Introduction

The contact center has always been where fraudsters go when the website and mobile app get too hard. Now they have better tools. In October 2026, Mexican authorities announced the arrest of a group accused of using AI-cloned voices of judges, governors and presidential staff to pressure officials into paying money and handing out jobs and contracts, a scheme investigators say ran since 2021 (Arab News). These are allegations, but the pattern is familiar: a convincing voice, a believable story, and a person who wants to be helpful.

Industry surveys point the same way. A Pindrop survey of U.S. security leaders found that 74% had encountered or suspected a deepfake attack, while only about 10% had purpose-built defenses in place. Pindrop sells voice security, so treat the numbers with some caution, but the direction is clear. Market analysts tracked by Biometric Update expect combined face and voice deepfake detection revenue to roughly double, from about $3 billion this year to about $6 billion by 2028.

Voice biometrics is a strong defense for the contact center, and callers never see it working. It can confirm who is calling, compare suspicious callers to fraudster databases, and check whether the speech is human. No biometric is perfect, but catching most fraud that would otherwise go unnoticed can dramatically reduce losses.

Three Types of Contact Center Fraud

Account Takeover (ATO)

Fraudsters use stolen credentials or social engineering, such as guessing knowledge-based answers, to take over a real customer's account. Simple IVR questions are easy to defeat, and agents can be fooled by callers who know account numbers and personal details. Once an account is compromised, losses can mount quickly.

Duplicate Registration

Here the fraudster already has an account and successfully creates another under a different identity. It is common in retail and telesales, where "new customer" offers or device promotions create an incentive. A single fraudster collecting multiple phones at multiple addresses is a typical example.

Deepfake Fraud

Voice cloning, also called synthetic speech, can model a voice from seconds of audio, and it is no longer a novelty. Cloned voices are used to impersonate executives and officials, as in the Mexico case, and to call customers and contact centers directly. Pindrop has reported detecting AI agents placing calls to contact centers in its own customer traffic, a sign that automated callers are becoming part of the threat mix. For agents and callers who "know" each other, the effect is corrosive: a familiar voice is no longer proof.

Key Voice Biometric Functions

Enrollment

One or more speech samples are processed by deep learning models to create a voiceprint, which is typically encrypted and stored in a database. The IngenID One™ Platform can maintain an unlimited number of voiceprint databases, such as one for validated customers and one or more for known fraudsters.

Verification (1:1)

A caller claims an identity and supplies a speech sample, which is compared to the stored voiceprint for that identity. The resulting score is a confidence indicator that drives follow-up actions. It works in IVRs, conversational AI and live agent calls, and is usually the first line of defense.

Identification (1:N)

When there is no identity claim, or the claim is suspect, a speech sample is compared against a database of voiceprints to find a likely match. It is the key tool for spotting known fraudsters and duplicate registrations.

Spoof and Synthetic Speech Detection

Spoofing means passing off fraudulent speech as a real speaker: text-to-speech or voice cloning, recorded playback, voice conversion, or a skilled mimic. IngenID has developed its own spoof detection and its own speech synthesis tools. Below are two samples, one real and one synthetic. Can you tell the difference?

It is genuinely hard to tell, and we won't say which is which. Our detection can. Research on human listeners agrees: even with training, people identify synthetic voices only about 73% of the time (University College London study).

Detecting Fraud with the IngenID One™ Platform

Not every client needs every tool, and the platform is configured to each client's needs. Common configurations:

Account Takeover

• Verification - repeated failures within a set time period can indicate ATO activity and automatically lock the account.

• Identification - failed samples are compared to fraudster databases, and new fraudsters are added as they are found.

• Alerting - account locks and fraudster matches trigger alerts to monitoring consoles and email lists.

Duplicate Registration

• Enrollment + Identification - each new enrollment sample is compared against existing customer voiceprints, and strong matches are flagged.

• Assessment - a fraud team member reviews flagged cases. Confirmed duplicates move to the fraudster database and the registration is cancelled.

Deepfakes

Deepfake detection is a core part of the platform, and IngenID does not charge extra for it.

• Every sample checked - enrollment, verification and identification samples are all evaluated by our synthetic speech detection engine, in parallel with normal processing.

• Response code - the calling application receives a specific code so it can act, while consoles are alerted and logs updated.

Don't Rely on a Single Check

A voice match at the start of a call is a strong signal, but not the only one a determined attacker has to beat. The more resilient approach is layered: check continuously through the conversation, and combine voice with other signals such as call-back to a known number, step-up verification for high-value requests, and a second approver for large transactions. We explore this in Continuous Authentication and Deepfake Detection.

How Much Does It Cost?

There is no per-use pricing for our anti-fraud tools. Brute-force detection, playback attack detection and synthetic speech detection are included in the base cost of our services.

Identification (1:N matching) and fraudster database orchestration require a private hosted or premise deployment with the Anti-Fraud Tools add-on, and possibly the Streaming Media Adapter add-on (details here). Pricing for these is "hybrid": a fixed monthly fee with unlimited users and usage, in tiers based on aggregate volume.

Get Started

If fraud is a concern, contact IngenID. We will do a brief discovery to see how our tools fit your environment, what obstacles exist, and what fraud savings and ROI are realistic.