AI + HumanContact center operations

Run your phone queue with AI and agents working as one team

The AI takes status checks, callbacks and overflow at any hour. Your agents keep billing credits, disputes and upset callers, which beats a people-only queue or an AI-only line.

  • Overflow, not rip-and-replace
  • Hand-offs that keep the context
  • One report for AI and agents
Contact centers: Monday 9:15 am · queue wait past 45 secondsAI agent on the line
AI agentOverflow, not rip-and-replace
OutcomeTrue read ordered, late fee held · notes ready to accept · call scored

Where is my order or claim?

The AI reads the status from your system and texts the link.

Callback when the wait is long

At 9:15 am the AI books a 10:30 callback and texts the confirmation.

Why is my bill higher?

The AI explains the invoice. A credit goes to an agent.

Claims, disputes, complaints

Straight to an agent who has the authority and the time.

01 · Why AI + Human

Why contact center phones will run on AI and agents together

A US center loses 30% of its agents a year, answers in 99 seconds on average and abandons 8.9% of calls. Wrap-up takes 13.7% of agent time. Hiring faster cannot fix that, so a people-only queue is hard to sustain. Yet 87% of customers say a route to a human is essential when GenAI is used, so an AI-only line is unsafe.

AI agentsLow cost per call, instant scale, consistent
  • Answers the 9:15 am queue spike on the first ring, at any volume
  • Reads order, claim and ticket status from your system of record
  • Runs the same two-point identity check the same way on every call
On its ownIt cannot decide a billing credit, a denied claim or a retention offer, and it can misread a caller in tears.
Your peopleJudgment, authority, care, new solutions
  • Decide credits, waivers and exceptions with real authority
  • Calm an upset caller and hear what the words leave out
  • Find a fix when the case matches no script or article
On their ownA people-only queue cannot cover every peak, every night and every status call without long waits.
Shorter waits and better judgment calls on the same queueOne desk, one record of every call
  • Overflow callers get a status answer or a callback instead of hanging up
  • The agent opens each hand-off with a summary, transcript and mood, with no re-asking
  • Peak hours go to disputes and credits, not to status lookups
Side by side
People only (traditional)
AI only
AI + Human
Answer speed
Waits grow at peaks; 99 s average
First ring, at any volume
First ring, with an agent ready for hard calls
Cost of a routine call
Highest: one agent per call
Lowest: billed per second
Low on lookups; agent time saved for hard calls
Consistency
Varies by agent and shift
Same answer and same check every time
Same checks; agents adapt where needed
Judgment calls
Strong, with authority to decide
Weak: cannot grant a credit or waiver
Agent decides, already briefed
Upset or high-stakes callers
Strong on tone and trust
Poor: risks a doom loop
Reaches a person fast; needs good hand-off rules
How to think about the split
Give the AI the lookups

If the answer sits in a system of record, the AI should say it. Tag your top 15 call reasons first.

Keep judgment, money and emotion

Credits, disputes, retention and distress go to an agent who has the authority and the time.

Design the hand-off first

Agree what an agent sees and how a caller re-enters the queue. Audit 20 hand-offs a week.

The short answer

How do you run AI and human agents together in a contact center?

Put the AI where callers already wait or call after hours: overflow after a queue time-out, an after-hours forward, or one IVR option. Let it finish lookups and callbacks, and hand every other call to a person with a summary, transcript and mood. Manage both on one report: service level, abandonment, occupancy and resolution. Widen only when your written thresholds hold for two weeks.

Reviewed November 2026 · 18 sources cited below
  • US agent attrition averaged 30% in 2024; 18% of centers surveyed lose over 25% of new agents in 3 months (ContactBabel).
  • US average speed to answer was 99 seconds in 2024, up from 31 in 2012; 8.9% of calls were abandoned (ContactBabel).
  • Post-call wrap-up takes 13.7% of agent time, more than idle time at 12.2% (ContactBabel, 2025 guide).
  • 87% of customers say a way to reach a human is essential when GenAI is used (Gartner, 2026).
02 · The stakes

What a contact center director is managing

Four numbers show why the phone queue is hard to staff, and why any new tool has to prove itself against service level, attrition and hand-off quality, not against a demo.

