Steps to Building a Successful Speech Analytics Program

By Claire Bakken, MainTrax

A common analogy to represent various aspects of culture and society is the physics of an iceberg.

Because 90% of an iceberg’s body is actually beneath the water surface level, we also get the phrase “tip of the iceberg,” meaning that what you see is only a small part of a much larger situation or problem that remains hidden. Just like seeing the tip of the iceberg may not initially seem as threatening, we commonly find that when companies begin deploying their speech analytics tool, they quickly realize there’s actually a lot more internally that needs to be considered for a successful program to be created. The truth is, utilizing a speech analytics tool is seldom as simple as the vendors say it will be and not as easy as the economic buyers think it is to manage.

During our experience in the industry for more than a decade, we’ve seen our share of victories and challenges that have contributed to a company’s successes and failures with their new speech analytics technology. Having a solid foundation and strategy set up is critical for an effective program. While it may be tempting to try a few shortcuts, it’s important to consider and plan for each step on our list of how to build a successful speech analytics program:

  1. Create a vision: Lack of vision means lack of organization which will eventually lead to either an ineffective speech program.  Or worse, the technology will collect dust.
    • What we recommend: Create mapping strategies, determine the most important KPIs and select high-value initial projects with intended outcomes. Additionally, create a culture of continued focus on the success of the tool.
  2. Line up resources: Knowing what’s needed and who’s “owning” different aspects of the tool’s use (from strategy to hands-on use) is important for a smooth process and ongoing success.
    • What we recommend: Design action plans and processes along with specific roles: who will do what with what. This includes acquiring or developing the right skilled talent. Along with hiring and training analysts, keep in mind the time it will require to build speech models and conduct iterations of testing and analysis.
  3. Understand the true voice of the customer: You may think you know what phrases are commonly spoken by your customers and agents (or what you wish they were saying), but you’ll never know unless you sit down and listen.
    • What we recommend: Perform a Content Audit to fully understand the voice of the customer, the voice of the agent. Besides helping you establish benchmarks of how often issues actually occur, a Content Audit will also identify factors such as audio issues and strong dialects that could affect the results.
  4. Build language models:  Knowing query building and analytics is important but understanding the tool’s features and functions will improve accuracy and insights.
    • What we recommend: Dedicate time to learn the speech tools features and functions. Experiment with combinations of these functions to build the most effective models.
  5. Capitalize on metadata: Important information like call direction, department, agent ID, and call duration can be pulled and utilized to narrow in on valuable calls or projects. A plethora of data is also available to you in the wrap-up codes of your recordings.
    • What we recommend: Incorporate available metadata and learn what data can and cannot be pulled from calls. Don’t wait. Make metadata part of your initial system integration or, if applicable, add it into your data lake.
  6. Analyze the patterns: Once your data is ingested into your speech tool, it’s time to begin your analysis.
    • What we recommend: Analyze results for patterns, develop hypotheses, test against control groups, and tune your models based on findings
  7. Validate the results: Once you’ve built models for specific outcomes, a thorough validation process needs to be conducted to ensure that the models built are producing accurate results.
    • What we recommend: Test and validate results to ensure insights align with outcomes.
  8. Visualize the results for executives: Many companies prefer to push their post-processed data into their BI visualization tools to display the data in their existing dashboards and reports.
    • What we recommend: Plan in advance to ensure you have the voice interaction templates required. Talk to the data scientists in advance
  9. Commit to continuous improvement: As initiatives, ideas, and expectations change, your speech analytics program will have to change, too. This ensures that you’re pulling the most accurate data for current business issues.
    • What we recommend: Establish a feedback loop for continuous improvement. Assign roles, create processes, and calibration schedules. This includes making adjustments based on organizational change and feedback from utilization strategies.
  10. Solicit stakeholder support: Those who have invested in the process of acquiring and deploying your speech analytics tool will want to be kept in the loop on status updates, progress and outcomes.
    • What we recommend: Share high-value insights and specific examples of interactions with stakeholders early and often in the process to maintain their commitment.

Which step in the process have you experienced roadblocks? If you’re considering a speech tool, what seems the most daunting?

About MainTrax
MainTrax is a leading provider of speech analytics professional services to end users and industry partners. Free of allegiance to any one solution or supplier, MainTrax has earned a reputation as an independent, unbiased resource for consulting expertise across a variety of products and providers.