AI + HumanCustomer support

Tier-1 phone support where AI and your team answer every call

AI answers status, policy and simple account calls on the first ring. Your team keeps the exceptions, the cancellations and the upset callers, so callers get speed and a person who can decide.

  • Starts with one call type, not the queue
  • Says it's an AI. People are one ask away
  • Sits behind your current IVR and carrier
Customer support: Wednesday 2:14 pm · IVR option 2, order statusAI agent on the line
AI agentStarts with one call type, not the queue
OutcomeDelayed-order call handed over with order, ETA and request. Caller did not repeat herself.

Where is my order?

A late tent, a missing parcel. One lookup and a tracking text.

Return or exchange

A damaged item. The AI opens the return, an agent handles exceptions.

Why was I charged more?

The AI explains the invoice. Refunds and credits go to a person.

I want to cancel

Revenue and goodwill at stake. An agent decides the offer.

01 · Why AI + Human

Why support phones will run on AI and people together

Callers still pick up the phone when something has gone wrong, and 8.9% of US contact center calls were abandoned in 2024. Most status and policy calls repeat all day, so staffing for them is costly and slow to scale. Yet 87% of customers say a way to reach a human is essential when GenAI is used. A people-only line queues on Monday. An AI-only line strands the caller who needs a refund approved.

AI agentsLow cost per call, instant scale, consistent
  • Answers the where-is-my-order call on the first ring, at 8 pm too
  • Absorbs the Monday spike with no queue message
  • Gives the same return-window and warranty answer on every call
On its ownCannot approve an expedite, a refund or goodwill, and can miss the mood of a caller on their third call.
Your peopleJudgment, authority, care, new solutions
  • Approve an overnight replacement or a refund on the spot
  • Read an upset caller and decide what to offer
  • Spot the pattern behind the third call about one outage
On their ownA people-only line queues on Mondays and spends agent hours on status and password calls.
Fast answers on routine calls, better decisions on hard onesOne desk, one record of every call
  • Status and policy calls finish with no wait, so agents have time for exceptions
  • Expedite and refund calls reach an agent with order, ETA and request, no repeated story
  • Every call, AI and human, is scored, so repeat calls show what to fix
Side by side
People only (traditional)
AI only
AI + Human
Answer speed
Queue at peaks, 99 s average in 2024
First ring, any hour
AI on the first ring, a person when needed
Cost per routine call
Highest: a paid agent on every status call
Lowest: billed per second
Low on routine calls, people spent where they matter
Consistency
Varies by agent and shift
Same policy answer every call
Same policy answer, plus a person's context
Judgment calls
Strong on refunds, goodwill, retention
Cannot approve or weigh goodwill
Person decides with the AI's summary in hand
Upset or repeat caller
Can calm and fix, once reached
Misses mood, may loop
Routed to a senior agent with full history
How to think about the split
Split by judgment, not by topic

Sort calls by how much judgment the answer needs and what the caller stands to lose. Lookups go to the AI. Exceptions go to people.

Give facts to AI, authority to people

Let the AI read the order and the policy. Keep refunds, expedites and retention offers with an agent who can say yes.

Widen the AI only when numbers hold

Add a call type when its transfers and 7-day repeat rate match your people's. Roll back when they do not.

The short answer

Which customer support calls can an AI handle, and which should a person take?

An AI can fully handle the predictable tier-1 calls: order or case status, policy and hours questions, and simple account changes after an identity check. It should start but not finish returns, basic troubleshooting and billing questions, then hand over with context. Complaints, cancellations, fraud and safety calls go straight to a person. Start with one call type, such as order status, and measure resolution, not containment.

Reviewed November 2026 · 15 sources cited below
  • Own: status lookups, policy answers, simple changes after ID check
  • Share: returns, first-step troubleshooting, billing explanations
  • Keep with people: complaints, cancellations, fraud, safety
  • Judge success by resolution and 7-day repeat calls, not containment
02 · The stakes

Phone is where customers go when it matters, and it is getting harder to reach.

