Topic
Date & Time
Speakers
Summary
Coffee with Callbi #17 explored how speech analytics and AI are evolving beyond traditional QA to turn customer conversations into actionable business intelligence and measurable business value.
Overview
The 17th edition of Coffee with Callbi brought together the growing Callbi user community for a wide-ranging discussion covering speech analytics, Artificial Intelligence, Quality Assurance, agent performance, customer experience, operational efficiency and the increasingly important question of how organisations demonstrate measurable business value from analytics.
Hosted by Rod Jones and Corey Springett, the session also provided an update on the forthcoming AI-enhanced capabilities within Callbi and explored an important underlying theme: speech analytics is rapidly evolving beyond its traditional role as a Quality Assurance tool. Increasingly, the real opportunity lies in turning customer conversations into actionable business intelligence.
One of the strongest themes emerging from the discussion was the changing role of Quality Assurance.
Traditional QA typically involves listening to a relatively small sample of interactions and evaluating agent performance against predetermined criteria. Speech analytics fundamentally changes this model by allowing organisations to analyse potentially 100% of customer interactions and identify patterns that would simply not be visible through conventional sampling.
Corey emphasised that the objective is not to replace QA professionals, but to make them substantially more effective. Analytics provides greater coverage and far more granular insight into individual agent behaviour, allowing coaching to become more targeted and evidence-based.
An interesting consequence is already becoming apparent. Some experienced Callbi users are beginning to describe themselves as Insights Analysts rather than Quality Assessors. Rod related the experience of one Callbi power user who observed that, as a QA, operational management sometimes listened to her recommendations. As an Insights Analyst, however, “the boardroom listens to what I have to say.”
That distinction is significant.
The information contained within customer conversations is relevant far beyond the contact centre. Marketing can identify customer attitudes towards products and competitors. Sales can understand objections and buying behaviour. Operations can identify broken processes and repeat contacts. Risk and compliance teams can monitor regulatory exposure. Executive management can gain direct insight into what customers are actually experiencing.
The contact centre therefore becomes not simply a service operation, but potentially one of the richest sources of business intelligence within the organisation.
The session provided an important update on Callbi’s forthcoming AI add-on, which was undergoing final refinement following User Acceptance Testing.
Two major capabilities were highlighted.
The first is AI-generated intelligent call summaries. These will provide an English-language summary of individual interactions together with structured information including call driver, escalation, sentiment and resolution. The resulting information can also become dashboard metadata and potentially be exported into other systems such as CRM platforms.
The second capability introduces Large Language Model prompt-based analysis.
This opens an important new dimension. Traditional queries remain extremely powerful where organisations know precisely what they are looking for. Keywords and phrases can identify whether required statements were made, whether processes were followed or whether particular customer behaviours occurred.
LLMs become particularly useful when the question is more subjective.
Was the customer frustrated?
Did the agent demonstrate adequate product knowledge?
Why does the customer appear dissatisfied?
Was there something within the conversation suggesting that the interaction should have been escalated?
Rather than replacing Callbi’s existing query engine, AI therefore complements it. Corey stressed that the conventional query engine remains central because thousands of defined queries can be run across every interaction, while LLM prompting is particularly valuable for exploratory and subjective questions.
This distinction produced one of the most useful concepts of the session:
Measures such as NPS and CSAT can tell an organisation that customers are unhappy. They do not necessarily explain why.
Speech analytics can expose the underlying drivers: pricing, service failures, processes, products, waiting times, repeated contacts or other issues. AI-based contextual analysis potentially takes this further by allowing organisations to interrogate conversations through more open-ended questions.
The discussion also returned to one of the industry’s biggest current debates: will AI replace contact centre agents?
The consensus was considerably more nuanced.
Rather than asking whether AI will replace people, Corey suggested that organisations should ask which parts of the job AI can perform better, and which parts should remain human.
Routine verification, repetitive data gathering, basic administrative processes and some back-office tasks are obvious candidates for automation. Removing these activities could allow agents to concentrate on higher-value conversations requiring judgement, persuasion, problem-solving and empathy.
