In the quest for superior customer experience, businesses traditionally rely on direct feedback channels like surveys, feedback forms, and product reviews. While invaluable, these methods often present a curated view of customer sentiment. The true goldmine of customer insights lies in the vast, unstructured world of indirect customer feedback – the unsolicited data generated through everyday interactions.
Artificial Intelligence (AI) is revolutionising our ability to understand these unspoken truths, the challenges involved, and the transformative business outcomes that emerge. Michael Michaloudis, Regional Director ANZ & ASEAN, Medallia, comments, “AI-powered text and speech analytics are no longer optional. They are crucial for processing the vast amounts of unstructured data that reveal true customer sentiment, hidden insights, and emerging trends which manual analysis would invariably miss.”
He adds, “Notably, unstructured data provides a vast treasure trove of experience data that a survey program alone can’t come close to capturing. The crucial step then becomes the skillful synthesis of these diverse data points, enabling us to pinpoint and address the most critical issues that will deliver the greatest positive impact for the business and the customer.”
The hidden landscape: Challenges of indirect vs direct feedback
Direct feedback, by its nature, is explicit. Customers answer specific questions, providing structured, quantifiable data that is relatively straightforward to analyse. In contrast, indirect feedback is often spontaneous, unprompted, and found in diverse, unstructured formats.Dave Flanagan, CX Strategy and Innovation Lead, Nexon Asia Pacific, says, “Indirect feedback — such as what customers say to agents on a call, how they behave on digital channels, or when they abandon an online journey — is often the most honest source of truth, but also the hardest to analyse at scale. Unlike survey data, it’s unstructured and embedded in transcripts, chat logs, and behavioural signals”.
Organisations and their VoC (Voice-of Customer) teams face a range of key challenges in collating and analysing this data. First and foremost is the sheer volume and complexity of the data. Lisa Khatri, Head of Customer Experience GTM and Strategy, Qualtrics, comments, “The biggest hurdle for most organisations is the sheer volume of unsolicited feedback. Think about it: calls, chats, and agent notes pour in at a rate of hundreds of thousands, dwarfing the feedback you’d get from surveys or social media. This data comes to you because customers need to contact you, not because you asked for it. That overwhelming volume is precisely where most companies struggle”.
Unlike survey responses with predefined answer options, indirect feedback is free-form text or even speech. Extracting meaningful data requires sophisticated linguistic analysis. Human language is complex, rife with sarcasm, slang, cultural references, and contextual subtleties. Social media, in particular, contains a lot of irrelevant chatter. Filtering out noise to identify genuine customer feedback is crucial.
Flanagan says, “Indirect feedback is spread across channels and often cluttered with “noise”, like small talk, repeated questions, or vague statements. Making sense of it requires strong data foundations and the right AI models to identify meaningful patterns and trends that drive business impact”.
Indirect feedback resides in disparate systems (CRM, contact centre software, social media platforms), making unified analysis difficult. Each team within the organisation, from the social media department to the customer service desk, often operates within its own data sphere, making it incredibly difficult to create a holistic view of the customer journey.
Khatri observes, “A significant hurdle for many organisations, particularly Voice of Customer (VoC) teams, is obtaining data from internal stakeholders who control contact centres or complaints departments. This is often due to internal politics and channel silos. However, there’s a growing global and increasingly Australian trend towards integrating all feedback sources, though this integration takes time and requires addressing security and privacy concerns. Many VoC professionals are not well-versed in cybersecurity implications, often leading to data integration being deemed ‘too hard’.”
AI to the rescue: techniques for analysing indirect feedback
AI is uniquely positioned to overcome the complex challenges of analysing vast amounts of indirect customer feedback. Indirect feedback comes in unstructured forms. Humans struggle to manually sift through and synthesise this volume, but AI, particularly NLP, excels at extracting sentiment, themes, and entities from such data. Different types of indirect feedback require different AI techniques to be applied
Salman Shahid, Chief Data Scientist, Asia Pacific & Japan Consulting Services, Teradata, says, “In general, different types of customer feedback require tailored AI techniques based on their structure and intent. Task specific language models were developed that targeted specific NLP problems”.