30%

mean US agent attrition in 2024; 18% of surveyed centers lose over 25% of new agents within 3 months. ContactBabel.1

99 s

average speed to answer in US centers in 2024, up from 31 s in 2012. Abandonment was 8.9%. ContactBabel.1

13.7%

of US agent time is post-call wrap-up, more than idle time at 12.2%. ContactBabel, 2025 guide.1

4%

of service organizations reached the AI return they expected; 24% report any positive return. Gartner, reported Nov 2026.2

03 · The whole journey

Where AI and agents share the desk, from first call to win-back

A contact center touches the customer long after the sale. Here is what the queue sees at each stage, what the AI takes, and what your agents keep, so the caller never repeats themselves.

Pre-sale questionsSTAGE 1
Activation and setupSTAGE 2
First useSTAGE 3
Everyday supportSTAGE 4
Upgrades and add-onsSTAGE 5
Cancel-or-keepSTAGE 6
Lapsed customersSTAGE 7
The moment
A prospect calls to ask about plans and hears 'all agents are busy' at 9:15 am.
A new customer cannot finish first login or setup, and waits on hold.
The first bill or order looks wrong, so the customer calls before trusting you.
Status and how-to calls fill the queue, and peaks bury the hard calls.
A caller asks about a plan change, but no agent has time to offer an option.
A caller says 'cancel' after a third call on one case, and no one owns it.
Lapsed customers are never called, so their reasons for leaving go unheard.
AI agent
Answers plan and coverage questions from the knowledge base, or books a callback.
Walks through the first steps, checks identity, and texts the setup link.
Finds the invoice or order, explains each line, and states the status.
Takes status checks, how-to and account changes, and tries the first fixes.
Answers plan questions and flags callers who fit an upgrade for your team.
Spots repeat callers and the word 'cancel', then routes to a retention agent with the history.
Calls lapsed accounts within quiet hours and consent rules to ask why they left.
Your team
Takes the buyer ready to compare options and hands warm leads to sales.
Steps in when the setup fails or the customer is stuck and frustrated.
Handles first-bill disputes and early billing credits with authority.
Takes claims, complaints and technical cases, already briefed by AI Assist.
Presents the offer and closes the plan change on a call with a ready customer.
One agent with authority listens, fixes the cause and decides the offer.
Handles the replies that need an offer, an apology or a fix.
What to track
Abandon rate on pre-sale calls
Setup completed without a repeat call
First-call resolution on new accounts
Service level and hand-off quality
Plan changes from support calls
Save rate on cancel calls
Reasons captured and accounts returned

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 calls should the AI own, start, or leave alone?

Sort your top 15 call reasons into three piles before you change any routing. Volumes are typical for a mixed consumer line. First-call resolution by call type is SQM's benchmark. Tag your own calls for two weeks before you copy this.

AI handles it end to endAI starts, a person finishes with the contextStraight to a person
Work
Volume
Who takes it
Why
Status: order, claim, ticket, booking
'Where is my claim?' The AI reads the status from the system of record, says it, and texts the link.
high
AI handles it
A lookup, not a judgment. Orders resolve first time 72% and general inquiries 74% (SQM), so the ceiling is high.
Hours, policy, coverage and how-to
Return window, office hours, what the outage notice says, how to reset the modem.
high
AI handles it
Answers live in the knowledge base. Scoring every call finds a wrong or stale answer in days, not at the next audit.
Callback booking when the queue is long
The wait is 4 minutes at 9:15 am. The AI books a 10:30 callback and texts the confirmation.
medium
AI handles it
Turns an abandon into a scheduled contact without adding a seat. Skip it if your callbacks already miss their window.
Payments and balance calls
The AI reads the balance and due date, then sends a secure pay link or moves the caller to your payment IVR.
medium
AI, then a person
Card digits spoken to the AI or an agent can put recordings in PCI scope. Keep them out of the recording path.
Identity checks and account changes
Address change, password reset, plan swap. The AI runs your standard two-point check first.
high
AI, then a person
Manual checks cost time on every call (ContactBabel puts US spend at $12.3bn a year). A failed check goes to a person.
Billing questions and credits
'Why is this bill higher?' The AI finds the invoice and explains it. A credit or waiver goes to billing.
high
AI, then a person
Explaining is a lookup. Deciding a credit needs authority and tone. Billing resolves first time 71% (SQM).
Technical troubleshooting
The AI runs the first steps (power cycle, status page, model check), then hands off with the steps already tried.
medium
AI, then a person
Technical calls resolve first time 63% (SQM). Never make the caller repeat the steps.
Claims, disputes and complaints
A denied claim, a disputed charge, a complaint about last week's visit.
medium
Straight to a person
Claims resolve first time 59% and complaints 47% (SQM). Judgment, authority and tone decide the outcome.
Cancel-or-keep, distress, safety, repeat callers
Retention offers, an injury report, a caller in tears, a third call on one case, anyone who says 'agent'.
low
Straight to a person
Trust is spent or the stakes are high. One owner with authority beats another round of questions.
TakeawayThe AI owns lookups and callbacks. People own judgment, money and emotion. The blend rows are where hand-off design decides whether callers feel helped or processed.
05 · One call, start to finish

