AI + HumanContact center for financial services

Banks and lenders win when AI and people share every call

The AI answers first, checks identity and handles status, due dates and fraud alerts. Your staff keep disputes, credit and hardship, so callers get speed and judgment.

  • Answers on the first ring
  • Hands off with the full record
  • Your staff keep credit decisions
Financial services: Monday 9:20 am · peak queue, 11 callers waitingAI agent on the line
AI agentAnswers on the first ring
OutcomeCard locked · dispute time-stamped 9:22 am · replacement card with fraud specialist

Disputed charge

Caller reports a charge from a store they never visited. Lock, stamp, hand off.

Loan application status

"Did you get my pay stubs?" The AI reads status and what is missing.

Fraud-alert call-back

"Did you make this purchase?" A yes releases the hold; a no goes to fraud.

Locked out of online banking

A one-time code to the phone on file, then a reset or a person.

01 · Why AI + Human

Why bank and lender phones will run on AI and people together

Callers now wait 99 seconds on average, and 63% of mobile users missed a bank call they did not recognize. A people-only team cannot staff every Monday 9:20 am peak. An AI-only line cannot judge a scam in progress, a hardship call or a credit decision. Money, rules and trust sit on both sides.

AI agentsLow cost per call, instant scale, consistent
  • Answers on the first ring, even with 11 callers waiting
  • Reads loan status and due dates the same way every time
  • Places fraud-alert call-backs and reminders at set pacing
On its ownIt cannot weigh a scam in progress, explain a credit decision or comfort a member in hardship.
Your peopleJudgment, authority, care, new solutions
  • Spot a scam story that sounds true and stop the transfer
  • Decide disputes, credit and exceptions with real authority
  • Calm a member after a death, job loss or missed payment
On their ownA people-only desk cannot answer every status call at peak or after hours without long waits.
The AI clears the routine so staff can judge the restOne desk, one record of every call
  • A disputed charge is locked and time-stamped at 9:22 am, then the fraud desk takes over
  • Loan officers start each call knowing what is missing, with no re-asking
  • Fraud-alert call-backs go out from your name, and any doubt reaches a specialist
Side by side
People only (traditional)
AI only
AI + Human
Answer speed at peak
Waits grow when the queue passes staffing
First ring, every caller, any hour
AI answers first; people take the calls that need them
Cost per routine call
Full staff time for status and due dates
Lowest, billed per second
Low on routine calls; staff time kept for hard ones
Consistency of answers
Varies by agent and by shift
Same approved answer every time
Approved answers, reviewed by staff through QA
Judgment calls
Strong on scams, hardship, exceptions
Weak; cannot weigh context or tone
A person decides, with the full record in hand
Dispute and fraud handling
Strong, but the report may be stamped late
Fast stamp, but cannot investigate or decide
Stamped at once, investigated by a specialist
How to think about the split
Sort calls by cost of a wrong answer

If a wrong answer costs a re-call, let the AI finish it. If it moves money or starts a clock, a person finishes it.

Give credit and crises to people

Decisions, reasons for a decline, scams in progress and hardship stay with staff. The AI gathers the facts first.

Measure resolution, not containment

Count calls solved with no repeat in 3 days. Review transcripts so people and AI both get better each week.

The short answer

How can banks and lenders use AI on phone calls without risking compliance?

Let the AI answer first and own the calls with one right answer: hours, application status, payment due dates and fraud-alert confirmations. Hand disputes, card locks, access resets and new loan inquiries to a person with full context. Keep credit decisions, scams in progress and hardship with your staff. Start with one call type on overflow, and measure resolution, not containment.

Reviewed November 2026 · 16 sources cited below
  • AI discloses itself, is recorded, and offers a person at any point
  • Identity is checked before any account detail is spoken
  • A phone report of a dispute is time-stamped at the first minute
  • Credit decisions and advice never come from the AI
02 · The stakes

Callers wait longer, miss your calls, and lose money to people pretending to be you.

Phone is still the main live channel in financial services. When the queue is long or your outbound call looks like spam, the customer who needed you most is the one who gives up.

