By 2032, Fortune Insights predicts the market for voice biometrics technology will be worth $15.69 billion. It’s a massively growing market, but the folks we’ve spoken to across industries and roles for the last few years have expressed frustration and dissatisfaction with their voiceprint implementations. Why?
Voiceprint solutions make a big bet: that account holders will opt in, enroll early, and trust you with their biometric data.
Spoiler: They don’t.
Even in high-trust industries like banking and insurance, enrollment rates are low and the cost of failing to identify a bad actor is high. That leaves contact center teams in an awkward spot: spending enterprise-level budget on a low adoption solution but still manually authenticating most callers.
From there, they have a choice to make. How much can they mitigate the low adoption with increased agent verification, push for customer adoption through resistance, or layer other solutions on top of what’s already implemented. In short, they’re balancing the risk of letting a bad actor through with reducing customer (and agent) frustration.
The Problems with Voiceprint Implementation
Voiceprint tech should help. But too often, it’s holding teams back. Most 'modern' implementations of voiceprint feel more like legacy solutions to an ever-evolving problem, built on outdated assumptions and requirements.
Let’s break down why.
1. It’s Expensive (and Slow) to Stand Up
Voiceprint systems often require months of setup, system overhauls, and tuning before you see any value.
Most traditional voiceprint implementations come from large vendors or are deployed through a professional services provider. So, you’re looking at months of contract negotiation, requirements gathering, and implementation standing between the kickoff meeting and handling the first call with the system, all while a team takes up your conference room and drinks your coffee.
The end result is a vendor-locked system that’s slow and difficult to change, even though you have the technical expertise on your team to support it.
Then what?
2. Nobody’s Enrolled
Voiceprints only work if customers enroll. But most don’t.
With only 30% of your customers enrolled on average, you’re spending big bucks and still manually authenticating the vast majority of your callers. That’s not just ineffective, it’s frustrating for both sides of your calls.
But why aren’t they enrolling?
There are numerous reasons depending on everything from the industry to the caller. Some customers may simply not have called since the system rolled out, others aren’t thinking about enrolling because they called for a specific reason and are trying to get that task done, others may not believe or understand the value of enrolling and trust always plays a part in enrollment.
3. Even When It Works, It Breaks Down
Let’s say you put in the time, educate your customers and your team, run a coordinated campaign, and you get decent enrollment.
Almost as soon as you deploy the system, impostors begin to enroll the wrong voiceprints for legitimate customers. The system flags real customers, adds frustration, and increases costs handling false positives. Contact center leaders we’ve spoken with know that they’re detecting fraudulent enrollments in voiceprint within a few months of deployment, and they know it’s not isolated but aren’t certain the exact scale or what accounts have been compromised.
What started as fraud prevention has turned into a huge source of friction in all directions.
But let’s take a step back, because voiceprint is intended to provide a very specific value, and that shouldn’t get lost in the shuffle.
The Intended Value of Voiceprint
The intention behind voiceprint is simple: use the customer’s voice as a unique identifier to allow for authentication as part of a normal conversation and thereby improving both account security and the customer experience.
That comes with a simple set of requirements:
- One solution for everyone
- Simplifies the experience for both sides of the call
- Keeps impostors out
That’s it, at the most basic level. Every vendor and provider adds their own spin, their own bells and whistles, and their own tradeoffs in implementation, but it’s all ultimately in service to a simple set of requirements that every voiceprint solution is trying to solve for.
If Voiceprint Were Invented Today, What Would It Look Like?
(Spoiler: It might look a lot like VoxEQ Verify)
Knowing everything we do about voiceprint implementations, user behavior, and the needs of a modern contact center, what would we do differently if we were building voiceprint today?
As mentioned above, we need it to work for everyone, simplify the experience on both sides, and keep impostors out.
What else?
1. Fast (and easy) to set up
For the most part, customers have existing tech teams who’re in the best position to add something new to their tech stack (and maintain it). For that to be viable, new additions need to be easy to understand and integrate with what’s already there. Integrating a solution needs to be doable in hours or days, not months.
What if we avoided a packaged solution that requires a team to learn a new tool? At the same time, we’re able to skip the vendor lock, middleware, and middle management, and provide something that in house tech teams can adapt into the existing tech stack, as long as it’s easy for them to do so.
On that same note, it should be easy to adjust the system after it’s live. That might include anything from changes to threat tolerance to what actions are taken on a flagged account, but ultimately these changes should be quick and easy to make.
2. No enrollment required
We know that enrollment is an issue for voiceprint. What if we didn’t need it?
An ideal solution protects every caller, every time, and that includes those that haven’t enrolled a voiceprint (or haven’t enrolled one yet).
3. Fits into the existing agent flows and systems
We want this tool to fit seamlessly into existing agent flows. We want to avoid a solution that adds three to five clicks per call, because those clicks add more time, cognitive load, and friction to the agent side. So where possible, we want to fit into what an agent is already doing.
In action, an ideal solution is almost invisible. An agent sees the voiceprint results as the call is routed from an IVR, giving them everything they need to authenticate faster. The same action that escalates to the fraud team also passes them the voiceprint information and how confident we are in the match, as well as any additional processing that’s happening behind the scenes.
4. No PII storage
One of the major hurdles of modern voiceprint implementations is storing and securing those prints, and ensuring that customers trust the security measures in place.
Let’s assume that one day, despite all the best efforts, we’re the victims of a cyber breach. How do we minimize the impact to our customers and their data, while balancing keeping them safe day to day?
The easiest way to make sure our customers’ information isn’t stolen is to ensure there’s nothing to steal. In this case, secure by design doesn’t mean building a bigger, stronger lock, it means there’s nothing to lock. And that means minimizing or eliminating the storage of PII.
Putting it all together
So where does that leave us?
Voiceprint was a compelling idea, and still is, in theory. But in practice, it’s become a cautionary tale: expensive to implement, hard to scale, easy to game. And most importantly, it leaves too many people out. Customers who don’t enroll. Customers who can’t. Customers who shouldn’t have to.
This goes beyond the theoretical for us. These are specific questions we’ve asked, requirements we’ve gathered, and conversations we’ve had in the process of building solutions here at VoxEQ. What if, for every call, you could answer the question ‘is this voice the customer you expect?’ And we can, for every call.
That’s not to say that our solution is the silver bullet, or that it meets every need every time. It’s to say that we’ve heard from folks across industries who are struggling with the same problems in their voiceprint solutions and continuing to refine and look for a better solution. We wanted to make something that holds up in the messy, unpredictable context of real calls, with all the noise, the forgotten passwords, and the callers who never got the enrollment memo.
If this sort of thing is on your mind too, please get in touch. We'd love to chat.