“For example, with social media posts, which are unstructured and emotionally charged, techniques like sentiment analysis, topic modelling, named entity recognition, emotion detection, and trend detection via time-series clustering are most effective and have been used extensively. These are still applicable even if the availability of LLMs has meant we don’t require as fine-tuning or training.”
Flanagan elaborates, “The right AI approach depends on the type of feedback we are analysing. For call transcripts and agent notes, large language models are ideal. They understand context, summarise conversations, extract key issues, and surface intent, even when it’s not explicitly stated during that conversation”.
“For social media, a mix of sentiment and emotion analysis can help track tone and intent in real time. To analyse website behaviour, machine learning and behavioural analytics are valuable tools. They help identify friction points, uncover patterns, and predict user actions such as abandonment or escalation points”.
Natural Language Processing (NLP) is the foundational AI technique for understanding human language. “Using natural language processing (NLP), sentiment analysis, and machine learning, we can interpret conversations and behaviours to uncover pain points, intent mismatches, and general friction in the customer journey”, says Flanagan.
NLP allows for:
- Sentiment analysis: Identifies the emotional tone (positive, negative, neutral) of text. Crucial for social media posts, reviews, and agent notes. Advanced sentiment analysis can even detect specific emotions like anger, frustration, or joy.
- Topic modelling: Discovers hidden thematic structures in large collections of text. This helps group similar feedback points, such as recurring product issues or frequently asked questions in support interactions.
- Named entity recognition (NER): Identifies and categorises key information like product names, company names, locations, and customer names within the text, providing crucial context.
- Keyword extraction: Pulls out the most important terms and phrases from a body of text, highlighting key discussion points.
- Intent recognition: Determines the underlying purpose of a customer’s communication (e.g., complaint, query, suggestion, purchase intent). This is particularly valuable for support interactions.
Machine Learning (ML) techniques are crucial for understanding customer touchpoints. Clustering and classification algorithms can group similar customer behaviours or feedback patterns, which helps identify distinct customer segments or common issues. Through predictive analytics, ML models can analyse historical data to forecast future customer behaviour, such as predicting churn risk based on interaction patterns or the likelihood of a specific product purchase based on website activity. Additionally, anomaly detection identifies unusual patterns or outliers in data, potentially indicating emerging problems or even fraudulent activity.
Shahid comments, “We can now develop tools that an LLM (Large Language Model)-based AI agent can use to apply relevant machine learning techniques to data. This allows the agent to generate insights that are quickly visualised. The agent can then swiftly pinpoint friction points by following instructions provided by a business analyst through a conversational interface. We’re witnessing the emergence of versatile, multi-purpose agents whose capabilities are primarily shaped by the tools they can access or are permitted to use”.
Real-world Impact: AI-driven success stories
AI’s ability to analyse indirect customer feedback is translating into tangible business benefits. Kahtri provides an example of how a major US financial services firm leveraged AI to analyse call centre interactions. By unifying call data with survey feedback, they reduced the time to act on customer issues from days to hours. This real-time insight allows for automated closed-loop feedback and proactive engagement, leading to significantly improved customer satisfaction. She says, “Previously, a survey sent two days after a call, with a two-day response time, meant a four-day delay in addressing customer issues. Now, with integrated analysis, they can act within hours, enabling automated closed-loop feedback for a larger customer base”.
Many contact centres traditionally review only a small fraction (3-5%) of calls for quality assurance. With AI-powered speech analytics, companies can analyse 100% of calls. This enables objective coaching, identifies agents needing support, ensures compliance with regulations, and reduces regulatory risk by eliminating bias in manual review.
Shahid provides an example of a global automaker used AI to better service to customers and identify upsell and cross-sell. He comments, “By using natural language understanding, reasoning, and retrieval-augmented generation (RAG), AI is improving their IVR systems to better resolve customer queries and enable upselling through propensity models. This also helps human agents by providing faster information retrieval, which cuts down the average time spent handling calls. This solution is expected to save the company $28 million annually in operational costs and generate an additional $390 million in revenue through AI-driven upselling and cross-selling, based on 118 million yearly calls”.
Companies are using AI to analyse product reviews and support tickets to identify common bugs, feature requests, and areas of dissatisfaction. By consolidating hundreds of thousands of feedback points into actionable reports, product teams can prioritise development efforts, leading to more user-centric products and features.
