Call Deflection and AI Call Routing: Resolve More Calls Without Adding Agents

Ryan Stevens
Contact center agent with a headset handling a call at his workstation

Last updated: November 16, 2026

Call deflection means handling a customer's request somewhere other than a live agent's queue: in a text, a callback, a self-service page or a conversation with an AI agent. It works when the request gets done. AI call routing handles the calls that still need a person, sending each one to the right team from what the caller says.

Call deflection is any method that resolves a customer contact without a live agent taking the call.

This article is for contact centers and multi-location operators who want fewer calls waiting in the queue without making customers work harder. For the wider picture of how AI fits a contact center, see how an AI call center handles peak volume across locations.

What is call deflection, and how is it different from containment and routing?

Call deflection is the work of moving a request away from a live agent to a channel that can finish it. Call containment is a narrower measure: the share of calls an automated system keeps to the end without handing the caller to a person. Call routing decides which person or team gets the calls that do need one.

The terms overlap, so fix what each one means before you set targets.

Term What it means What it counts
Call deflection A request handled without a live agent, by any channel Contacts kept away from the agent queue
Call containment A call kept by the IVR or AI agent to the end, with no transfer Calls the system finished without a handoff
Call resolution The caller's request is done and they don't need to contact you again about it Requests completed, checked by repeat contacts
Call routing Choosing where a call goes: which team, location or person Calls sent to the right place on the first try
AI call routing Routing decided by an AI agent from what the caller says in their own words Intent picked up and acted on without a phone menu
Intelligent call routing Routing that uses data, such as caller history, location, language or skills, to pick the destination Better matches between caller and destination

The call deflection meaning that matters for planning is the third row. A deflected or contained call that ends with the task undone usually comes back as a second contact, so measure resolution alongside the other two.

What separates good call deflection from bad?

Good call deflection finishes the caller's task in fewer steps than waiting for an agent would, and keeps a person within reach. Bad deflection pushes callers to another channel that can't finish the job, or makes reaching a person hard.

Two Gartner surveys show how often self-service falls short. In a survey of 5,728 customers conducted in December 2023, Gartner found that only 14% of customer service issues were fully resolved in self-service, and only 36% of issues customers described as "very simple" (Gartner press release, August 19, 2024). In the same release, 45% of customers who started in self-service said the company didn't understand what they were trying to do.

A later Gartner survey of 3,566 B2B and B2C customers, run in February and March 2026, found that 87% say it is essential for companies to provide an option to reach a human agent when they use GenAI for service (Gartner, August 4, 2026). Half of the customers in that survey said their interactions are easier when companies use GenAI.

Put together, the findings point to a few tests every deflection method should pass:

Callers tend to accept deflection when Callers push back when
The new channel finishes the whole task They're sent to a page or menu that doesn't cover their issue
They choose it, such as accepting a text link or a callback They're moved without being asked, or the call ends
What they said carries over, so they don't repeat it They have to explain the problem again to a person
A person is one clear step away The way to a person is hidden or takes several failed attempts
It's faster than the queue they would have waited in It takes longer than holding would have

Call deflection strategies, ranked by how callers experience them

The eight call deflection strategies below are ranked by one test: how likely the caller is to get the task done without repeating themselves, with a person still in reach. The order is our judgment, built from the tests above, and the right mix depends on your call reasons.

Rank Strategy What the caller experiences Works best for Watch out for
1 Proactive notifications The answer arrives before they think to call Delivery updates, appointment reminders, outages, payment due dates Messages that invite a reply nobody reads
2 AI resolution on the call They say what they need and it gets done on the same call Bookings, reschedules, order status, account questions, routing Agents that can read your systems but can't write to them
3 Callback that keeps their place They hang up and get called back at their turn Peak hours, long queues Callbacks that arrive late or at a bad time
4 Text link offered during the call They accept a link to finish a form, payment or upload on their phone Document uploads, payments, address changes Links sent without asking, or forms that don't work on a phone
5 Self-service portal or app They do it themselves when it suits them Account changes, invoices, order history Portals that miss common tasks
6 Web chat or messaging with a handoff They type instead of talk, with a person available Questions that need a link or a picture Bots with no path to a person
7 Keypad self-service in the IVR They press keys to hear a balance or confirm an appointment One-step lookups with a number to enter Deep menus and options that don't match the caller's words
8 "Visit our website" messages They're told on hold to go online, often without a link Rarely; at most a short line while they wait Messages that end the call or repeat on a loop

Proactive notifications rank first because the customer never has to pick up the phone. Pull your top call reasons and ask which ones a timely text would have answered, such as "where is my technician?" or "is my order shipped?".

