Eliot Cohen Bacrie Lire en français

Six Problems. One Voice. A Billion-Dollar Blind Spot.

April 12, 2026 · 11 min read

How voice intelligence is becoming the missing layer in insurance, legal, security, HR, due diligence, and consumer protection.

On March 26, 2025, a finance director in Singapore joined a routine Zoom call with his company’s senior leadership. He recognized every face. He recognized every voice. He authorized a $499,000 wire transfer. None of those executives were real. Every face was a deepfake. Every voice was artificially generated. HSBC flagged the anomaly the next day. Singapore’s Anti-Scam Centre recovered the funds by March 28.

That was one call, one company, one afternoon. Multiply it across insurance claims, legal depositions, investor calls, job interviews, and customer support lines, and you start to see the gap. Voice touches every industry. Every industry trusts it. Almost none of them verify it.

I spent three years building ORAVYS to fill that gap. Today, the platform runs thousands of independent analysis engines on a single voice recording. It identifies synthetic speech. It maps stress, emotion, personality, cognitive load. It generates forensic reports where every finding traces back to peer-reviewed science. The technology is not the story here. The story is what it solves, industry by industry, and what that is worth in concrete terms.

This is what ORAVYS Group does. This is what we stand for.

What ORAVYS Group stands for

Three principles govern everything we build. First: the voice belongs to its owner. We never store voiceprints, never link recordings across sessions, never retrain on client data without explicit opt-in. GDPR-compliant from the first line of code. Audio processed and deleted by default.

Second: every result must be verifiable. When our system flags an anomaly, the report cites the published study that established that feature as an indicator. A fraud investigator, a judge, an insurance adjuster can trace each finding to its scientific source. No black boxes.

Third: probabilistic assessments, never certainties. We do not produce verdicts. We produce measurements with confidence intervals, documented error rates, and transparent methodology. The human making the decision stays in the loop.

The six verticals at a glance

Claims Detect (claimsdetect.com) covers insurance fraud at 13 billion euros/year in Europe. ROI: 30x+ first year (ClearSpeed documented).

Anti-Deepfake (anti-deepfake.com) covers CEO voice fraud at $25M per incident, +3,000% attacks in 2023. EU AI Act compliance: avoid 7.5M euro fine.

Voice Forensic (voiceforensic.com) replaces the forensic audio expert at $500–2,000/hr and 4–8 weeks. Same quality in minutes, $199–999 per report.

Voice DD (voicedd.com) addresses the fact that 75% of startups pitch with misleading info. $2,500 vs $50K-150K traditional due diligence.

ORAVYS Enterprise (app.oravys.com) tackles call centers that review only 2–5% of calls. 10% less turnover = $1M+ saved per 1,000 agents.

Voice Intuition (voiceintuition.com) addresses $25.4B lost to phone scams in 2023 (US alone). $14.90 per scan, protection for everyone.

Insurance: the thirteen-billion-euro wound

Insurance fraud costs European insurers an estimated 13 billion euros per year, according to Insurance Europe, the federation representing 91% of European premium income. That figure represents roughly 10% of all claims paid. It gets absorbed into premiums. Every honest policyholder pays for it.

Voice sits at the center of the problem. Claims are filed over the phone. Recorded statements are reviewed by adjusters who handle dozens of cases a day. They listen for inconsistencies, but human ears are imprecise instruments. They miss acoustic indicators of rehearsed speech. They cannot separate a genuine stress response from a performed one. And now, a threat that did not exist five years ago: synthetic voices filing claims on behalf of people who never called.

Voice-related fraud attacks increased by 475% in 2024 (Pindrop Voice Intelligence Report, 2025). Insurers designed their processes for a world where the person on the phone was at least a real person. That premise broke.

Claims Detect is our insurance vertical. It analyzes recorded claims calls across hundreds of dimensions, using detection accuracy validated in independent benchmarks. The output is a forensic report that an adjuster can attach to a case file, not a black-box score.

The economics: El Roble Seguros, the largest auto insurer in Guatemala, deployed ClearSpeed’s voice analytics and documented a 31x return on investment in its first year. ClearSpeed reports a 30x+ average ROI across its insurance clients, with a 35% increase in fraud loss savings per claim. An insurer processing three million minutes of claims calls annually, at $0.05 per minute, spends $150,000 on analysis. If that intercepts even a fraction of the 13 billion in annual fraud, the math does not need explanation.

Cybersecurity: the three-second threat

Three seconds of audio. A conference keynote, a podcast clip, a LinkedIn video. That is all it takes for tools like ElevenLabs or Resemble AI to produce a voice clone that carries the speaker’s intonation, rhythm, and accent. The $25 million Hong Kong fraud in January 2024 (confirmed by CNN and the World Economic Forum) used this method. So did the $243,000 CEO impersonation in 2019 (reported by the Wall Street Journal). So did the UK energy firm that lost 220,000 euros to a cloned executive voice in 2019.