One overflow call, from queue time-out to scored record

Your ACD, IVR and agents stay. The AI sits behind one overflow rule. This is a single call at 9:15 on a Monday, when the queue passes 45 seconds.

0:00

The queue passes 45 seconds

Your ACD overflow rule forwards the call over SIP to a VoiceGenie number. The caller does not hang up or call back.

0:03

The AI says it is an AI

It names the company, says it is an AI assistant and that the call is recorded, then offers help or a callback. 'Agent' always reaches a person.

0:30

Identify and check

It matches the caller by number, asks the two-point check your policy requires, and reads back the name and account it found.

1:30

Finish or hand off

A status or callback finishes here. Anything else goes to the right skill queue with a one-line summary, the transcript, mood and steps done.

2:00

The agent takes it with AI Assist

The agent sees the live transcript, does not re-ask, decides, then accepts the proposed disposition, callback time and notes in one click.

End

Both halves land in one report

Recording, transcript and QA score are saved. Supervisors see AI and human calls together: hand-offs, service level and outcomes by interval.

06 · Start small

How to start: one queue, peak overflow, eight weeks

Do not put the AI at the front door. Put it where callers already wait or hang up, write your pass marks before the first call, and change one thing at a time so you know what moved the numbers.

Start with this

Peak-interval overflow on one queue: status lookups and callback booking

Overflow callers are already waiting or abandoning (8.9% of US calls in 2024). Status and callbacks are lookups with a system of record, so the AI cannot make their experience worse than a long wait.

How calls reach the AI

Your ACD's overflow rule (queue time-out of 45 to 60 seconds) forwards one queue's calls over SIP to one VoiceGenie number, in the two worst intervals of the day. IVR and ACD stay as they are.

Weeks 1 to 2

Baseline by interval

  • Pull 30-minute data for the queue: offered, answered, service level, ASA, abandons, average time to abandon.
  • Tag the top 15 call reasons for two weeks and pick the AI scope from your own mix.
  • Write escalation rules, the hand-off fields an agent sees, and how a transferred caller re-enters your queue.
  • Run 40 test calls with staff: angry callers, bad lines, and 'agent' said at every step.
Weeks 3 to 4

Overflow in two intervals

  • Turn on the overflow rule for the two worst intervals only. The AI does status and callback booking, nothing else.
  • A supervisor listens live in week three. The team reviews every transferred or failed call each morning.
  • Send WFM the AI-handled and transferred counts per interval so the forecast sees the new flow.
  • Fix knowledge gaps and tool errors daily, one change at a time.
Weeks 5 to 6

All peaks, plus after hours

  • Extend the rule to every interval where the longest wait passes your threshold, and to the after-hours forward.
  • Turn on AI Assist for agents on that queue: live transcript, proposed disposition and notes.
  • Re-forecast AHT for human calls. The easy calls are gone, so the average on what is left may rise.
  • Hand-score 30 calls and compare with automatic scores. Fix the scorecard before widening.
Weeks 7 to 8

Gate review, then widen

  • Hold the review against the pass marks written in week one. Do not move them now.
  • If met, add one IVR option (status) in business hours, with 'agent' always reaching a person.
  • If not met, name one cause, one owner and one date. No decision by week eight means no widening.
  • Share the interval report with the floor: what the AI took, what came back, what agents gained.