Callers still choose the phone to fix problems, yet more of them hang up before an answer and self-service rarely finishes the job. Adding AI helps only if a person stays within reach.

8.9%

of calls were abandoned in US contact centers in 2024, up from 7.1% in 2023 (ContactBabel survey)1

9%

of customers fully resolve an issue in self-service (Gartner survey of 8,398 customers, reported 2019)2

83%

of consumers say they at least sometimes repeat themselves after an AI-to-human transfer (2026 vendor survey of consumers)3

87%

say a way to reach a human is essential when GenAI is used (Gartner, 3,566 customers, Feb to Mar 2026)4

03 · The whole journey

Where AI and your team help a customer across their whole life with you

Support calls are not one moment. They run from the first question before a purchase to the call that decides whether a customer stays. Here is who does what at each stage on one desk.

Pre-sale questionsSTAGE 1
Order and onboardingSTAGE 2
First useSTAGE 3
Everyday supportSTAGE 4
Upgrades and add-onsSTAGE 5
Renewals and cancellationsSTAGE 6
Lapsed customersSTAGE 7
The moment
A shopper asks about shipping or fit before buying and gets voicemail.
The customer placed an order or signed up and wants to know what happens next.
The product arrives and the customer is stuck on step one, or it is damaged.
Status, billing and account calls come in daily. The wait is the problem.
A happy customer wants an upgrade, but the call ends once the issue is fixed.
A customer says cancel. The right offer depends on the account and the reason.
A past customer went quiet after an unresolved problem or a missed renewal.
AI agent
Answers shipping, stock and return-policy questions from the knowledge base on the first ring.
Confirms the order, sends setup steps or a tracking link by text, and flags missing details.
Walks through known first-step fixes and opens a return when policy allows.
Owns status lookups and simple account changes after an ID check, and explains invoice lines.
Notes interest in upgrades or accessories and offers a call-back or a text link.
Recognizes the intent, says a person will help, and transfers at once with the reason.
Places a polite outbound call within quiet hours and consent rules to check in.
Your team
Takes the shopper who needs advice on a complex order or a custom quote.
Steps in when an address, payment or account setup needs a decision.
Tier 2 takes the unresolved case with the steps already tried.
Handles refunds, disputes and the repeat caller on an unfixed issue, with full history.
Follows up on the interested customers, with the call summary and account in view.
Listens, decides what to offer, and keeps the relationship where it can be kept.
Takes the customer who wants to return or still has a complaint to settle.
What to track
Pre-sale calls answered; calls turned into orders
Fewer where-is-my-order calls after purchase
First-call resolution on first-use issues
Resolution rate and 7-day repeat calls
Add-on interest captured and followed up
Saved accounts and cancellation reasons logged
Lapsed customers reached and reactivated

Read down a column for one stage, or across a row to see what the AI and your team each do from first contact to win-back.

04 · AI or person

Which support calls the AI owns, which it starts, and which go straight to your team.

Sort calls by how much judgment the answer needs and what the caller stands to lose, not by how they phrase it. Tag your own top ten call reasons for two weeks first. This table is a starting point for that exercise.