99 sec

Average speed to answer in US contact centers, 2024, up from 31 seconds in 2012 (ContactBabel decision-makers' survey)1

63%

of US mobile subscribers said they missed a call from their bank because they did not recognize the number (First Orion survey of 5,000+, late 2021)2

$3.5B

reported lost to US imposter scams in 2025; bank impersonators led business-impersonation losses (FTC, June 2026)3

31%

of 100 US banking customers surveyed said they stopped using their institution after repeated bad contact-center experiences (Deloitte, 2026, small sample)4

03 · The whole journey

How AI and your staff serve a member from first inquiry to coming back

A bank or lender relationship runs for years and phone calls mark the key points. The AI handles the routine step at each point. Your staff take the money, credit and trust decisions, with the full record in front of them.

Account or loan inquirySTAGE 1
Application and approvalSTAGE 2
First use and setupSTAGE 3
Servicing and fraudSTAGE 4
Added productsSTAGE 5
Payment help and renewalSTAGE 6
Lapsed or closed accountsSTAGE 7
The moment
Caller asks about rates or a car loan and reaches voicemail or a queue.
Applicant calls to ask what is missing or if the file moved.
New member cannot log in, activate a card or set up bill pay.
A charge looks wrong, a card is lost, or a payment date is unclear.
A member with a car loan may need a card or savings, but nobody asks.
A payment is late, a CD is maturing, or a member is in hardship.
A member goes quiet, moves money out, or closes an account.
AI agent
Answers on the first ring, learns the need and timeline, gets consent, and books the loan officer.
Verifies identity, reads status and missing documents, and texts the upload link.
Sends a one-time code to the phone on file, activates the card, and walks through app setup.
Locks the card, time-stamps the report, collects facts, and hands off with the record.
Calls members with consent, asks one question about a need, and books a banker.
Sends due-date reminders with a secure pay link, and flags maturing products for a call.
Places a courteous check-in call, asks what changed, and logs the reason.
Your team
The licensed loan officer quotes rates, checks eligibility and gives advice.
Underwriters explain decisions, handle exceptions and work with applicants who are declined.
Staff step in when the code fails or the member needs help with a joint account.
Fraud specialists decide disputes, spot scams in progress and arrange replacements.
The banker reviews goals, explains terms and makes suitable offers.
Staff handle hardship, set up plans and talk through renewal choices.
A relationship banker calls back with an offer to fix the real problem.
What to track
Inquiries answered and booked
Fewer status re-calls
Accounts active in first week
Time to card lock
Referrals booked
On-time payments
Members won back

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 financial services calls should the AI own, start, or never touch?

Sort every call by what a wrong answer costs. If the cost is a re-call, let the AI finish it. If the cost is money moved or a regulator's clock, let the AI start it and a person finish it. If it is credit or a crisis, go straight to a person.

AI handles it end to endAI starts, a person finishes with the contextStraight to a person
Call type
Volume
Who takes it
Why
Hours, branch, routing number, wire cutoff
"Is the Elm Street branch open Saturday?" or "What time is the wire cutoff?"
high
AI handles it
Public facts, one right answer, no account data. Answer in seconds and text the address.
Application status and missing documents
"Did you get my pay stubs?" "What is still outstanding on my auto loan?"
high
AI handles it
Read-only from the loan system after an identity check. The AI reports status; it never explains a decision.
Payment due date, amount due, how to pay
"When is my mortgage payment due?" then a secure pay link by text
high
AI handles it
Routine and repeatable. Card numbers never enter the AI path; the link or your payment system takes them.
Fraud-alert call-back (outbound)
"Did you make a $412 purchase in Columbus at 8:14 am?" Yes releases the hold.
medium
AI handles it
A yes or no answer, short call, and an opt-out. Any 'no' or doubt moves to a fraud specialist.
Lost, stolen or compromised card
"I left my debit card in a taxi." Lock now, replace later.
medium
AI, then a person
The AI verifies and locks fast. A person confirms the mailing address and any pending charges.
Locked out of online or mobile banking
"I changed phones and can't log in." Reset after a one-time code.
high
AI, then a person
A top target of social engineering. The AI sends a code to the device on file; failures go to a person.
Dispute a charge or transfer
"I don't recognize this $89 ACH debit." Gather date, payee, amount, and notice time.
medium
AI, then a person
Oral notice can start Regulation E timelines. The AI captures facts and the time; staff investigate and decide.
New loan or account inquiry
"I want a used-car loan, about $22,000." Needs, timeline, consent to be contacted.
medium
AI, then a person
The AI qualifies interest and books the loan officer. Rates, eligibility and pre-approval stay with licensed staff.
Scam in progress, hardship, estate or POA
"A man says my account is hacked and told me to move my money." Or a death in the family.
low
Straight to a person
Money may be leaving now, or the caller is distressed. Judgment, empathy and authority beat speed here.
TakeawayIn the table above, the AI finishes only the first four rows, but it starts the next four, so a person never opens a call cold. Tag your own two weeks of calls to see how your mix compares.
05 · One call, start to finish

One disputed-charge call, minute by minute, from first ring to fraud desk.