Expand when all of these are true

  • Abandonment in covered intervals is down by at least a third from baseline, with service level at or above target.
  • At least 90% of audited hand-offs reach the right queue first time, and callers do not repeat themselves.
  • 7-day repeat rate on AI-handled calls is no higher than on agent-handled calls with the same reason.
  • Recording notice and AI disclosure are on 100% of scored calls, and no agent group runs over 90% occupancy.
07 · What to measure

Metrics a contact center leader can defend

Baselines are national figures, not yours. Capture your own by 30-minute interval in weeks 1 and 2. Planning ranges are what to aim for with AI and people together, not promises.

Metric
Typical today
Planning range with AI and people
Service level, ASA and abandonment1
Share answered within threshold (80/20 is the convention), average speed to answer, and abandons. Pull by interval from the ACD.
ASA 99 s; 8.9% abandoned (2024)
Abandons 4 to 6% in covered hours
Occupancy and shrinkage3
Occupancy is time handling calls over logged-in time. Shrinkage is paid time off the phones. Track both by interval.
Avg 83% in calculator data; shrinkage 30 to 35%
Peaks held at 85 to 90%, not above
AHT and wrap-up (human calls)1
Talk plus hold plus after-call work for agent-handled calls, split by call reason. Expect the mix to get harder.
Wrap-up 13.7% of agent time
Wrap-up down a third on assisted calls
Agent attrition, total and 90-day1
Agents who left in the last 12 months over average headcount, plus new agents who left within 90 days. Read monthly.
30% (2024); 18% lose over 25% of new hires
2 to 5 points lower after 12 months
AI resolution rate (in-scope calls)
Share of in-scope AI calls solved with no repeat contact on the same reason within 7 days. Not the same as containment.
No independent voice benchmark
55 to 75% of in-scope calls
Hand-off quality4
Share of transfers that reached the right queue first time, and share of audited hand-offs where the caller repeated themselves.
12.0% of calls transferred (2022)
Under 10% of audited calls repeat
First-call resolution and 7-day repeat5
Issue solved with no further contact on it. Measure by post-call survey and by 7-day repeat. Internal counts run high.
71% by survey; 78% center-reported
Hold, then gain 1 to 3 points
QA coverage and calibration4
Share of calls scored, and how closely automatic scores match a human reviewer on the same 30-call sample.
All calls recorded at 60% of recording users
100% scored, 30 hand-scored a month

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 the numbers look like for a 40-agent center

Capacity is the gain: calls answered instead of abandoned, agent hours moved from the busiest intervals to calls that need judgment, and less overtime. Change the inputs to match your own queue.

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
$13 · 50%value of one abandoned call answered and resolved · share of recovered calls that become one
Estimated monthly gain
$84,963a month
Today, per call
$6.24
With VoiceGenie
$3.43
2,700more calls answered a month
1,350extra abandoned call answered and resolveds a month
$17,550value of those outcomes
22,500calls 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

  • Call length: the average US service call ran 426 seconds in 2022, so 7 minutes 4. 30,000 calls a month is about 42 agents of talk time at 34 calls a day.
  • Cost per person: BLS median CSR pay of $21.53 an hour 6 times the 1.43 total-to-wages ratio in BLS compensation data 7 is about $64,000 a year, or $5,300 a month, before software and supervision.
  • Calls per person per day: 8 hours times 49.6% talk time 1, divided by 426 seconds, is about 34.
  • AI finishes 25%: Planning range for peak overflow, after-hours and one IVR option. No independent voice benchmark exists. Missed calls of 9% match the 8.9% abandonment rate in 2024 1.
  • Outcome value: $13 is two avoided repeat contacts at the $6.55 average cost of an inbound call in 2022 4. It leaves out retention value on purpose. The 50% conversion is a placeholder.

What the dollar figure leaves out

  • Peaks stop pushing agents past 90% occupancy, where burnout and early exits concentrate.
  • New agents get a live transcript and notes; a chat-support study found novices gained most (34%).
  • Every call, AI and human, is recorded and scored, so QA and coaching see all of it, not a sample.
09 · Rules to get right

What you must get right on a recorded, AI-assisted line

This is what to get right and what to ask any vendor, not legal advice. Rules differ by state and change often, so have counsel review your call flow before launch.