AI handles it end to endAI starts, a person finishes with the contextStraight to a person
Call type
Volume
Who takes it
Why
Order or case status (WISMO)
"Where is my order?" "What is the status of my claim or ticket?"
high
AI handles it
One lookup, one answer, same whoever picks up. Verify the caller, read the status, send the tracking link by text.
Policy, hours and how-to questions
Return window, warranty terms, store hours, "does it ship to Canada?"
high
AI handles it
The answer is in your knowledge base. Keep the articles current and the AI will not guess.
Simple account changes after ID check
Update an email or address, resend an invoice, reset a portal link
medium
AI handles it
Rule-based and reversible. Two facts from the record first, and a person if the check fails twice.
Returns and exchanges
"This arrived damaged, I want to send it back."
medium
AI, then a person
AI gets order, reason and photos link and opens the return if policy allows. Exceptions and goodwill go to a person.
First-step troubleshooting
Router will not connect, app will not log in, printer shows an error
medium
AI, then a person
AI runs the known first steps, then hands to tier 2 with what was tried. SQM: technical calls resolve first time 63%.
Billing questions
"Why is this charge higher?" "I was billed twice."
medium
AI, then a person
AI pulls the invoice and explains the line items. Refunds, credits and disputes need a person with authority.
Repeat caller on an unresolved issue
Third call about the same outage or the same unfixed order
low
Straight to a person
The customer has already been failed. Route on the repeat flag to a senior agent with the full history.
Cancellation and retention
"I want to cancel my plan." Renewals at risk.
low
Straight to a person
Revenue and goodwill are at stake and the right offer depends on the account. A person decides what to offer.
Fraud, safety, legal or vulnerable caller
Unrecognized transactions, injury, a threat to sue, a caller in distress
low
Straight to a person
High harm and high liability. The AI recognizes the signal, says so, and transfers at once with a summary.
TakeawayThe AI owns lookups, your team owns exceptions, and the middle is shared. Move a call type toward the AI only when its transfers and repeat calls look as good as your people's.
05 · One call, start to finish

One delayed-order call, from first ring to resolution.

The AI does the lookup and the explaining. A person does the one thing that needs authority. The caller tells the story once.

0:00

Answers, discloses, listens

The AI says it is an AI, that the call is recorded and that a person is available. The caller describes the problem in their own words.

0:20

Verifies the caller

It asks for the order number and billing ZIP and checks them against the order system. Caller ID alone is not identity.

0:35

Looks up the order

It reads status, carrier scan and promised date from the order system. It finds a carrier delay and a new delivery date.

1:00

Tells the truth, offers what policy allows

It gives the new date and offers anything within its limits, such as a delay text. The caller needs the item sooner.

1:25

Hands off with context

Expedite or refund needs approval, so the AI says so and routes to the order-issues queue with a one-line summary, transcript, the caller's mood and the lookup.

1:45

Person decides, AI Assist writes it up

The agent approves an overnight replacement. AI Assist proposes the disposition and notes, accepted in one click. The whole call is scored, AI and human.

06 · Start small

Start with one call type behind one IVR option. Widen only when the numbers hold.

Do not move the queue. Pick the call type with the most predictable answer, point one door at the AI, review every call, and write the pass marks before you start so the go or no-go is not a debate.

Start with this

Order or case status calls ("where is my order")

High volume, the answer sits in one system, callers are rarely upset, and a wrong answer is easy to spot. A failed call costs a transfer, not a customer.

How calls reach the AI

Add one IVR option, such as "press 2 for order status", that forwards to a VoiceGenie number. Saying "agent" or pressing 0 still goes to your queue. A failed ID check also goes to your queue.

Week 1

Baseline and build

  • Pull 60 days of ACD data: tag top call reasons, then record AHT, transfer rate and 7-day repeat rate for status calls
  • Write the knowledge base and the verification rule, for example order number plus billing ZIP
  • Give the AI read access to the order system and set the transfer rules and queue hours
  • Run 30 test calls: bad audio, an angry caller, a wrong order number, a request for a person
Weeks 2 to 3

Live in one window, every call reviewed

  • Turn the IVR option on for weekdays 10 am to 2 pm only
  • Review every call next morning and tag each failure: lookup, understanding, policy or hand-off
  • Check that each transfer arrived with summary and transcript, and time how fast a person picked up
  • Count callers who asked for a person and whether they got one on the first ask
Weeks 4 to 6

All hours, sample review

  • Open the option all day and add shipping, returns-policy and hours questions
  • Review 20 calls a day, plus every transfer and every caller who calls back within 7 days
  • Compare resolution, repeat contacts and CSAT with the baseline from week 1
  • Hold a weekly review with the agents who take the transferred calls
Weeks 7 to 10

Add the next call type

  • Add returns: the AI collects order and reason and opens the return when policy allows
  • Offer the AI or a callback when queue wait passes 60 seconds, for status calls only
  • Set pass marks for each added call type before turning it on
  • Decide in writing: widen, hold or roll back

Expand when all of these are true

  • At least 70% of status calls resolved with no person, checked on a 100-call sample by listening
  • 7-day repeat rate on AI-handled calls no higher than your human rate for the same calls
  • At least 9 in 10 transfers were needed, and fewer than 10% of transferred callers had to repeat themselves
  • CSAT on AI calls within 3 points of human calls, with no ID-check or disclosure failures
07 · What to measure

Eight numbers that tell you whether the line is working.