The technique keeps reaching higher. In early 2025, fraudsters cloned the voice of Italy’s Defense Minister to call business leaders and request urgent transfers tied to a fake state operation. At least one payment of close to one million euros went through before anyone caught it.

Deepfake attacks against businesses surged 3,000% in 2023. Deloitte’s Center for Financial Services estimates that generative AI fraud losses will reach $40 billion by 2027, up from $12 billion in 2023. In a 2024 Deloitte survey, more than one in four executives said their organization had already faced at least one deepfake incident.

The regulatory clock is running. The EU AI Act, Article 50, mandates that AI-generated content be labeled and detectable. Enforcement of these transparency obligations starts August 2, 2026. Penalties reach up to 7.5 million euros or 1.5% of global annual turnover. The European Commission published its first draft Code of Practice on marking AI-generated content in December 2025. A final version is expected by June 2026.

Anti-Deepfake is our cybersecurity vertical. Enterprise-grade synthetic voice detection with thousands of engines running in parallel. When a call comes in, the system answers the question that legacy voiceprint matching was never built to ask: is this a real speaker, or a synthesis model?

For a CISO: one successful CEO fraud costs more than a decade of detection service. For a DPO facing the August 2 deadline, compliance is not optional. The alternative is a fine that could end the company.

Legal and forensics: the expert witness problem

Audio evidence is everywhere in legal proceedings. Depositions, witness statements, recorded threats, contested phone calls, wiretaps. The question of whether a recording is authentic used to surface occasionally. Now it surfaces in almost every case involving audio.

And the recording itself is now a target. A synthetic clip can be submitted as evidence, a confession that was never spoken, a threatening voicemail that never happened, an admission stitched into a real conversation. Once a fabricated recording enters a case file, it shapes an investigation or a verdict until someone can prove it was generated. That proof is exactly what a courtroom needs and what an untrained ear cannot give. This is the heart of why audio authenticity stopped being a niche question and became a procedural one.

The traditional answer is a forensic audio expert. Rates run $500 to $2,000 per hour. A full analysis takes four to eight weeks. Courts require Daubert-compliant testimony: testable methodology, known error rates, peer review, general acceptance in the field. Finding an expert who satisfies all four prongs, is available, and can testify before the trial date is increasingly hard.

Voice Forensic is our legal vertical. Every ORAVYS forensic report is built for court admissibility. Each metric cites the peer-reviewed study that validates it. Error rates are documented. Chain of custody is maintained. The methodology is testable and reproducible because every engine’s logic is transparent.

Cost: Traditional expert at $15,000+ per case vs. Voice Forensic at $199 to $2,500.

Turnaround: Traditional expert at 4 to 8 weeks vs. Voice Forensic in minutes.

Daubert compliance: Traditional expert depends on the individual. Voice Forensic has it built-in: citations, error rates, chain of custody.

Reproducibility: Traditional expert varies. Voice Forensic is deterministic: same input = same output.

For attorneys, this changes the economics of audio evidence. Cases that could not justify the cost of a forensic expert can now access the same level of analysis. Justice becomes less dependent on budget.

Due diligence: the voice in the pitch

CB Insights has documented that weak founding teams are among the top reasons startups fail (CB Insights, 2024).

The cases that make headlines are worse. Charlie Javice fabricated user numbers to sell her startup Frank to JPMorgan for $175 million. JPMorgan’s own due diligence missed it. Cluely’s CEO publicly admitted that the $7 million ARR he had been claiming was entirely fabricated. These are not outliers. They are symptoms of a system where due diligence often stops at the spreadsheet.

Synthetic voice raises the stakes again. A founder’s voice is easy to clone from a single podcast or demo, which means a fabricated voice update claiming traction the company does not have can land in an investor’s inbox as a voicemail or a WhatsApp note. The harder version is the reference call. Diligence teams still phone customers and advisors to confirm a story, and those calls are rarely verified for authenticity. A coordinated set of cloned “references”, a fake flagship customer, a fake advisor vouching for the team, can manufacture the social proof an investor relies on. The call sounds normal because acoustically it is convincing. The signal it carries is fabricated.

A voice carries information that a pitch deck cannot. Cognitive load appears in speech rate changes and pause patterns. Rehearsed answers produce different acoustic signatures than spontaneous ones. Acoustic markers validated by decades of psychoacoustic research show up in ways that published science by Scherer (2003), Tsanas et al. (2012), and Cummins et al. (2015) has documented and measured. None of this is mind-reading. It is measurement, grounded in published science.