US states

Consent to record comes before recording8

California and other all-party states require consent before the call is intercepted. A generic 'may be monitored' line or a later click is likely to fail. Name the AI vendor and what it captures.

US states

Say it is an AI in the first sentence9

Utah requires disclosure when a caller asks, and up front in high-risk interactions that collect sensitive data and give advice (law runs to July 2027). One rule meets all: say 'I'm an AI assistant' at the start of every call.

Global

Keep card numbers out of recordings10

PCI guidance says card validation codes must not be stored after authorization, even encrypted, and recordings that can be queried count. Send a pay link or use masked capture so the AI and the transcript never hear digits.

US

Outbound AI calls generally need consent11

The FCC ruled in 2024 that AI-generated voices are 'artificial' under the TCPA, so outbound calls generally need the called party's prior express consent (written for marketing). Keep the consent record, quiet hours and opt-out handling for callbacks and reminders.

US

Never trap a caller away from a person12

The CFPB's 2023 chatbot report described 'doom loops' with no off-ramp to a person and said firms may face liability when chatbots fail legal requirements. Make 'agent' work at every step and escalate on failed attempts.

Four questions to ask any voice AI vendor, including us

  1. If your service or my ACD link is down, do calls fall back to my own queue, and how fast?
  2. Do you use my call audio or transcripts to train your models, and how long is each kept?
  3. Is consent captured before the first recorded word, and can I change the wording by state?
  4. Can AI and human calls be in one report that reconciles to my ACD's service level and abandons?
10 · Where it plugs in

The systems this sits next to

VoiceGenie typically connects to these through SIP trunks, the open API, post-call webhooks or Zapier. Check your ACD's overflow and transfer options first.

ACD, IVR and telephony

  • GCGenesys Cloud
  • NCNICE CXone
  • ACAmazon Connect
  • AVAvaya
  • CWCisco Webex Contact Center
  • 8X8x8

Workforce management

  • VEVerint
  • NWNICE WFM
  • CACalabrio
  • ASAssembled
  • INinjixo

QA and speech analytics

  • VEVerint
  • NNNICE Nexidia
  • CACalabrio
  • MAMaestroQA
  • SCScorebuddy

CRM, ticketing and knowledge

  • SASalesforce
  • HUHubSpot
  • ZOZoho
  • ZEZendesk
  • SEServiceNow
  • COConfluence
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 for the dialer, AI agents and your agent desk

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 for agents. VoiceGenie covers all four on one record. You set it up once, and each input, whether an integration, a knowledge article, a corrected answer or a QA finding, improves both the AI and your people.

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 connection to your systems

Connect your CRM, billing and ticketing once through the open API, webhooks or Zapier. The AI and your agents work from the same record.

One knowledge base

Write a policy once. The AI answers from it, and every call is scored, so a wrong or stale answer shows up in days.

One recording and QA view

AI and agent calls are recorded, transcribed and scored together, so one finding can coach an agent and fix the AI.

12 · What goes wrong

Five ways AI contact center projects go wrong

Replacing the IVR menu on day one
Start with overflow or one option. 'Agent' must always reach a person. Move the next option only after the gate is met. Only 9% of customers fully resolve an issue in self-service (Gartner, 2019).
Cutting headcount before results are in
Spend early gains on peak relief and training. Reduce through attrition once thresholds hold. In 2026, 21% of leaders cut for AI and 7% had already rehired (Gartner).
Forecasting human AHT as if nothing changed
The AI takes the easy calls, so what is left is longer. Re-forecast AHT on the new mix, send WFM per-interval overflow counts, and do not judge agents against old AHT.
A hand-off that drops context or the queue place
Define the payload: summary, transcript, mood, steps done. Agree how a transferred caller re-enters the queue. Audit 20 a week. 83% of surveyed consumers repeat themselves after transfers (2026 survey).
Judging by containment and a QA sample
Containment counts callers who gave up. Report resolution and 7-day repeat. Score every call, including the AI's, and hand-score 30 a month to calibrate.
13 · Questions

Questions people ask before they start

Does adding AI mean fewer agents?

Treat AI as capacity first. It answers overflow and after-hours calls and finishes lookups so people take the calls that need judgment. Gartner's 2026 survey found 21% of service leaders cut headcount for AI and 7% had already rehired. Reduce through attrition once your thresholds hold, and spend early gains on peak relief and training.