Measure whether the problem was solved, not whether a person was avoided. Take baselines from your own ACD and CRM first. Published figures are context, and the targets are planning ranges, not promises.

Metric
Typical today
Planning range with AI and people
Resolution rate (in-scope calls)
In-scope calls where the need was met with no person and no repeat call in 7 days. Sample 100 calls a week by listening.
No independent benchmark found
55% to 75% of in-scope calls
Containment rate
Calls that ended without a transfer. Never report it alone: a caller who hangs up in frustration counts as contained.
No independent benchmark found
25% to 40% of all calls at scale
7-day repeat contact rate5
Callers who contact you again about the same issue within 7 days, by any channel. Match on customer ID and reason.
29% call back at 71% FCR (survey)
No worse than humans, then down 2 to 4 pts
First-call resolution5
Issues solved on first contact. Survey customers after the call. Internal reports run 10 to 20 percent higher.
71% by survey, 78% self-reported
Hold in pilot, then up 2 to 4 pts
Transfer rate and precision6
Share of calls transferred, and share of transfers a person confirms were needed. Sample 20 a week.
12.0% transfer rate, US centers, 2022
Precision 90% or higher
Hand-off quality3
Share of transferred calls where the caller repeated their reason or details. Sample 20 transfers a week.
83% repeat at least sometimes (survey)
Under 10%
Abandonment and speed to answer1
Callers who hang up before an answer, and seconds to a live voice. Report AI answer time and time to a person separately.
8.9% abandoned, 99 s average, 2024
AI under 5 s, abandonment under 5%
CSAT on AI-handled calls
Post-call score on AI-handled calls, compared with human-handled calls of the same type.
No independent benchmark found
Within 3 points of human calls

Planning ranges are targets to test in a pilot, not promises. Measure your own baseline for two weeks before you change anything.

08 · The return

What tier-1 automation is worth to a 6,000-call-a-month support line.

Use your own numbers in the calculator. The example is one support line taking about 6,000 calls a month, with the AI finishing the predictable third and answering calls that would otherwise be abandoned.

A worked example. Move the sliders to match yours.

7 minaverage call length
$5,300loaded monthly cost per person
34 a daycalls one person handles
$30 · 50%value of one abandoned call answered and resolved, saving a sale, a renewal or a repeat call · share of recovered calls that become one
Estimated monthly gain
$22,608a month
Today, per call
$6.24
With VoiceGenie
$3.26
540more calls answered a month
270extra abandoned call answered and resolved, saving a sale, a renewal or a repeat calls a month
$8,100value of those outcomes
4,200calls your team still takes, with context

AI at $0.14 a minute with volume credit applied, plus Insights seats for the people who take handed-over calls. Phone line charges are your carrier's and are not included. Illustration, not a quote. Open the full calculator

Where these numbers come from

  • Cost per person: BLS median CSR pay of $21.53 an hour 7 times 1.43 for benefits and taxes 8, over 2,080 hours, is about $64,000 a year. Our estimate, before software, supervision and facilities.
  • Calls per person per day: 8 hours x 49.6% talk time 1 divided by a 426-second average service call 6 gives about 34. Derived. Replace it with your ACD data.
  • Missed share of 9% matches the 8.9% US abandonment rate in 2024 1. Your peak-hour rate is probably higher.
  • AI finishes 30% is a planning range for a mature rollout, not a benchmark. The first pilot call type will be far below it. Count it from your own sample.
  • Outcome value and conversion are placeholders for one saved sale, renewal or repeat call. For context, the average US inbound call costs $7.20 9.