This is the Monday 9:20 am call from the top of the page. The AI does the fast, safe parts. The person does the judging. Nothing is repeated.

0:00

Answers on the first ring

The AI says who it is, says it is an AI, says the call is recorded, and tells the caller a person is one step away at any time.

0:20

Verifies out of band

It texts a one-time code to the mobile number on file. No account number or SSN digits are spoken. A failed code goes to a person, not a second guess.

0:50

Stops the loss and starts the clock

With the code matched, the AI sends the card-lock request to your card system and records the exact time of the report. Oral notice can start the Regulation E clock.

1:20

Collects the facts

Date, merchant, amount, card in hand or not, other charges. It reads each one back so the fraud team does not have to re-ask.

2:00

Hands off with the record

The fraud specialist sees a one-line summary, the transcript, the caller's mood and the actions taken. The caller hears a name, not a transfer into silence.

After the call

Disposition in one click

AI Assist proposes the case notes, disposition and call-back time. The specialist accepts them. The supervisor sees the call, the QA score and the clock time in one view.

06 · Start small

Start with loan status and missing-document calls on overflow.

Pick the call that has one checkable answer and no advice in it. Send only overflow and after-hours calls to the AI. Read every transcript. Expand only when the conditions below hold.

Start with this

Loan application status and missing-document calls, on overflow and after hours

Read-only lookups, no credit advice, no money moved, and a steady stream of "where is my application?" calls. It builds your identity check, hand-off and QA habits before disputes.

How calls reach the AI

Your main number and IVR stay the same. If no one answers in four rings, after 5 pm, or when the queue passes 90 seconds, the PBX or carrier forwards the call to a VoiceGenie number.

Weeks 1-2

Baseline and rules

  • Tag two weeks of calls by reason; log rings-to-answer, abandons and voicemails by hour and by branch phone
  • Write the do-not-answer list: rates, eligibility, decisions, fees waived, anything in dispute
  • Set the identity check with compliance, using a code to the number on file for anything beyond public facts
  • Finish your vendor-risk review and decide who gets the hand-off and after-hours messages
Weeks 3-4

Staff test

  • Staff place 50 scripted calls, 15 of them traps such as 'why was I declined?' or 'what's my rate?'
  • Read every transcript; fix wrong answers, long pauses and failed hand-offs
  • Test the hand-off to a loan officer in hours and the next-morning queue after hours
  • Test the caller who says 'person' at the first sentence; it must work every time
Weeks 5-8

Live on overflow

  • Turn on forwarding for no answer in four rings and for after hours only
  • Review every call in week one, then a 20 percent sample plus every transfer
  • Hold a 15-minute weekly review with the lending and member-service leads
  • Post four numbers weekly: answered, finished, transferred, wrong-answer count
Weeks 9-12

Decide and add one thing

  • Compare results to the baseline and the exit conditions below
  • If they hold, add one call type: due dates with a pay link, or fraud-alert call-backs
  • Put your name on outbound caller ID before any outbound list goes live
  • If they do not hold, change the script or stop; no other process depended on it

Expand when all of these are true

  • At least 85 of 100 reviewed status calls end with a correct answer, no person, and no repeat call in 3 business days
  • Zero calls where the AI gave a rate, an eligibility opinion or a reason for a decision
  • Zero account details spoken before the identity check passed, confirmed by a full audit
  • Unanswered and abandoned calls in covered hours cut at least in half against the baseline weeks
07 · What to measure

What to measure, how to define it, and where you probably start.

Published baselines are for all US contact centers, not banks alone, and no independent voice-AI benchmark exists for banks. Measure your own two weeks first and judge the pilot against that.

