客戶正派自己的 AI 致電給你:聯絡中心軟件準備好了嗎?
Your Customers Are Sending Their Own AI to Call You. Is Your Contact Center Software Ready?
客戶開始讓自己的 AI 助理代為致電客服,三份調查顯示服務團隊尚未準備好應對這種轉變。Zendesk 調查發現,78% 商界領袖尚無法核實代表客戶行事的 AI Agent 身份,43% 消費者希望個人 AI 助理可直接替他們處理服務問題。文章建議按請求風險訂立處理規則、獨立統計 AI 來電,並在客戶親自回電時保留來電紀錄。
Originally published at ictcontact.com
Your next caller might not be a person. Customers are starting to hand service calls to their own AI assistants, and three reports released on 18 and 19 November 2026 show how unprepared most service teams are. Zendesk found 78% of business leaders have no way yet to verify an AI agent acting for a customer. Your contact center software needs a policy before the volume shows up.
The good news is that you don't need new technology to write that policy. You need a clear rule for which requests a bot can make, a way to count those calls apart from human ones, and a record that follows the customer when they ring back themselves.
What the November reports actually say
Three surveys landed within two days of each other, and they describe the same shift from different seats.
Zendesk's CX Trends 2027 report (released 19 November) is the one that names the problem directly. In its UK release, 78% of business leaders said verifying the identity of AI agents acting on customers' behalf is a challenge they currently have no solution for. The same report found 43% of consumers want to be able to ask their personal AI assistant to resolve a service issue without contacting the company directly. It draws on two global surveys run in July and August 2026, of 3,100+ consumers and 4,400+ business respondents.
The Liveops 2026 AI-Empowered Customer Report (18 November) surveyed 750 US adults in October, all of whom had used their own AI tool, such as ChatGPT or Gemini, for a customer service problem in the past year. Of that group, 72% said they'd used personal AI to fix a problem without contacting the company at all. And 84% said they'd had to repeat to a real person something they had already told a company's chatbot. Keep the sample in mind: these are people who already use AI this way, not the general public.
Then there's AnswerConnect's 2026 AI Customer Experience Survey (19 November), run by OnePoll with 6,000 adults across the US, UK and Canada in May 2026. It found 59% find AI agents frustrating when they call customer service, up from 54% the year before, and 31% would hang up if an AI answered, up from 29%. AnswerConnect sells human receptionists, so read it as a vendor survey with a point of view. The year-on-year rise is still the bit that stuck with us.
Are AI callers really reaching contact centers yet?
Some, not many. At the Contact Centre and CX Expo in London on 18 and 19 November, CCMA chief executive Leigh Hopwood told CXM she'd been asking members whether they had "knowingly received a call from a consumer bot". At a recent CCMA summit, five or six of 70 people said yes. She called it "a bit sporadic still", held back by the cost of running your own bot and the personal data you have to hand it.
That's small. It's also how every channel shift starts. In September, Business Insider described a claims agent at a California insurer who hears the same AI caller, working for a firm of accident attorneys, a couple of times every shift, and who was docked on her employer's automated empathy score for not being nice enough to it. The bot couldn't tell it was being scored, and the scoring tool couldn't tell it wasn't scoring a human.
Our view: you don't need to panic about volume. You do need to decide what happens on the first call, because your agents are already improvising.
Why AI callers break contact center software metrics
Almost every number your floor runs on assumes a human on the other end. Average handle time, satisfaction scores, quality ratings and containment all quietly depend on a caller who gets tired, gets annoyed, or gives up.
A bot does none of that. It'll wait on hold for an hour and call back the next morning. If your save offers and IVR menus were designed partly to wear people down, an assistant built to hold out will walk straight past them, and those contacts land back in the queue.
The fix starts with counting. If AI calls get mixed into human ones, your staffing model and your agent scores both drift and nobody can explain why.
Write the policy by request type, not by caller
The most sensible approach we've seen comes from operators already handling these calls: sort requests by what they can change, then decide what a bot can do in each tier. A bot booking a table is not the same risk as a bot moving money, even if it's the same bot.
Low risk. Order status, opening hours, booking or moving an appointment. Nothing leaves the account. Serve these, and don't make your agents feel they've done something wrong by helping a bot.
