours from their help desk, and documentation about technical and user training. Make sure they differentiate between training for business users, technical users and system support.
8. Don’t fall prey to an impressive demo. Get clear about what highvalue insights you desire for your business, and ask to test the system against a subset of your own calls. Consider conducting a “bake-off” of multiple speech vendors with your own recordings. This can take additional time but you’ll gain valuable knowledge about exactly what you’re purchasing.
11 Best Practices for Deploying a Speech Analytics Tool
Selecting the speech tool that’s right for your company is only the first step. Organizations are often so eager to unwrap their new toy, they’ve given too little consideration to how it will be used.
A speech tool is not a plug-and-play technology. Your organization must be prepared to dedicate the proper resources that will ensure its success, and have the infrastructure and processes in place to capitalize on the timely business intelligence that will be generated.
1. Develop a team of analysts. Leveraging the full power of your speech analytics tool requires performing a variety of administrative, analytical and fact-finding tasks on an ongoing basis. The learning curve can be steep and confusing without the presence of internal or external speech analytics experts who know how to apply operational subtleties and strategies for capitalizing on raw data.
Whether you tap internal resources or rely on outside consultants to manage your speech analytics initiative, maximizing the technology’s capabilities requires certain skill sets. Start with the stakeholders in departments that will be impacted, and build a team of inquisitive and curious analysts who have a passion for analytics.
If you choose to manage the process from within, your ideal candidates will be able to toggle between left-brain (logical) thinking and right-brain (creative) thinking. Individuals who can synthesize speech analytics data from both viewpoints will be able to analyze and articulate business opportunities and make the case for necessary changes.
These whole-brain thinkers are a bit of a rare breed, so here are minimum requirements for three essential skill sets.
ADMINISTRATOR. The Administrator is responsible for deployment of the speech analytics tool and manages all server and database connections.
BUSINESS ANALYST. The Business Analyst is responsible for executing all the analytical tasks necessary to compile, analyze and deliver key findings of speech analytics projects to upper management.
INTERACTIONS MONITORING ANALYST. The Interactions Monitoring Analyst is responsible for listening to and reviewing the content of selected calls based on criteria provided by the Business Analyst.
Depending on the depth of your internal resources, one staff member could perform multiple functions—but it’s best to train two resources in the event you’re unable to retain your lead.
2. Plan to leverage the data to effect change. Think in advance of the actions you’ll take as a result of knowing that certain phrases, patterns of phrases and trends were detected. What will you do with the significant amount of data you’ll be generating? Have you appointed one or more change agents with responsibility for identifying and capitalizing on opportunities for change? Make sure you put processes in place that enable you to get ahead and stay ahead of the technology.
3. Emphasize to your agents that this tool isn’t intended as a “gotcha” device. Your agents will likely “sit up straighter” knowing that all calls are being analyzed, even while they harbor suspicions about the new speech tool. Take pains to assure agents that speech analytics will impact them positively—both in identifying “champagne moments” to celebrate and helping build cultural trust within your organization.
Consider building the first speech project based on a positive business issue while identifying other process and performance issues for future projects that you can roll out once trust is established. If at some point constructive criticism is necessary, be sure to deliver it respectfully to maintain morale and head count.
4. Determine your first business application. Get clarity about the short-term and long-term business applications you’ll be targeting. What is it you ultimately want to learn? Is what you’re seeking obvious or subtle? Why is it important? What is your utilization strategy?
5. Define success measurements. For every business application— whether first-call resolution, agent performance, customer satisfaction or process improvements—agree on what success will look like. identify what metrics need to be measured and how to measure them before the first call is analyzed. Then factor in your business rules and objectives, define the endgame and determine how to evaluate results.
6. Walk before you run. Use your speech tool as a filter to access important calls to review for evidence to support a business issue or hypothesis, but it’s important to actually listen to those calls before jumping to any conclusions. Speech analytics is not a perfect science.
7. Put a sharp focus on a specific business issue, and start working on that issue with a single workgroup. Examples of business issues that are good candidates for speech analytics include:
- First-call resolution
- Agent performance
- Sales effectiveness
- Customer effort
- Customer complaints
- Customer churn
- Process improvements
8. Don’t rely solely on generic, outof-the-box canned phrase libraries that may have been included by the speech vendor. Instead, customize your phrase library in a way that’s meaningful for your business environment, your customer population, and your specific objectives for speech analytics insights.
Then develop your queries methodically, taking the time to solicit and integrate input from relevant stakeholders. The quality of your speech library will make or break your speech analytics initiative.
If your business issue is complex, such as predicting churn or improving sales methods, spend whatever time is necessary to build complex rules to identify patterns that will drive improvements. Be prepared to spend months building complex projects, starting with building a hypothesis and eventually testing against control groups.
9. Spend time listening to recordings of actual calls. If you forge ahead on a speech analytics project without listening to recordings reflective of the business issues you’re targeting, you lower your odds of producing useful business intelligence significantly. Guessing about “the voice of the customer” is not a suitable substitute for actually hearing it. Reviewing hundreds of agent-customer conversations is a laborious task, but it’s the only way to identify the meaningful key phrases that determine customer intent and predict outcomes.
10. Focus attention on both sides of the agent-customer conversation. Focusing on what agents are telling customers ensures that agents are following scripts correctly, communicating necessary information and avoiding prohibited words and phrases. The customer side of conversations provides equal or greater value by pinpointing the reasons behind customer complaints and identifying quality gaps in products, services, internal processes and agent-customer interactions.
11. Establish benchmarks by performing a speech analytics content audit. Listen to a sizable sample (typically 300 to 500) of random recordings from the appropriate business queue to determine how often a specific phrase or group of phrases occur. Gaining a sense of how often particular business issues surface and how often related phrases are uttered provides a benchmark from which to start tuning the speech engine.
Kicking off a speech analytics project without first conducting a content audit can lead to second-guessing the results. Without a real-world benchmark established by listening to hundreds of actual calls, you cannot be confident that, say, detecting a certain phrase in one out of every 10 calls is an accurate assessment of how often that phrase actually occurs.
Your carefully selected and expertly deployed speech analytics solution will allow you to access your call data by turning unstructured phone call data into searchable data that can reveal helpful analytic insights for multiple departments.
For instance, data harvested from your voice interactions can be used to improve operations in a number of ways. It can help you create insights that can improve the customer experience, improve the efficiency of agents, identify and manage key risks, identify opportunities to reduce call volume, and when coupled with other data can improve existing predictive models and create new models.
What’s more, you can use your analytic capabilities to assist in making “near real-time” decisions in the contact center space to improve the customer experience, prevent escalation of calls and ensure adherence to regulatory and legal requirements.
It won’t be long before contact centers that are not using speech analytics will be the exception rather than the rule. If you’d like to gain a competitive advantage, the time to invest in speech is now.
Scott Bakken, Founder and President of MainTrax, is a highly respected independent voice in the speech analytics industry.
– Reprinted with permission from Contact Center Pipeline, http://www.contactcenterpipeline.com