AI resolution ranks second because it doesn't ask the caller to change channels. An AI agent that answers in natural language (or as we call it, a Voice AI employee) picks up the call, works out the request and does it in your systems. A current one, like Newo, can book the appointment, read the order and write the outcome back before the call ends, then hand the call to a person when it should.

Callbacks rank third because they remove hold time without solving anything early. They move the work to later, so staff them like any other queue.

The bottom two need the most care. An IVR that answers balance checks is useful. A message that tells people to go online and then hangs up turns a phone call into two contacts. If you already run an IVR, our guide to how IVR phone menus work and where they fall short covers what to keep.

How does AI call routing work?

AI call routing listens to what the caller says, works out the intent and the details that matter, and sends the call to the destination your rules set for that intent. It resolves the request itself when it can. It replaces "press 1 for billing" with "tell me what you're calling about", so callers who don't know your departments still reach the right one.

Most call routing AI follows the same five steps on every call:

  1. Listen. The caller explains the request in their own words, in their language.
  2. Identify. The AI picks out the intent (reschedule, billing dispute, emergency) and the details (location, account, date), and looks up the caller by phone number where it can.
  3. Check the rules. It applies what you set for that intent: business hours, which location or team owns it, language, priority and any customer tiers.
  4. Resolve or route. If the AI is allowed to finish the request, it does. If not, it tells the caller what happens next and transfers.
  5. Pass the context. The person who picks up gets the intent, the details collected and a short summary, so the caller doesn't start again.

AI call routing examples

Your rules decide what the AI finishes and what it hands over.

What the caller says What the AI picks up What happens
"I need to move Thursday's appointment at the Elm Street office to Friday." Reschedule, Elm Street location, Friday Checks Elm Street's calendar, moves the booking and texts a confirmation
"Where's my order? It was meant to arrive yesterday." Order status, order found from the caller's number Reads the status from the order system and offers a tracking link by text
"Water is coming through my ceiling." (at 11 p.m.) Emergency, after hours Tells the caller it's connecting them, then transfers to the on-call number
"I was charged twice this month." Billing dispute, account found Transfers to billing with the account and a summary of the charge
"¿Puedo hablar con alguien sobre mi tarjeta?" Spanish, card account question Continues in Spanish and routes to the team that owns that card product
"I want to cancel my membership." Cancellation Routes to the retention team, as your rule for cancellations says

Intelligent call routing methods compared

Intelligent call routing is a family of methods, and AI routing usually combines several of them on one call.

Routing method Decides on What the caller does
Menu (IVR) routing The key pressed Listens to options and picks one
Time-based routing Time of day and holidays Nothing; the call goes to the open team or the after-hours number
Location routing Which site the caller wants or is near Names the location, or it's matched from the number dialed
Skills-based routing Agent skills, such as language or product Nothing, once the need is known
Data-directed routing Caller history in your CRM, such as an open ticket or account tier Nothing; known callers go straight to their team
Intent-based AI routing What the caller says they need Speaks normally

Routing rules to write before go-live

The AI follows the routing rules you write, so write them before launch. Fill in a table like this one for each call reason, and review it with the team leads who receive the transfers.

Call reason Can the AI finish it? If not, route to After hours Context the person needs
Book or reschedule Yes, within your booking rules Front desk of that location AI books the next open slot Requested time, location, patient or customer name
Billing dispute No Billing team Take details, promise a callback Account, charge amount and date, what the caller wants
Emergency No On-call number On-call number Address, what happened, callback number
Cancellation No Retention team Take details, promise a callback Account, reason given

How do you measure call deflection?

Measure call deflection with five numbers side by side: deflection rate, containment rate, resolution rate, transfer rate and misroute rate. Add customer satisfaction from post-call surveys. Any one of them alone can look good while callers have a bad time, so report them together, next to the handle time and first contact resolution figures you already track.

Metric Formula What it tells you
Deflection rate Contacts handled without a live agent ÷ contacts that would have reached one How much work left the agent queue
Containment rate Calls the AI finished with no transfer ÷ calls the AI answered How often the AI kept the call
Resolution rate Contained calls with no repeat contact on the same issue within 7 days ÷ calls the AI answered How often the task got done
Transfer rate Calls handed to a person ÷ calls the AI answered How much still needs your team
Misroute rate Transfers that had to be sent on to another team ÷ all transfers Whether routing works

A worked example

These are example numbers to show the working, not benchmarks. Say the AI takes 6,000 calls a month for two reasons, bookings and order status, and the people who handled them before spent 6 minutes on each call on average.

