FeelGPT & AppTone: AI and Voice Analysis Tools for Fraud Detection in the Insurance Industry

Whether you prefer your customers to speak to human agents to file their claims or you prefer to automate this process using an AI agent, you can now incorporate a fraud detection system directly into the process with minimal changes to your existing workflows. Emotion Logic’s AppTone sends a carefully crafted questionnaire to the applicant, querying various topics related to the application. The risk analysis engine identifies issues that contain significant emotional distress and recurring risk cues and flags them for more in-depth investigation. Through a simple phone call, all incoming claims can be screened for potential fraud immediately after the session with FeelGPT.

Both systems are offered as SaaS services, with no on-site installation required. The entire process is fully automated, from the enquiry stage through to reporting. Both the AppTone and FeelGPT voice verification technologies also utilise advanced AI to associate words and emotions into coherent and meaningful reports that can be adapted to new fraudulent tactics across all lines of insurance.

Just a few days ago, Emotion Logic won first place in InsurTech Israel’s startup competition. The Insurtech was selected to participate for Israel in the Global Rapid Accelerator Programme, which will take place from 11 to 14 November in Des Moines, Iowa (USA).

Interview with Amir Liberman, CEO of Emotion Logic, and Mauro Nadav, Fraud Solutions Expert about AI Tools and Fraud Solutions.

In a recent report from Cifas, a leading fraud prevention organization in the UK, a worrying increase in so-called “first-party fraud” or “crimes of everyday life” was highlighted, such as inflated insurance claims and chargeback fraud. One in eight consumers admitted to having committed some form of fraud in the past year. This is where EmotionLogic’s solutions come in. How so?

Amir Liberman: Lying to your insurer is easy. The current fraud detection systems are tuned to known criminal profiles or what I would call highly biased criteria of risk. But if you are from the „right“ color, age, or side of the city, what would stop you from exaggerating your claim? What would stop you from demanding a chargeback on services you have consumed? The cost and risk for the claimant is zero, even if caught red-handed. Emotion Logic’s technology combines very sensitive voice analysis capable of detecting many subtle cues of unexpressed emotional responses, as well as signs of “risk”—the feeling we all have when we know we are doing something wrong or forbidden.

Our technology is not swayed by your expressed confidence or your skin color; it analyzes everyone’s voices with the exact same level of sensitivity, looking for the tells that exist in all human voices, regardless of socioeconomics or geographical location. If you feel at risk when you tell us your story, we will know.

How does AppTone work via a messaging app?

Mauro Nadav: AppTone integrates with messaging apps, primarily WhatsApp, used by insurers to communicate with clients. Here’s how it works:

1. When a policyholder files a claim, the insurer sends a specific 12-question survey via WhatsApp.
2. The policyholder responds with voice messages.
3. AppTone utilizes advanced voice analysis technology to extract vocal biomarkers, which are then integrated with transcripts generated through speech-to-text (STT). The combined emotional analysis and transcript are subsequently processed by AI to draw insightful conclusions.
4. AppTone generates a detailed risk assessment report within seconds of completion.
5. This report is then made available to the insurer’s claims team for quick, objective evaluation.

AppTone is designed specifically for analyzing audio responses sent through messaging apps as part of the claims process. The system prioritizes privacy and complies with data protection regulations.

And on a phone?

Amir: For those who prefer the human touch and the whole claim submission process handled by live call center agents, FeelGPT offers a quick and accurate way to automatically post-process all these recordings and provide the same insights as AppTone. FeelGPT is a post-call processor that, like AppTone, combines speech-to-text with our voice analysis technology and modern AI to create a context-aware emotion and risk assessment on the recorded audio, crafting easily understood and highly detailed reports. These reports not only assess the risk but also follow the principles of responsible AI by showing exactly how the decision was reached, allowing a human manager to review and make their own assessment if needed.

With AppTone, you say the future of fraud detection is here. Explain this.

Mauro: AppTone represents the future of fraud detection in insurance by introducing innovative approaches that enhance the assessment process. By analyzing vocal biomarkers, AppTone brings a new dimension to fraud detection – the ability to interpret emotional states and stress levels in claimants‘ voices. This emotional intelligence in risk assessment provides insurers with deeper insights into potential fraudulent activities, offering a more nuanced understanding of each claim.

AppTone’s ability to detect fraud on a case-by-case basis, even when facing new and previously unseen types of fraud, empowers human claim managers to identify and address these emerging threats quickly. This proactive support helps insurance companies stay ahead of evolving fraudulent activities and adapt their strategies effectively.

AppTone’s voice analysis provides a layer of insight that goes beyond conventional methods, offering a more comprehensive assessment of risk. By focusing on the present situation and the individual claimant’s responses, AppTone can capture the full context of a claim in real-time. This approach allows for a more accurate evaluation of each case, potentially reducing false positives and improving the overall efficiency of the claims process.

These capabilities not only enhance fraud detection but also have the potential to reshape the entire claims process. By providing more accurate and timely risk assessments, AppTone can help make the claims process more efficient and fair.

This innovative approach to fraud detection represents a significant step forward in balancing the need for thorough risk assessment with improved customer experience in the insurance industry.

What is the difference between AppTone and traditional Scoring Solutions?

Mauro: AppTone and traditional Scoring Solutions differ fundamentally in their approach to fraud detection. Traditional Scoring Solutions rely on historical data and statistical models, using past patterns to predict future behavior. These methods apply pre-determined criteria based on aggregate data to assess risk.

