Picture the call. An AI agent has waited on hold for 20 minutes on a customer’s behalf. Your employee finally picks up. The agent has the customer’s name, address, account number, and the answers to every security question. It wants to cancel a service. And your employee realizes there is no person on the line. That may sound futuristic, but it’s happening now.
Consumers are beginning to use AI agents to contact companies for them. These tools can call businesses, wait on hold, negotiate bills, request refunds, cancel subscriptions, and schedule appointments. Google, for example, can already call local businesses to ask about prices and availability.
If an assistant can save people from waiting on hold, repeating information, or arguing about a bill, many will be happy to delegate the work. For organizations, that raises some hard questions. Is the AI agent legitimate? Does it really represent the customer? And how should an employee respond when the voice on the other end is not a person?
Here are five recommendations for preparing your organization.
1. Ask why customers would send a bot. This may be the most important question. If contacting a company becomes a job, customers will eventually hire technology to do the job for them. Some use of consumer AI agents is inevitable. But improving your own service may reduce the incentive to send bots into your organization.
2. Plan for consumer bots in your customer access strategy. That’s your overall plan for how customers reach you and get served, including the channels, routing, people, technology, and information behind it. Consumer AI agents now need to be part of that plan.
3. Verify the request. Separate identity, authentication, authority, and scope. These are four different issues.
- Identity asks: Which customer and account are involved?
- Authentication asks: Can we verify the customer’s identity and securely connect this request to them?
- Authority asks: Has the customer authorized this AI agent to act on their behalf?
- Scope asks: What, specifically, has the customer authorized it to do?
An AI agent may know the customer’s name, address, account number, PIN, and security answers. But knowing the customer’s information is not the same as having the customer’s authority. And permission to review a bill does not necessarily include permission to change a service, accept a contract, transfer money, or close an account.
4. Match safeguards to the risk. Use risk-based protocols, and match your tools to the risk. Not every interaction needs the same level of control. An agent asking for store hours or general policy information presents little risk, so keep it easy. But changing an address, canceling insurance, accessing medical information, or transferring money is, of course, very different.
5. Give employees a clear playbook. Frontline employees should not be making identity, privacy, fraud, and authorization decisions on the fly. Does the AI have to disclose that it is automated? When must the customer join the interaction? When should the contact be escalated, and how should it be documented? And what if the bot becomes repetitive, deceptive, or appears to be manipulating the process?
Employees will also need training and support. An AI agent can be endlessly persistent. It does not get tired, embarrassed, or concerned about taking too much of someone’s time. Your employees should not be left alone to manage that.
Organizations have spent years deploying AI to reduce their own service costs and effort. We should not be surprised when customers begin using AI to reduce theirs. Your next customer may not personally be on the phone. But the AI contacting you may still represent a real person, with a real need and a real relationship with your organization.
These calls are coming. Don’t wing it. Plan for them now.
0 Comments