4 Pillars of A High-Performing Speech Program

Recommended Reading

As collections agencies develop a more strategic approach to flexibility and scalability in the post-COVID economy, speech technology has become an integral part of their growth journey. To boost productivity without neglecting compliance, call center-based companies must look to advanced speech analytics (SA) platforms to provide a positive consumer experience while bolstering effective operations.  

The speech analytics market, which was once valued at US$ 1.6 billion in 2020 is now forecasted to reach US$ 9 billion by 2030 as per a study by Report Ocean – a compound annual growth rate (CAGR) of more than 20%. This growth can largely be attributed to the need for technologies that can automatically track trends and performance in contact centers.

Nowhere is this truer than the accounts receivable industry, where an increasing regulatory crunch is making it harder for agencies to run efficiently while upholding the latest Consumer Financial Protection Bureau (CFPB) and Reg F rules.

While speech analytics can be an incredibly powerful tool to solve new-age call center issues, we commonly see agencies run into the same problems as they launch their speech analytics programs. Many mistakenly believe that if they just plug-in SA into their workflows, the platform will immediately deliver insights that can transform their call centers on the fly. Unfortunately, it’s usually not that easy.

What Makes Speech Analytics Hard?

It is worth accepting that speech technology is not a perfect science. As we continue living in the age of AI, we expect the transcripts to be perfect, the data to be 100% correct, and the recordings to be spotless once we implement a new SA tool into our collections business. If you have those expectations, you need a reality check around using a conventional SA platform without best practices.

  • It is never easy to categorize calls without scanning every word within a conversation.
  • You don’t know if you’re tracking your agents effectively as their scorecards remain forever unchangeable.
  • Customizing dashboards and reports is hard because speech to text isn’t always perfect. Last-minute data pulls are hard as you never have the visualized data ready with you.
  • Sentiment analysis of calls isn’t always accurate because you are rarely able to understand the context and then categorize the portion of the call into meaningful data for your reports or dashboards.
  • Poor audio quality and the inability to integrate your SA to the recorder hold you back from tapping the full potential of your contact center.
  • You don’t always have the budget to hire a speech domain expert who can handle all aspects of a speech solution.

From everything mentioned above, it is clear that just investing in a speech solution isn’t enough to improve your collections yield and scale your business. It is not a solution that you can set and forget. It requires ongoing involvement and inputs by different stakeholders and needs to be navigated with a clear strategy and planning.

So, what do you do to avoid losing focus? Begin with clear initial goals and then dig deep into what we call the “Four Pillars of a Competitive, Foolproof Speech Program.” Need help? We have compiled this guide that should help you achieve everything mentioned above and more in a step-by-step, stress-free manner, one goal at a time.

1. Dynamic Agent Scorecards

In a typical call center setting, agent performance is continually reviewed. In a manual setting, managers are required to spend hours parsing through sample call recordings. It’s nearly impossible to cover 100% of calls, and potential issues or non-compliance can fall through the cracks. In absence of automated agent scorecards, there exists no formal system to capture performance issues objectively, benchmark them against industry standards, and make informed decisions about agent training and retention.

Interestingly, speech platforms can also enable you to immediately see agent performance trends, increase training speed, and even assess the effectiveness of your existing training programs. With effective scorecards aligned with but not limited to measures such as the average call times, silence analysis, or times of maximum activity, collections leaders can identify areas where agents can be coached to improve their performance. Looking at data in the form of scorecards, not only eliminates the need for manual performance reviews but also frees a manager of any confirmation bias they might have against an agent.

Small Tips That Make a Big Impact:

  • Switch to a SA platform that lets you create custom scorecards for each agent and project the same information to the agent panel to enable them to check their daily score and fix their errors. This way, agents will only see clutter-free information before starting their day for better focus, including their overall score trend for the month.
  • Score all calls. Scorecards based on 100% of calls rather than a small sample are much more accurate and support better training and learning output. By monitoring and scoring 100% of calls, the opportunity exists to connect analytics, quality assurance, and performance management, collecting information right down to the individual agent level.
  • Certain KPIs can always be part of your agent scorecards. That being said, choosing call center metrics for scorecards cannot be a one-time exercise. As per your current collections pattern, give more weightage to one metric that solves your biggest problem. For instance, if you spot an increase in call abandonments, focus more on the CSAT component of the scorecard vs the first call resolution (FCR) rate.