Mapping the Journey: AI for friction point identification
AI is particularly adept at understanding the customer journey and pinpointing areas of friction. “Understanding friction in the customer journey — whether online or in conversation — has traditionally been time-consuming and difficult to scale. AI changes that. It connects the dots across digital behaviour, call transcripts, and support interactions to uncover patterns that would otherwise be missed if done manually”, says Flanagan.
By analysing data from multiple touchpoints, AI creates a holistic view that wouldn’t be apparent from isolated data sets. Kahtri observes, AI models can flag unusual behaviour patterns on a website or app that might indicate a customer is struggling. For example, repeated clicks on a non-interactive element, multiple attempts to navigate a certain section, or sudden abandonment of a checkout process.
Shahid adds, “To pinpoint friction, anomaly detection identifies unusual behaviours such as repeated clicks or long dwell times, sentiment and emotion analysis reveals distress in support interactions, topic modelling uncovers recurring complaints tied to journey stages, and AI-enhanced heatmaps or session replays show where users hesitate, rage-click, or abandon sessions. These insights enable businesses to proactively optimise digital experiences”.
By tracking sentiment across a series of interactions (e.g., website visit, chatbot interaction, support call), AI can identify where customer frustration escalates. A negative sentiment in a chat that leads to a call can highlight a failing self-service option. AI can correlate specific online behaviours (e.g., viewing a product page multiple times) with subsequent support interactions (e.g., a call about product features). This helps identify information gaps on the website or common pre-purchase questions.
AI can anticipate future customer needs and potential friction points before they occur. By analysing patterns in support tickets and customer notes, AI can identify the underlying causes of recurring problems, whether it’s a technical bug, unclear instructions, or an agent training gap. If a customer exhibits certain browsing behaviours, AI might flag them as likely to encounter a specific issue, allowing for proactive outreach.
Shahid says, “AI enhances operational efficiency by streamlining processes and resource allocation, while real-time insights provide a 360° view of customer interactions without the need for predefined journeys. Behavioural clustering groups users by similar navigation patterns to distinguish successful from unsuccessful journeys. Predictive modelling helps anticipate outcomes, with churn prediction flagging users likely to abandon and conversion models estimating the likelihood of completing goals like purchases”.
Michaloudis adds, “Leveraging AI allows us to move beyond reactive analysis. Its predictive and prescriptive capabilities act as an ‘early warning system,’ detecting patterns to forecast future customer behaviours like churn risk and proactively identifying problems before they escalate.”
What Australian businesses need to do
Shahid highlights how Australian organisations and contact centres are making significant strides in using AI to analyse customer feedback. He says, “Almost all the customers we’ve been working with are deploying or looking to deploy Generative AI methodologies. They’re keen to better extract, consolidate, and analyse feedback from customer interactions. This also involves substantial workflow automation through Agentic AI wherever possible, aiming to significantly boost productivity and reduce technical debt”.
Popular use cases include conversational AI for routine inquiries, sentiment analysis for unstructured feedback, AI-driven quality assurance, and voice analytics for gleaning insights from call recordings. “Despite this progress”, says Shahid, “many organisations are still in the early to mid-stages of AI maturity. They’re grappling with challenges like data silos, integration complexity, and a shortage of skilled AI talent. Establishing the necessary infrastructure and information flows for production-level deployment, while adhering to regulatory constraints, remains a hurdle”.
Australian enterprises seeking to leverage AI for analysing indirect customer feedback should focus on a number of key areas. Michaloudis advises, “”The real challenge lies in integrating and consolidating customer interaction data across disparate channels to create a unified view. Once achieved, AI-driven insights must be democratised, made accessible and actionable at every level, from the C-suite to the frontline, to drive real-time decision-making and continuous improvement.”
He adds, “Ultimately, AI initiatives in customer feedback analysis must be directly linked to tangible business outcomes. This includes not only enhanced customer satisfaction and loyalty but also significant financial impacts like operational cost savings and revenue growth.”
The era of relying solely on solicited customer feedback is evolving. AI provides the tools to listen to the “unspoken voice” of the customer, transforming vast amounts of indirect data into actionable insights. By embracing AI for comprehensive customer feedback analysis, businesses can not only understand what customers are saying, but also what they are doing and feeling, leading to more empathetic customer experiences, optimisd operations, and sustained growth in today’s competitive landscape.