Medium risk. Refunds, plan changes, a new delivery address, a dispute. Take the request, but confirm it with the account holder on the phone number or email already on file before you apply it. The bot's word isn't the customer's word.
High risk. Closing an account, moving money, password resets, SIM swaps. Only the account holder does these, full stop. Scammers have exactly the same tools as your customers, and they can place thousands of calls a day.
Honestly, the medium tier is where you'll spend most of your design time. It's also where a polite, clear script saves the most money, because the assistant will report back to its owner exactly what you told it.
Five changes to make in your contact center software this quarter
Add an "AI caller" disposition. Make it a normal wrap-up code agents can pick when the caller says it's an assistant, or obviously is. You'll get a count in your reports within a week, which is more than most teams have today.
Keep those calls out of agent scoring. Tell QA to exclude or flag them. Nobody should lose points for not charming software.
Put the tier rules in the agent script. One short paragraph per tier, including the exact sentence agents say when a request needs the account holder.
Make the customer's history follow them. When the person calls back after their bot did, the agent should see what was asked and what was promised. Liveops found 69% of its respondents rate that as extremely or very important.
Decide where your own AI fits. Low-risk bot calls are a good fit for a voice agent on your side. Let software talk to software for order status, and keep your people for the calls that need judgment.
How ICTContact contact center software handles this today
We'll be straight about it: ICTContact doesn't label AI callers for you, and we're not aware of any platform that does it reliably yet. What it gives you is the plumbing to run the policy above.
Dispositions are custom in ICTContact, so an "AI caller" code takes a minute to add, and it shows up in the Disposition Report alongside your other outcomes. Each disposition can map to a CRM status, so the contact record notes that an assistant called. The built-in CRM keeps every call, SMS, email, note, recording and WhatsApp chat on one time-ordered record, which covers point four without a connector. Call recording and the supervisor panel (listen, whisper, barge) let team leads review how agents handled the first few bot calls. The full list is on the ICTContact features page.
If you run the Service Provider Edition, the AI Voice Agent is built in. You drop it into a call flow in the IVR Designer, write its system prompt, and choose which tools it can use, such as crm_lookup, write_disposition and transfer to a live agent queue. So you can tell it, in plain words, to answer status questions from any caller, record the outcome, and transfer anything about refunds or account changes to a person. Every AI call gets a turn-by-turn transcript with each tool call shown, which is where you'll check that the rule held. The AI Voice Agent guide walks through the setup and testing.
On-premises licences are sold by concurrent channels with unlimited agents, so adding a few seats to cover a new kind of call doesn't change what you pay. You can compare editions on the pricing page.
Frequently asked questions
Is it legal for a customer to send an AI assistant to call my contact center?
There's no general rule stopping it, and the surveys above don't suggest one is coming soon. The real question is authority: whether the assistant can prove it acts for the account holder. Until there's a standard way to check that, use the risk tiers above.
How can agents tell they're talking to an AI caller?
Common signs are a number that isn't tied to the account, no answer to security questions, and the same phrasing repeated. Don't rely on voice quality, though. Synthetic voices are good enough now that sound alone won't tell you.
Should my contact center block AI callers completely?
We wouldn't. Blocking them pushes the customer back onto your phone line in person, angrier and a day later. Serving the low-risk requests and politely refusing the high-risk ones costs less and keeps the customer on side.
Do AI callers affect contact center metrics like AHT and CSAT?
Yes. A bot that waits patiently and never gets frustrated skews handle time, satisfaction and containment. Tag those calls with their own disposition and report on them separately so your staffing and quality numbers stay honest.
Does ICTContact include an AI voice agent?
Yes. The AI Voice Agent comes built into the ICTContact Service Provider Edition. It runs as a node in the IVR Designer, looks callers up in the CRM, records dispositions, and transfers to a live agent queue when a person is needed.
Get your AI caller policy in place
Bot-to-agent calls are rare today, and that's exactly why now is the cheap time to set the rules. Start with the disposition code and the three tiers, then watch the count for a month. If you'd like to see how ICTContact would fit your call flows, contact us and we'll walk you through it.
來源:DEV Community · dev.to