2. QA Intelligence

Traditionally, contact centers analyze a random sample of around 1% or less of calls. This method is rather self-selecting and not statistically representative. Speech analytics can provide 100% analysis across all calls and deliver a much more comprehensive, accurate view of what is happening in the contact center. While an analytics solution can be implemented to look at one particular pressing issue, such as compliance or automating the QA process, it can further develop into a tool for business intelligence, process optimization, consumers experience improvements and collections increase.

One of the real strengths of speech analytics from a QA perspective, the ability to automatically classify types of calls, might not seem very important. But being able to do this without asking QA personnel to sit for days listening to hundreds of calls is the difference between basing decisions on people’s opinions and gut feelings over basing them on hard verifiable facts. The automatic categorization of calls, based on the types of words and phrases that typically get used within collections calls is a starting point. By determining how many calls fit various categories, you can quickly keep a constant lookout for changes in consumer interactions.

The previous point leads us into a perfect segue into agent script compliance. With real-time prompts, SA can also make sure collectors say the right things throughout the conversation. It tracks each call and alerts collectors when they deviate from the pre-approved script. The benefit? You end up ensuring that all your agents are reading the mini-Miranda, respecting two-party consent rules, and following other obligatory requirements within the FDCPA and Reg F framework. If an agent skips an important keyword or phrase that is mandated by compliance norms, speech analytics flags it immediately. This way, agents get to correct themselves to avoid a compliance violation on the spot.

Small Tips That Make a Big Impact:

  • Check the ratio of classified and unclassified calls more often to determine whether you have covered the topics properly. If 30% of calls in a group are unclassified, the chances are the search terms don’t match many words in the calls, and other categories must be found.
  • Tap into transcription-based analytics (TBA), which will help you convert conversational speech to text. This way, you will be able to automatically create a text version of an unsuccessful phone call, allowing unlimited text searches when trying to determine the root cause of the issues. The context of words can be considered to avoid false positives. A text version of calls will make it possible to process data with other techniques such as data mining. This way, TBA will also automatically alert analysts to new topics for trend analysis.
  • Reflect on important questions such as “Why is the business witnessing a sudden spike in silence time?” or “Why the number of unclassified calls is suddenly increasing? If you are a smaller agency, you may struggle to find someone with enough time to investigate and classify calls, especially to get to the root cause of the issues. If you have the bandwidth to hire full-time QA analysts, hold on to it! If not, many SA vendors such as Provana can add great value to your operation by providing off-roll data analysts with years of experience in the credit and collections domain for more comprehensive trend analysis and customized reporting.

3. Real-time Call Analytics

While speech solutions have certainly been successfully used for recording 100% of phone conversations to allow call centers a safeguard against compliance risks, adherence to compliant script in real-time has become a genuine concern for many agencies lately. This explains why collections agencies might not just analyze voice recordings under their post-call speech analytics goals, but also dig deep into live calls with real-time SA to gather actionable insights on items like talk-offs, compliance, consumer preferences, and silent time.

Both types of SA can be used to analyze 100% of calls, rather than having a supervisor listen to a random call sample that may not be representative of agent performance and training gaps. Nowadays, speech analytics solutions such as ICAP can even sift through unstructured real-time call data to identify the probable causes of failure and success, helping you perform a complete analysis of consumer discontent and root-cause analysis. This gets you ahead of issues before they impact performance and profitability and help you report with confidence to demanding clients and compliance teams.

Based on the caller’s tone or modulation, real-time SA can help busy managers and QA teams identify training requirements at an agent level, providing immediate feedback to agents right after a flawed call.  AI-driven agent prompts in real-time, however, can be confusing and ineffective. By listening to live agent calls randomly, SA can help design effective training tools to help agents comply with regulations, deal with difficult conversations, and develop the knowledge and skills they need.