Metric
Typical today
Planning range with AI and people
Average speed to answer1
Seconds from the end of the IVR to a person or the AI answering, by hour and queue. Include branch phones that ring through.
99 s average, US centers (2024)
AI under 5 s; people near 80/20
Abandonment rate1
Calls that hang up before an answer, divided by calls offered, with calls under 5 seconds excluded the same way every week.
8.9% average, US centers (2024)
Under 5% in covered hours
AI resolution rate
In-scope AI calls where the stated reason was handled and there was no repeat contact on the same matter within 3 business days.
No bank benchmark; measure for 2 weeks
55 to 75% of in-scope calls by week 8
First-call resolution (all calls)5
Share of issues solved on the first call. Choose one method, survey or repeat-call logs, and keep it fixed.
71% by survey method (SQM)
3 to 6 points up on in-scope types
Hand-off quality
Of 50 sampled hand-offs a month, the share where the person did not ask the caller to repeat the reason or identity.
Planning range: measure by listening
90% or better
Dispute notice-to-case time
Minutes from the caller's first report to a time-stamped case with date, payee and amount. The Regulation E clock needs this stamp.
Planning range: varies by desk
Under 5 minutes, every case stamped
Outbound answer rate2
Outbound calls answered by a person, divided by calls placed, split by whether caller ID shows your institution's name.
63% missed a bank call, unknown number (2021)
Set from your own first 500 calls
After-call work time1
Wrap-up minutes per human-handled call, measured before and after AI Assist drafts notes and the disposition.
13.7% of agent time is wrap-up (2024)
One-third less on AI Assist 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 a mid-sized credit union or community lender might see.

Change every number below to your own. The page calculates results from these inputs; they are planning assumptions for one example institution, not a promise.

A worked example. Move the sliders to match yours.

4.5 minaverage call length
$5,335loaded monthly cost per person
40 a daycalls one person handles
$180 · 12%value of one loan or account application started from a recovered call · share of recovered calls that become one
Estimated monthly gain
$50,493a month
Today, per call
$5.33
With VoiceGenie
$2.59
1,080more calls answered a month
130extra loan or account application started from a recovered calls a month
$23,328value of those outcomes
7,800calls 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

  • 12,000 calls a month is a planning size for a mid-sized credit union contact center; replace it with your own call log.
  • 35 percent of calls finished by the AI counts only the call types the split table assigns to the AI. No independent bank benchmark exists.
  • Cost per person is the BLS median CSR wage of $21.53 an hour (May 2025), loaded by about 1.43 for benefits, over 2,080 hours, before seats and supervision 6.
  • Missed share of 9 percent follows the 8.9 percent average abandonment rate in US centers 1. Use your own, including branch phones.
  • Outcome value of 180 is a planning figure: share of applications that fund times first-year margin. Use your own pull-through.

What the dollar figure leaves out

  • Fraud and dispute reports are time-stamped within minutes, even at 9:20 on a Monday
  • Loan officers start each call knowing what the caller needs and what is missing
  • Staff spend more of the day on disputes, hardship and relationships, less on 'where is my application?'
09 · Rules to get right

The rules that touch a phone call in financial services.

This is a practical map, not legal advice. Your compliance officer and counsel decide how each rule applies to your charter and products. Ask the same questions of any voice vendor.

US

Regulation E: oral notice and the dispute clock7

A notice of error can be oral. Investigations run 10 business days, or 45 days if you give provisional credit. Log and time-stamp every phone report the same day as good practice. Written confirmation within 10 business days can be required.

US

Regulation Z: credit card billing errors8

A credit card billing error notice must be written and received within 60 days of the statement. Acknowledge in 30 days; resolve within two billing cycles, 90 days at most. The AI should send your form or link, not say a phone call alone starts this clock.

US

FFIEC authentication guidance (2021)9

Do not rely only on knowledge-based questions. Use stronger controls for credential resets, such as a one-time code to a known device or a call-back to a pre-established number. The guidance names call center staff as a social-engineering target; treat an AI agent the same way.

US

TCPA on outbound calls and fraud-alert exemptions10

Marketing and most servicing calls need consent. Narrow 2015 exemptions cover fraud and breach alerts: free to the recipient, 3 per event over 3 days, about a minute, no marketing, opt-out. Collection is excluded. AI voices count as artificial [s17].

US

Regulation B: adverse action reasons11

Notice of a credit decision is due within 30 days of a completed application, with specific reasons; a failed score alone is not enough. The AI must never state, hint at or explain a decision. The CFPB withdrew its 2022 algorithm circular in 2025; the rule still applies.

US

FTC Safeguards Rule (non-bank lenders)12

Non-bank lenders must oversee service providers by contract, require multi-factor authentication for access to customer data, and report certain breaches of 500 or more consumers within 30 days. