From Know-Your-Customer to Know-Their-Agent

Personal agent calls are coming, but fraudsters, aggregators, and scrapers are already here.

Building the future of customer experience means understanding intent, value and resolution, whatever the channel. Customers want their problems resolved with as little friction as possible. When the stakes are high, they want it immediately, from someone (or something) they trust.

The quickest way to get something done without being bothered is to send someone (or something). That is Jeeves territory, P.G. Wodehouse’s valet who settles Bertie Wooster’s problems before Bertie has finished describing them (always a fun read). Modern AI assistants are a new breed of Jeeves, with as much autonomy as we want to give them and some firm views on what is best for the user. 

With the sharp increase in agent traffic, bank security leaders keep asking me what to do with it: how do we build common trust and continuous value for our customers? The customer is king, but what about trusted envoys? How do we welcome and guide agents without putting customers at risk? 

With Meta, Google and others putting personal assistants in front of large user bases, errands will be handed off and many of those will end up in your contact centers. It’s not hype, just reality. The agents that book restaurants today will be disputing a charge tomorrow. 

The phone is getting busier, not quieter

People have been predicting the end of the phone call for decades, on repeat. We tend to forget customers want to maintain a synchronous channel of communication when the stakes are high or when the need is immediate. They will also want their agents to get things done immediately in whatever channel suits that need for immediate resolution. Synchronous and immediate attention drives customer loyalty. 

When mobile apps took off, everyone said they’d kill the contact center. App usage grew, and banking call volume kept going up anyway. Agent adoption will start small and grow, and voice will still be there the whole time. So the front door needs work, because that’s where the new traffic shows up. 

Nobody is telling you how much of that growth is bots and agents. It’s hard to measure, and that’s part of the problem. When automation sits on the customer side, calls get cheaper to make, so more of them get made. Some come from fraud rings. A growing share comes from legitimate agents acting for real customers.

The blind spot

This part isn’t new. Aggregators and IVR scrapers have been calling banks for years, across millions of calls. If you run a large contact center, you already know that. Most still can’t see that traffic. One large bank estimates it gets roughly 10 million of those calls a year that it simply can’t see.

Those callers are a different kind of bot, but they’re still bots. Without visibility you’re flying blind. You can’t decide how to treat an aggregator call versus a personal assistant if you can’t tell them apart.

Personal agents are the next wave of a similar blind spot, and it’ll be bigger.

Why agents will show up on the phone first

I’ve seen this over and over in this business. New technology takes the path of least friction.

Personal assistants won’t wait for every bank and airline to adopt a new protocol. They’ll use what already works. That means web apps where they exist and phone calls everywhere else. Every company already has a phone number and already answers it. There’s no integration or partnership to set up first.

Until protocols catch up, the path of least resistance wins, and that path is voice.

So personal agents will reach your contact center by phone before they reach it through any clean integration. And we have to handle this today, in a far-from-perfect present. 

Detect first, then decide

You can’t set a policy for callers you can’t see. Detection comes first. Policy comes second, and the policy belongs to the enterprise, not the vendor.

On a phone call, detection can’t rely on one signal. Single-factor authentication hasn’t been enough for years. What works is stacking signals and scoring them together:

• Liveness. Is this a live human voice or a synthetic one? 
• ANI and spoof risk. Does the calling number look legit, or does the network data say it’s spoofed?
• Blocklists. Does the voice match a known fraudster or a known bad bot?
• Behavior and coherence. Does the conversation hang together like a real request?

Those signals roll up into a risk score, and the enterprise decides what to do with it. It can block the call, challenge it with a step-up, send it down a path built for bots, or welcome it.

Most security stacks don’t have that last option. If you can tell a legitimate personal agent from an attacker, you can actually serve the agent. Give it a structured path and let it finish the task fast. That’s a policy choice, and you can only make it if you can see who’s calling.

Detection tells you it’s an agent. Policy decides whether to trust it, what it’s allowed to do, and who carries consent and liability. Can it check a balance? Change an address? Move money? Saying yes safely takes trust and permissions on top of detection. That’s what makes “agents welcome agents” an honest line.

At Omilia we built this multi-signal layer into TalkGuard, and it has been shipping ahead of its time. I bring it up so you know this isn’t theoretical. The bigger point is the front door.

People blur one distinction here all the time. Liveness answers “is this a machine?” An allow list answers “is this a valid agent for this user?” and “did the speaker change mid-call?” Those are different questions. A legitimate personal agent fails a liveness check by design, because it is a machine. That doesn’t make it fraud. If you collapse liveness and identity into one yes-or-no gate, you’ll block every helpful agent and still miss some of the bad ones.

On the lab side, we have always worked on real problems end to end. Our models are proven in real conditions and we always built for purpose. 

This is still about people

It’s easy to read all this as a security story. Mostly it’s a customer experience story.

Voice of the customer only works if you know who the customer is. If your front door can’t classify callers, that signal breaks in one of two ways. Either every bot looks like fraud and you start blocking customers who chose to use an assistant. Or every agent looks like a person, and your sentiment scores, containment rates, and call reasons fill up with machine traffic that tells you nothing about how people feel.

The first costs you customers. The second costs you the truth about them. Either way, CX and VoC break as soon as you can’t tell who’s calling.

The hybrid world

I see two camps getting this wrong in opposite directions.

The first is CCaaS built only for human agents. Queues, routing, and workforce plans all assume every call is a person who needs a person. As personal-agent volume rises, those systems will drown. You can’t staff your way out of demand that moves at machine speed.

The second is the agentic-first crowd building only for web bots. They’ve got clean APIs and good demos. They’ll miss the phone, and that’s where a lot of this traffic lands first.

Both camps miss that it’s going to be hybrid.

You need both.

Personal agents will take more of the volume. People will still handle the judgment calls, the moments that need empathy, and the exceptions that don’t fit a script. A good platform has to run both on the same call path.

In practice that means a few things. Detect who’s calling. Let the bot resolve what it can. When it can’t, hand a clean record to a person so the customer doesn’t have to repeat themselves. And keep voice of the customer from both sides, human and agent, so you still learn from every call.

Agents welcome agents

“Agents welcome agents” sounds like a slogan, but it’s more of a policy. Companies that build a voice front door now will treat legitimate personal agents as customers. They’ll serve them fast and save people for the moments that need a person.

Companies that wait for a perfect world will end up treating every non-human call as fraud. Their customers will notice, and so will their competitors.

The agents are already dialing. Pick up.