Can I add AI without changing my ACD or IVR?

In most set-ups, yes. VoiceGenie works with your own carrier and numbers, so it can sit behind one IVR option, take overflow after a queue time-out, or answer after hours without changing your phone system. Confirm that your ACD can forward on a time-out and accept a transfer back with context. Then start with one queue.

How much does it cost?

AI calls are $0.14 per minute, billed per second, with the phone line not included, so a five-minute AI call is $0.70. Human seats are $29 (Desk), $59 (Insights) or $99 (Copilot) per person per month on annual billing. There is free trial credit and no minimum commitment. Enterprise pricing is private and covers in-country deployment and done-for-you set-up.

How long does a pilot take?

Plan eight weeks: two weeks measuring your own baseline by interval, two weeks of overflow in the worst intervals, two weeks extending to all peaks with AI Assist on, then a gate review. The technical connection depends on your carrier and ACD, so confirm the forwarding rule first. Do not shorten the baseline weeks, because they decide whether the pilot proves anything.

Will callers hate talking to an AI?

Some will, so give them an exit. Gartner's 2026 survey of 3,566 customers found 87% say a route to a human is essential when GenAI is used. A 2026 vendor survey found trust in AI resolving the issue rose from 26% to 55% when a human option was visible. Say it is an AI, finish quickly, and make 'agent' work every time.

What resolution rate should I expect from voice AI?

No independent benchmark exists for voice AI, and vendor figures rarely define the term. Containment counts callers who gave up. Track resolution: the issue solved with no repeat contact within 7 days. As a planning range, expect 55 to 75% resolution on in-scope calls, and compare it with your own agents on the same call reasons.

How does AI change my forecast, occupancy and shrinkage?

AI calls do not draw on agent shrinkage, but the calls left for people are harder, so their handle time may rise. Re-forecast AHT on the new mix, give WFM overflow volumes by interval, and hold peak occupancy near 85 to 90%. Above 90% is where practitioners warn of burnout and attrition.

Is it compliant to record calls and use AI?

It can be, if you get the basics right. Obtain consent before recording starts, because several states require all-party consent. Say it is an AI in the first sentence, keep card numbers out of recordings, and keep a working route to a person. Ask any vendor whether it trains on your audio. Have counsel review your state rules.

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

Sources

  1. ContactBabel, The US Contact Center Decision-Makers' Guide 2025 (17th edition) (2025).
  2. CMSWire (reporting a Gartner survey), Just 4% of Customer Service Leaders Hit Their Expected Automation Returns (2026).
  3. Call Centre Helper, What Are the Industry Standards for Call Centre Metrics? (2026).
  4. ContactBabel, The US Contact Center Decision-Makers' Guide 2023 (15th edition) (2023).
  5. SQM Group, What Is a Good First Call Resolution Rate? (2022).
  6. US Bureau of Labor Statistics, Customer Service Representatives: Occupational Outlook Handbook (2025).
  7. US Bureau of Labor Statistics, Employer Costs for Employee Compensation, Q2 2026 (2026).
  8. Epstein Becker & Green, Eavesdropping by Algorithm: How Courts Are Applying Decades-Old Privacy Laws to AI (2026).
  9. Davis Polk, Utah scales back reach of its generative AI consumer protection law (2025).
  10. PCI Security Standards Council, Information Supplement: Protecting Telephone-based Payment Card Data (v2.0) (2011).
  11. Cooley, FCC: AI-generated robocalls are illegal under the TCPA (2024).
  12. Consumer Financial Protection Bureau, Chatbots in consumer finance (2023).
  13. ContactBabel, The US Contact Center Decision-Makers' Guide 2026 (2026).
  14. Gartner, Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent (2026).
  15. Gartner, Gartner Says Only 9% of Customers Report Solving Their Issues Completely in Self-Service (2019).
  16. CMSWire (reporting Five9's 2026 CX report), Customers: The AI Handoff Is Broken, Contact Center Leaders Won't Hear It (2026).
  17. National Bureau of Economic Research, Generative AI at Work (Brynjolfsson, Li, Raymond) (2023).
  18. Call Centre Helper, How to Calculate Staffing in a Contact Centre (2026).