What the dollar figure leaves out

  • Agents take fewer status and password calls and more calls where they can change the outcome
  • Callers reach an answer at 8 pm and during the Monday spike, not a queue message
  • New hires get live help on the human part of the call; in a chat-support study, novices resolved 34% more issues per hour [s11]
09 · Rules to get right

Rules to get right before the AI takes a support call.

Inbound support carries fewer telemarketing rules than outbound, but recording consent, identity checks, card data and honesty about the AI all still apply. This is a checklist for your counsel, not legal advice.

US states

Get recording consent before the call starts10

California and other all-party consent states need consent before recording begins. Consent given after recording starts does not count. Say it in the first sentence, before anything is captured, and log the wording.

US states

Limit what your vendor does with call audio10

Courts are still deciding whether an AI vendor is your tool or a third-party eavesdropper. In a 2025 federal case in California, a claim over an AI call vendor went ahead on a capability theory, turning on what the vendor does with the data. Bar training use in the contract.

US states

Say it is an AI in the first sentence11

Utah requires disclosure on a clear request, and up front for high-risk interactions such as collecting sensitive data or giving advice. Rules differ by state and are changing. Disclosing in the first sentence is the safest practice.

Global

Keep card numbers out of the AI path12

If the AI, transcript or recording holds card numbers or security codes, they enter PCI scope, and security codes may not be stored after authorization. Send a payment link by text or hand off to a masked payment step.

US

Verify identity before account details13

Caller ID is not identity. Check two facts from the record before reading anything account-specific, and move to a person after two failures. The CFPB warns of weak safeguards around sensitive data. Telecom and banking add sector rules.

US

Never trap a caller or guess an answer13

The CFPB found chatbots that loop customers with no route to a person and give wrong information, and warned of legal exposure. Give every caller an exit to a person and test answers against policy.

Four questions to ask any voice AI vendor, including us

  1. Is any of our call audio or transcript used to train your models or anyone else's? Is that in the contract?
  2. How does the AI verify identity, and can it speak account details before the check passes?
  3. How are card numbers kept out of the AI, the transcript and the recording?
  4. Show a hand-off: what does the agent see, and what happens when no agent is free?
10 · Where it plugs in

Where the AI reads and writes during a support call.

A support call touches the ticket system, the order system and the customer record. VoiceGenie connects to HubSpot, Salesforce, Zoho and Google, and to anything else through the open API, webhooks or Zapier.

Help desk and CRM

  • ZEZendesk
  • FRFreshdesk
  • SSSalesforce Service Cloud
  • HSHubSpot Service Hub
  • JSJira Service Management
  • SEServiceNow

Orders, shipping and billing

  • SHShopify
  • NENetSuite
  • SHShipStation
  • UAUPS and FedEx tracking
  • STStripe
  • CHChargebee

Phone system and carriers

  • TWTwilio
  • TETelnyx
  • VOVonage
  • RIRingCentral
  • 8X8x8
  • Any SIP trunk or PBX

Knowledge and identity

  • COConfluence
  • NONotion
  • GUGuru
  • GDGoogle Drive
  • OKOkta
  • AUAuth0
Keep your numbers and carrierTwilio, Plivo, Telnyx, Vonage or any SIP trunk or PBX. Put the AI behind one menu option, on overflow or after hours.
CRM and calendarsHubSpot, Salesforce, Zoho and Google connect directly.
Everything elseThe systems above connect through the open API, post-call webhooks or Zapier.
11 · One platform

One platform, set up once, so every input improves AI and people

You do not need one vendor for the dialer, another for the AI agent, a third for the agent desk and a fourth for AI tools that help your people. VoiceGenie covers all four on one record. You connect your systems, write your knowledge base and set your rules once. Each new integration, article, corrected answer or QA finding then improves the AI and your agents together.

The usual way

Four tools, four set-ups

Dialerown contract · own integration
own training · own reports
Voice AI agentown contract · own integration
own training · own reports
Agent deskown contract · own integration
own training · own reports
Agent assist and QAown contract · own integration
own training · own reports
With VoiceGenie

One platform, set up once

AI agents
AI-native dialer
Human Agent Desk
AI Assist
Shared by the AI and your people
  • Your phone numbers and carrier
  • CRM and calendar connections
  • Knowledge and call rules
  • Recordings, QA and analytics
One order and ticket connection

Connect the order system and help desk once. The AI reads status and logs the call. Agents see the same record in the desk.