  1. The AI finishes 3,600 calls with no transfer, so containment is 3,600 ÷ 6,000 = 60%.
  2. It transfers 2,400 calls, so the transfer rate is 40%.
  3. Of the 3,600 contained calls, 300 callers contact you again about the same thing within 7 days. Resolution is 3,300 ÷ 6,000 = 55%.
  4. Agent time saved is 3,300 resolved calls × 6 minutes = 19,800 minutes.
  5. The median customer service representative earned $21.53 an hour in 2025, according to the Bureau of Labor Statistics. That's about $0.36 a minute in wages alone, before benefits and overhead.
  6. Wage cost saved is 19,800 × $0.3588 = about $7,100 a month, before you subtract what the AI costs to run.

Counting contained calls instead of resolved ones would give 21,600 minutes and about $7,750, an overstatement of roughly $650 a month from 300 callers who had to come back. Put resolution in every report so the savings figure holds up.

Call deflection implementation checklist

Use this checklist to roll out deflection and AI call routing one call reason at a time.

  1. Pull three months of call reasons from your IVR, call recordings and agent dispositions, and rank them by volume.
  2. Mark each reason as prevent (a proactive message could answer it), resolve (an AI agent or self-service can finish it) or route (it needs a person).
  3. For each "resolve" reason, write down what finished looks like in the system where the result lands.
  4. Fill in the routing rules table for every "route" reason, including after-hours destinations.
  5. Connect the AI to the systems it has to read and write, and check its knowledge: hours, prices, policies and transfer numbers for every location.
  6. Test with real recordings of your own calls, including accents, background noise and callers who change their mind.
  7. Make the way to a person one clear step, and tell callers before every transfer.
  8. Send a small share of live calls to the AI first, with your current flow as the fallback.
  9. Track the five metrics weekly, and read a sample of transcripts with a team lead.
  10. Add the next call reason only when resolution and misroutes hold steady.

How Newo handles deflection and routing on your existing line

Newo is a Voice AI agent platform that answers calls in front of the phone system you already run. Against the checklist and the top-ranked strategies above, Newo's own pages show the following.

  1. Newo identifies the caller, reads the account, completes what it can and writes the outcome back to your CRM or ticketing system before the call ends.
  2. Newo holds routing, hours and services per location, so a caller who names a site reaches that site's calendar and team.
  3. The agent tells the caller before it transfers, then transfers. You set transfer numbers for working and non-working hours, the situations that should go to a person, and separate numbers for situations such as emergencies or group bookings. Check how the call summary reaches the person who picks up in your setup.
  4. With SMS turned on, the agent sends the caller a text with content you approve, such as booking details or a link.
  5. The same agent handles callbacks, confirmations and follow-up, and can call back instantly when someone fills in a form on your website.
  6. Newo answers the number you already publish, and your numbers, routing and contact center platform stay where they are.
  7. The agent runs voice to voice with no text step in the middle. It waits for a real pause before it speaks, stops when the caller talks over it, and works in 15+ languages.
  8. Newo lists everything the agent knows, document by document, before it takes a call.
  9. Calls from every location show in one dashboard.

Newo holds ISO/IEC 27001:2022 and SOC 2 Type 1, supports HIPAA with a signed BAA and GDPR with Newo as processor, and uses 256-bit encryption.

Routing can be the whole first project. In a published case study, a certified Newo partner replaced the IVR flow at a Mexican contact center with 14 departments. Within two weeks, first-contact routing accuracy rose from 65% to 95%, and misroutes and transfers fell from 35% to 5%.

Build your own Voice AI employee

Start with one call reason you'd like off the queue. Create your own Voice AI employee from your website, call it, and hear how it handles your callers' requests.

Frequently asked questions

What is call deflection? Call deflection is resolving a customer's request without a live agent taking the call, through proactive messages, an AI agent, a callback, a text link or self-service. It only saves work when the request gets done, so measure resolution as well as deflection.

What is call containment? Call containment is the share of calls an IVR or AI agent finishes without transferring the caller to a person. A caller who hangs up in frustration can still count as contained, so check repeat contacts too.

What are some examples of call avoidance? In customer service, call avoidance usually means preventing calls before they happen: delivery and appointment texts, outage notices, payment reminders and order tracking pages. In call center management, people also use the term for agent behavior that dodges calls, such as staying in after-call work or releasing calls early.

What does it mean if someone is deflecting? In everyday speech, deflecting means turning a question or criticism away instead of answering it, often by changing the subject. In a contact center, call deflection has a neutral meaning: sending a request to a channel that can finish it.

What is intelligent call routing? Intelligent call routing sends each call to a destination chosen from data, such as the caller's history, location, language or the skills of available agents. AI call routing adds the caller's own words, so the request decides the destination instead of a menu choice.

Does AI call routing replace an IVR? It can replace the menu, with callers saying what they need instead of pressing keys. Many teams keep their phone system and existing routing and put an AI agent in front of it, which takes the calls it can finish and transfers the rest.

More from Newo

Book a demo