Small Tips That Make a Big Impact:

  • Continue enriching your real-time prompts to commonly flag compliance breaches and highlight words, phrases, and other verbal cues that could indicate an agent is faltering. This will help agents correct their mistakes most of the time before they move to the next call. Repeat the test-improve-test-improve the enrichment process as many times as possible.
  • Visualize real-time and post-call speech analytics as integral components of a wholesome speech program. While real-time speech analytics will enable you to alter the outcome of a call by intervening, post-call speech analytics will help design effective training programs.
  • Continuously measure all aspects of proper call handling and optimized consumer experience in real-time without having to rely on a tiny post-call sample. Also, find key training opportunities on compliance disclosures. This will help you ensure that those mistakes “one-off” issues or systemic are addressed by training.

 4. Top-Notch Sentiment Analysis

Agents, especially those with higher levels of empathy and interpersonal experience are mostly able to identify the emotions of the callers. However, the right agent-customer pairings are always a challenge that can otherwise result in significantly higher collections rates and better CSAT. By tracking agent performance across various personality types and EI scores optimal routing can indeed be a reality. And, sentiment analysis can help you achieve exactly that.

Many new-age SA tools such as ICAP come with real-time sentiment analysis that can evaluate the language used, as well as voice inflections, the rate of speech, and the amount of stress in the voice to decode the emotional outcome of ongoing interactions. It automatically mines calls for abusive language hence prompting agents to use a more upbeat, positive tone during the call. It is always helpful in assessing customer aims by examining the tone, pauses, and volume of speech.

Small Tips That Make a Big Impact:

  • Encourage call center managers to decode the language used by agents and customers, to understand their empathy profile, communication style, engagement levels, and transactional skills. Use insights gathered from historical calls to choose the agent whose skills and characteristics are most likely to achieve a positive response from the next consumer in the queue.
  • Track agent performance across various personality types. This information can be fed into the performance management process to help agents improve, and promote optimal routing.
  • Maintain behavior playbooks as per sentiment analysis trends. These playbooks with scorecards can help managers teach agents specific behaviors by outlining how to best demonstrate each behavior, showcasing best collections practice examples, and suggesting sample role-plays.
  • Instruct your QA team to dig deep into the sentiment analysis of audited calls. Glaring Emotional Intelligence (EI) gaps in agent-consumer interactions can be found in the form of condescending language or high rates of speech. Brainstorm a way to project the same EI gaps in visual formats for agents to see on their speech panels. This should set agents up for self-improvement or help them to be mindful of their tone before starting their day. Speak to your speech analytics solution provider to get scorecards amended for the EI component.


For the last few years, collections agencies have been using call center speech analytics to help reduce delinquencies, mitigate losses, and maximize their accounts receivable recovery. However, due to the lack of best practices and scarcity of up-to-the-speed solutions in the landscape, only a handful of agencies’ speech programs can quickly evolve as per customer preferences and lead to better collections yield.

Now, the question is what should you do if you realize the speech platform you implemented in your business isn’t perfect and limiting your potential to achieve all the best practices we talked about in this book? You certainly cannot bet on a solution that is marketed as a “business transformer,” but requires deep-seated change management for your agents and QA staff. What if you could have a solution that can easily be integrated into your system without having to affect your existing workflows and QA processes, calibrating change for one agent, one client at a time?

Power and Flexibility Out of the Box: Provana’s advanced speech analytics technology allows you to monitor consumer interactions with 100% coverage for regulatory compliance and increased collections volume. The platform not only mines, evaluates, and presents audio and transcribed data in rich formats, but also spots trends, coach agents on the consumer’s needs, and measures how well agents are fulfilling those needs. The SaaS-based SA platform offers scalability and cost-effectiveness — a sought-after combination in the current collections landscape.

Our solution is built on our depth and breadth of experience and knowledge across the collections and call center landscapes. At Provana, we have had more than 35 million call interactions, 500,000 client audits, and over 2 billion records that we’ve analyzed through our AI engine.

Want help with automating your QA process and all the benefits mentioned above? Book a demo here.