This aligns with observations from the recent Callbi webinar examining AI and jobs in CX and BPO. Rod noted that participating operators were not anticipating wholesale job losses. Their emphasis was instead on redeployment, upskilling and maintaining the “human in the loop.”
The message was clear: the strongest application of AI may not be replacing humans, but augmenting them.
Technology investment inevitably reaches the boardroom question:
What is the return on investment?
The discussion identified several areas where speech analytics can create measurable value.
Improving QA coverage from a small sample to potentially 100% of interactions provides substantially greater visibility without requiring equivalent increases in QA resources. Analytics can improve sales conversion by identifying objections and determining whether agents consistently communicate product benefits. In collections environments, similar techniques can identify behaviours associated with stronger payment outcomes.
Repeat contacts represent another significant opportunity. Understanding why customers repeatedly contact an organisation can expose failures in processes, service delivery or agent behaviour and create opportunities to improve First Contact Resolution.
Perhaps most importantly, analytics enables organisations to improve the performance of the people they already employ. Better insight creates better coaching. Better coaching changes behaviour. Better behaviour can improve sales, collections, customer experience and operational performance.
The Callbi ROI calculator was discussed as a means of quantifying some of these benefits, using factors such as agent numbers, interaction volumes, cost per seat, handle time and attrition. Importantly, Corey pointed out that such calculations primarily capture operational benefits and may substantially understate downstream value arising from improved sales, collections and customer outcomes.
See the Callbi ROI Calculator at >> Callbi | ROI Calculator – Callbi
Rod added another dimension: organisations frequently concentrate on cost reduction while failing to measure the financial value of customer retention, loyalty, repeat purchase and lifetime customer value.
This is precisely where listening systematically to the Voice of the Customer becomes strategically important.
The audience discussion moved into the practical realities of query building and highlighted an important lesson: creating the initial query is only the beginning.
Queries need to be tested against real conversations, validated and progressively matured.
Where calls fail to match, the question should be why. Agents may pronounce words differently. Transcription may interpret a phrase differently. Agents may use perfectly acceptable alternative wording that was not anticipated when the original query was created.
The solution is therefore iterative. Listen to the non-matching calls, identify acceptable variations and incorporate them into the query. Over time, the query becomes increasingly representative of how conversations actually occur.
This also explains why deterministic queries and AI-generated interpretation may occasionally produce different sentiment results. A purpose-built query searches for specifically defined indicators, while an AI model interprets the broader context of the interaction. Both approaches have value, but they answer subtly different questions.
Another valuable discussion addressed transcription accuracy, particularly within the complex South African contact centre environment.
Audio quality, background noise, pronunciation, microphone quality, accents, industry terminology, acronyms and brand names can all affect transcription.
South Africa adds another layer of complexity through multilingual conversations, vernacular language and code-switching.
Callbi’s approach has been to optimise its language models specifically around contact centre audio and continuously improve them through real-world customer data and feedback. The discussion cautioned against comparing headline transcription accuracy figures generated using clean broadcast or studio-quality audio with the far more demanding reality of contact centre conversations.
Context also matters enormously when dealing with South African languages. Word-for-word translation may fail to capture meaning where individual words carry different meanings depending upon how they are used. For this reason, contextual summarisation may ultimately provide more useful insight than literal translation.
One of the most strategically important points came towards the latter part of the session.
The most successful analytics implementations do not treat speech analytics as a standalone application.
Callbi-generated information can form part of the broader Business Intelligence ecosystem, feeding CRM systems, churn models, customer experience programmes and operational reporting.
For example, evidence of customer dissatisfaction within conversations could become an early-warning indicator within a churn model. Failure to follow a defined process could highlight potential downstream CRM or workflow problems.
This represents an important shift in thinking.
The value is not simply in analysing conversations. The value lies in what the organisation does with the intelligence those conversations reveal.
Perhaps the most important message from this Coffee with Callbi session was that the real value of speech analytics is no longer simply the ability to listen to more calls.
It is the ability to listen to the business through the conversations its customers and employees are already having every day.
When those conversations are systematically analysed, converted into insight and connected to operational and strategic decision-making, the contact centre becomes far more than a customer service function.
It becomes a source of organisational intelligence.