In contrast, AppTone analyzes the claimant’s current responses, providing a real-time assessment of the present situation. This focus on the individual claim, rather than historical patterns, allows for a more relevant evaluation. AppTone’s use of vocal biomarkers adds an emotional context to the analysis, offering insights not available through traditional scoring methods.

The adaptive nature of AppTone’s AI allows it to evolve more quickly in response to new fraud patterns compared to static scoring models, which typically require periodic updates. This flexibility, combined with its present-focused approach, can reduce the likelihood of unnecessarily investigating legitimate claims.

By providing a current, individualized analysis, AppTone complements traditional methods, enhancing the overall fraud detection process. It offers insurers a tool that can adapt to the dynamic nature of fraud while maintaining a focus on customer satisfaction by reducing false positives.

How can specific words and speech patterns indicate different types and levels of risks?

Amir: By applying the context-awareness capability of modern AI systems, we can carefully assess whether any specific emotional reaction is expected in that situation. For example, is it okay for me to be uncertain about the street where I had the accident? I think you may be unsure—I know I have a terrible memory for street names—but is it okay for you to be unsure about who was driving your car at the time you say you were? I think not.

Can you give us an example of how to use AppTone?

Amir: I’d be delighted to! Why don’t your readers try our demo test? It is a light 4 questions questionniare about fraud and technology. Scan this QR code to start a WhatsApp session, or use this link to take the browser test.

What data can be obtained with the risk assessment system for insurance companies?

Amir: The data we offer is very rich and can include personality traits, case-specific loopholes where the client reacts strongly to certain aspects of the case, or sensitivities around places, names, or events—all of which can guide the investigation. This comprehensive data allows insurers to focus on the most relevant areas, streamlining the investigative process and improving the accuracy of risk assessments.

Why is it important that AppTone not only identifies risks but shows exactly where they are in a conversation?

Mauro: AppTone helps us identify specific risks in conversations accurately. We pinpoint the exact words that signal risk, which helps us avoid mistakes and protect legitimate claims. It also allows our analysts to focus on the crucial parts of a conversation, speeding up their review process and boosting productivity. We provide clear guidance for investigators, making their work more straightforward and effective. This approach saves time and money for insurance companies and ensures that our decisions are both confident and fair. By making the claims assessment process precise and data-driven, AppTone not only improves our fraud detection but also enhances the overall efficiency and fairness of how we handle claims.

Exclusive biomarkers are being used for AppTone. How long did it take for them to be developed and refined, and in what respect does the technology offer a significant advantage in the market?

Amir: Emotion Logic is a young startup, soon to celebrate its third anniversary, but the technology behind it was developed over the last 27 years by my team and me, originally for security needs in Israel. All the biomarkers we discovered are proprietary to us and are very different from those normally used for voice analysis. We focus on a much deeper level than what is expressed and pronounced. We’ve been constantly working with academia and investigative bodies worldwide to validate our findings both academically and in real-life settings. I think this ability to detect subtle, unexpressed emotions is what makes our technology unique and the most capable in this field, known as EDR (Emotion Detection and Recognition).

Does the program comply with GDPR regulations?

Amir: We place great emphasis on privacy, fairness, and proper disclosure. We do our utmost to keep all data collected relevant to the case and not personal, separating identifying information as much as possible. All data is ideally removed from our system the moment the analysis is completed and delivered. Insurers have a valid need to assess the honesty of claims submitted, as they are entrusted by their clients to manage their money for legitimate claims. Validation of claims using our technology is far less intrusive and completely unbiased compared to other systems. The final responsibility for compliance lies with the insurer to properly notify their clients about the process. Recent statistics show that refusals to take the test were less than 1%, and people can simply refuse to take the test, which will just mean they have to take the longer investigative route, but that’s okay.

What are the savings that your insurance clients have reported so far?

Mauro: While respecting client confidentiality, we can share significant results from a major insurance company’s implementation of AppTone. We see potential for incredible savings in two key areas, based on a case study recently completed by one of our partners:

In this study, we applied AppTone to claims that had already been identified as HIGH RISK by another system. The results showed that only 52% of these cases were indeed high risk, while 13% contained some inaccuracies. After completing all investigations, about 70% of the cases flagged by AppTone as high risk were rejected, and the remaining 30% were paid due to a lack of conclusive proof. Furthermore, 75% of the cases initially flagged as high risk by the other system, but marked as low risk by AppTone, were eventually paid without issue.

These findings highlight the significant potential savings for insurers. By refining the initial risk assessment with AppTone, investigations can be more accurately focused on truly high-risk claims, reducing unnecessary investigations on low-risk claims. This not only minimizes the risk of fraudulent payouts but also optimizes the use of investigative resources, allowing insurers to allocate their efforts where they are most needed, thereby maximizing efficiency and cost-effectiveness.

What percentage of your analyses are actually correct?

Amir: In validated tests, our core technology showed amazing capabilities, far above the 95% markers. In insurance settings, due to the nature of the industry, it is hard to tell because not all cases we identify are investigated, and not all cases we recommend against investigating are verified. However, as you could see from the answer above, the savings potential and the ROI we can demonstrate are staggering, and the accuracy we can offer, after real-life validation, is second to none.

Read also: Nemesysco: How 007 knows if you are lying