One knowledge base

Write return windows and warranty terms once. The AI answers from them, and one fixed article corrects every later call.

One recording and QA view

Every AI and human call is recorded, transcribed and QA-scored in one view. A finding fixes an answer or coaches an agent.

12 · What goes wrong

Five ways phone automation projects go wrong.

Judging success by containment
Count a call as resolved only if the need was met and the caller did not call back within 7 days. Report containment and resolution side by side, by call type.
Making the AI the only door
Keep "agent" and 0 working at every point. Gartner found 87% of customers call a human option essential, and Gartner notes repeated failed AI attempts make people less likely to use the tool again.
Launching on a stale knowledge base
Name an owner for every article, load policies as they are today, and read failed calls first each morning. The AI is only as current as the page it reads.
Handing off without context
Pass a one-line summary, transcript, mood and what the AI already did. 83% of consumers say they repeat themselves after a transfer. Sample 20 transfers a week.
Starting with the hardest calls
Complaints resolve first time 47% of the time and general inquiries 74%. Begin with status and policy calls, earn the numbers, then add returns and billing.
13 · Questions

Questions people ask before they start

What customer support calls can AI handle?

AI can handle calls with a predictable answer found in one system: order or case status, policy and hours questions, and simple account changes after an identity check. It can start returns, first-step troubleshooting and billing explanations, then hand to a person. Complaints, cancellations, fraud and safety calls should go straight to your team.

How much does AI phone support cost per call?

VoiceGenie charges $0.14 per minute for the AI, billed per second, with the phone line extra. A six-minute call costs about $0.84 in AI time. Human seats start at $29 per person per month on annual billing. For context, ContactBabel's 2026 guide puts the average US inbound call at $7.20, though cost definitions vary by center.

How long does it take to set up AI on a support line?

Connecting a number behind one IVR option is the quick part. The weeks go into the knowledge base, the identity check, transfer rules and reviewing calls. Plan about four weeks to reach full-day coverage of one call type, and about ten to add a second. Set pass marks before you start and decide in writing at the end.

Will callers hate talking to an AI?

Some will. Gartner found 64% of customers would prefer companies not use AI in service, mainly because they fear it is harder to reach a person. Design around that: say it is an AI, answer fast, resolve the routine call, and let anyone reach a person on the first ask. People tolerate automation that finishes the job.

Do I have to tell callers it is an AI and that the call is recorded?

Treat both as required. Recording consent must come before recording starts in all-party consent states such as California, and a buried privacy link may not count. AI disclosure rules vary by state, and Utah requires it on a clear request or up front for high-risk calls. One first sentence covering both is the safest practice. Ask your counsel.

How is this different from my IVR or chatbot?

A phone menu makes callers fit your categories. Only about 9% of customers fully resolve an issue in self-service, per Gartner. A voice AI lets the caller describe the problem in their own words, checks your systems, and acts. When it cannot finish, a person gets the summary and transcript, so the caller does not start over.

Will this replace my agents?

The aim is capacity, not fewer people. The AI answers every call and finishes the routine ones, and your agents take the calls that need judgment. US center agent attrition averages about 30% a year, so many teams use the extra capacity to cover turnover and shorten waits. Keep strong human staffing so callers who ask for a person get one fast.

What is a good containment or resolution rate?

No independent benchmark exists, and vendor numbers rarely define terms. Judge resolution, not containment: calls where the need was met and nobody called back within 7 days. For the first call type, a planning range is 55% to 75% of in-scope calls. Set your pass mark from your own human baseline for the same call type.

Try it on one call type

Hear it handle your calls before you change anything.

Talk to our AI agent in your browser, or start a trial and point one line at it. Your team keeps every call that needs them.

No credit card · No minimum commitment · Keep your carrier