🚧 Feature is in Alpha and available on request
Alpha is the earliest stage of testing. Core functionality being tested with a very small number of highly engaged customers.
What that means for you
Things may break or behave unexpectedly
Features may change as we learn and iterate
Your feedback directly shapes what gets built next
To get access, contact support@voyc.ai or your Voyc Account Manager.
Fraud costs financial services firms real money and it often only comes to light after the loss has already happened. Deception Signals change that. Voyc listens to every conversation and surfaces the ones where deception patterns are building, pointing your team to the calls worth a closer look before it's too late to act on them.
What Are Deception Signals?
A Deception Signal is a targeted detector that listens for a specific pattern of behaviour known to appear in deceptive and fraudulent calls. When a call matches, the signal is raised and surfaces in your Signals queue.
The signals are built on research into real fraud cases in the vehicle finance and insurance sector. Voyc identified the patterns that appear repeatedly across confirmed fraud and built a detector for each one. The key finding from that research: fraud rarely announces itself in a single moment. The signals compound. One signal on a call is worth noting. Three signals on the same call is a meaningful fraud risk and that's the conversation to review first.
The Signals
Deception Signals cover three areas of risk within a conversation: customer behaviour, agent conduct and dealer activity. Each signal is built for specific industry contexts, which you can see on the signal card or when you open the signal detail. They run independently, so each one fires only when its particular pattern is present. The more signals a single call triggers, the stronger the case that something needs a closer look.
Signal Title | Description | Sector |
Dealer Deception: Poor Application Quality | Raised when a dealer's submitted applications are questioned for quality or integrity during a call. This typically surfaces when a broker challenges figures or details that don't hold up on review. Worth paying close attention to when it appears alongside other signals. |
|
Dealer Deception: Transaction Integrity | Raised when details submitted by a dealer don't match the broker's own records during a call. Where what was submitted and what's on file diverge, the conversation is flagged. |
|
Deceptive Behaviour: Avoiding Questions | Raised when a customer resists or challenges the verification questions being asked. Evasion, deflection and pushback on routine checks are the patterns it catches. |
|
Deceptive Behaviour: Incorrect Vehicle Details | Raised when a customer can't confirm basic details of the vehicle being financed. Someone who genuinely took delivery of a vehicle can usually describe it. Someone who can't is worth a closer look. |
|
Deceptive Behaviour: Third Party Prompting | Raised when a customer's delivery of basic personal details suggests they may be being prompted by someone else. The pattern is fragmented, hesitant responses where information arrives in pieces rather than naturally, often across date of birth, phone number and address. |
|
Deception: Hesitation | Raised when a customer hesitates while recalling personal details the policyholder should know. A genuine policyholder can usually recall their own details without pause. Significant hesitation when asked about basic policy information, medical history or identity details can indicate the customer doesn't hold that knowledge first-hand. |
|
Deception: Question Evasion | Raised when a customer avoids answering a question that bears on an underwriting or claims decision. Questions about health conditions, prior claims or risk factors are routine in insurance calls. When a customer consistently deflects or fails to answer questions that could affect the outcome, that pattern is worth investigating. |
|
Deception: Agent Prompting on Medical Questions | Raised when an agent steers a customer away from disclosing a medical condition. This signal flags potential misconduct on the agent side: if an agent's handling of medical questions appears to discourage full disclosure, it creates compliance and regulatory risk for the business. |
|
Deception: Contradiction | Raised when a customer's account of events conflicts with what they said earlier or with the record. Inconsistencies between statements within a call, or between what the customer says and what's on file, are a strong indicator that something in the account isn't accurate. |
|
Deception: Third Party Coaching | Raised when a customer's delivery of factual details suggests someone off the call is supplying them. When answers arrive with unnatural pacing or in fragments, in a way that suggests the customer is being fed information rather than recalling it, this signal fires. |
|
Setting Up Deception Signals
Signals are configured per Channel, so you can choose which ones run where and set up your team's notifications.
Navigate to the Channel you want to configure. Select Configurations from the side navigation, then select the Deception Signals tab.
Each signal card shows the signal name, the category and a short description of when it fires. Toggle on the ones relevant to your operation either directly on the card or by viewing the card.
To receive an email when a specific signal fires, click View & Add Notifications on the signal's card. Search by username, team name or email address and click + Add.
📝 Note: Users must be Channel members to receive signal notifications.
Viewing Your Signals (Signal Page)
Open Signals from the left navigation to see all signals raised across your Channel. Each Signal is linked to a Conversation and ordered by Signal Date.
The All Found Signals column is where triage starts. It shows how many different signals fired on the same conversation. One signal is a prompt to investigate. Three signals on the same call is a stronger fraud indicator, and that's the conversation to open first.
Each row also shows:
Signal Title: which signal fired
Agent: who handled the call
Conversation Date: when the call happened
Resolution: the current state of the signal
Each column is sortable so you can order the data in whichever way is suitable to your needs.
Resolution states
Each signal sits in one of four states:
Unseen: the signal has fired but the conversation hasn't been opened yet
Seen: you've opened it on the conversation, but haven't marked it yet
Valid: marked as a legitimate fraud indicator in the conversation
Invalid: marked as not suspicious on review of the conversation
Invalid signals are excluded from the list by default. To include them, open the Exclude dropdown in the top right and uncheck Invalid signals. Archived conversations work the same way.
Filtering
Use + Add filter to narrow the list by Signal Title, Agent, Conversation Date or Resolution. The default date range is the last 30 days. To change it, select the arrow on the Signal Date filter to pick a different range, or remove it and add a new Signal Date filter with a custom range.
Exporting Your Signal Data
Use the download button in the top right of the Signals page to export the current view as a CSV. The export includes: Signal Title, Agent, Date Triggered, Review Status, Notes, Date Reviewed, Reviewed by, Conversation URL and Attributes.
📝 Note: The CSV export reflects your current view based on the filters applied. It is also limited to the page you are on: if your filtered results span more than 250 signals across multiple pages, you will need to export each page separately.
Reviewing and Actioning a Signal
Triggered Deception Signals can be reviewed and action within your conversation, similarly to how you would an Alert.
Open the relevant conversation, either from your Channel Conversation List or via the Signals page.
You will find a section called Deception Signals. For each signal you'll see:
Signal Description: what the signal listens for
Reasoning: Voyc's explanation of why this specific call triggered it, in plain language
Confidence: how strong the match is based on the transcript
Evidence: the exact moment in the call that triggered it. Click to play the audio and jump to that point in the transcript.
Actioning Your Signal
Once you've reviewed a signal, mark it to record what you found:
Valid: the behaviour this signal looks for was found in the call. The signal fired correctly.
Invalid: on review, the behaviour this signal looks for was not found in the call. The signal fired, but the pattern wasn't there.
Your team can add a note to explain the decision. These assessments build a record that supports any action you need to take, whether that's escalating internally, informing a downstream process or coaching your team on what the patterns look like in real calls.
💡Tip: You may consider tagging certain conversations to create your own filter or dataset. Read more about how to use Tags our Tag Report.
The Gist
Fraud related behaviours in vehicle finance and insurance calls rarely announce themselves, but your conversations are one of your richest touchpoints of data. Deception Signals surface the ones where the patterns are building, giving your team what they need to prioritise investigations and dig further into applications and claims before a loss is confirmed. Activate the signals relevant to your business, triage by how many fired on the same call, and mark your outcomes to build a record you can act on.