Voice DD is our due diligence vertical. Investor-grade voice analysis of CEO calls, pitch recordings, board presentations, and the reference calls that back them. The output is a credibility profile mapping stress, confidence, cognitive load, and emotional congruence across the conversation, and a verdict on whether the voice itself is human or synthetic. Not a lie detector. A measurement instrument with documented methodology.

A due diligence pack costs $2,500. A typical M&A due diligence engagement runs $50,000 to $150,000. Adding voice analysis is a rounding error on the budget and a material addition to the signal.

HR and call centers: the 2% sampling problem

Call centers manually review 2 to 5% of their calls for quality assurance. The other 95% goes unexamined. At scale, that means millions of hours of voice data containing intelligence about agent performance, customer satisfaction, emotional escalation patterns, and fraud signals. All of it discarded.

Agent turnover in call centers runs 30 to 45% annually. Replacing a single agent costs $10,000 to $20,000 when accounting for recruitment, training, and ramp-up. Early identification of burnout, stress, or disengagement through voice patterns could reduce that attrition meaningfully.

On the hiring side: a 30-second voice sample contains more behavioral information than most companies extract from a 45-minute interview. AssessFirst, a French HR tech company with 15 million euros in annual revenue, already sells a voice-based assessment product. The market exists.

ORAVYS Enterprise handles HR and call center use cases. 100% call coverage with a broad set of behavioral signals extracted per recording. For call centers: every call is analyzed, not 2%. For HR: a structured, bias-reduced voice assessment that complements traditional interviews.

Call coverage: Manual QA at 2–5% sampled vs. ORAVYS Enterprise at 100% analyzed.

Turnaround: Manual QA in days to weeks vs. ORAVYS Enterprise in under 30 seconds.

Bias: Manual QA is reviewer-dependent. ORAVYS Enterprise is algorithmic and documented.

Burnout detection: Manual QA is reactive (after resignation). ORAVYS Enterprise is proactive (voice pattern shift).

If voice analysis reduces turnover by even 10% in a 1,000-agent center, the savings exceed $1 million per year. The analysis cost is a fraction of that.

Consumer protection: the call you should not have trusted

Americans lost $25.4 billion to phone scams in 2023 according to Truecaller’s U.S. Spam and Scam Report, with over 56 million adults victimized. The elderly are disproportionately targeted. A scammer clones a grandchild’s voice from a social media video and calls the grandparent: “I am in trouble, I need money.” The voice sounds right. The fear is real. The money is gone.

There is no consumer tool today that tells an ordinary person whether the voice on the other end of a call is real. Caller ID is trivially spoofed. A voice sounds the same whether it comes from a real speaker or a synthesis model.

Voice Intuition is our consumer vertical. Individual voice authenticity analysis at $14.90 per scan. Upload a recording, receive a forensic-grade assessment of whether the voice is human or synthetic, with emotional and stress analysis included. No expertise required. No subscription needed.

The value is not abstract. It is the difference between sending $5,000 to a scammer and keeping it. For a parent receiving a panicked call from a child’s cloned voice, it is the difference between the worst day of their life and a close call.

The platform behind all of it

Every vertical described above runs on the same foundation: the ORAVYS voice intelligence platform.

Independent analysis engines: thousands. Emotion dimensions: 40+. Clinical biomarkers: documented, peer-reviewed. Processing time (full analysis): under 30 seconds. Data retention: processed and deleted by default. Privacy: GDPR-compliant, Global Privacy Control. Infrastructure: EU-only (europe-west1).

Six verticals, one platform. Each is calibrated for a specific industry’s needs, its regulations, and the decisions its people make every day.

The holding company is ORAVYS Group. It coordinates the platform, the verticals, and the research arm that keeps the detection ahead of the attacks.

What this is worth

The deepfake detection market is projected to reach $7.27 billion by 2031, growing at 42.8% annually (MarketsandMarkets, 2025). Insurance fraud detection, voice biometrics, emotion AI, and HR tech analytics add billions more. The total addressable market for voice intelligence spans sectors that together represent tens of billions in annual spending.

The real proof of value is not a market projection. It is a fraud investigator who intercepts a synthetic claim that would have cost $200,000. It is an attorney who gets forensic audio analysis in hours instead of weeks. It is a CFO who does not wire $25 million to a fake executive. It is a grandmother who does not send her savings to a voice that sounded like her grandson.

That is what we built. That is what ORAVYS Group stands for.

Eliot Cohen Bacrie is the Founder & CEO of ORAVYS Group, an Israel-based voice intelligence company building forensic-grade audio analysis across six industry verticals.

ORAVYS Group: oravysgroup.com

Platform: app.oravys.com Anti-Deepfake: anti-deepfake.com Voice Forensic: voiceforensic.com Claims Detect: claimsdetect.com Voice DD: voicedd.com Voice Intuition: voiceintuition.com YouTube: youtube.com/@oravys

References

More writing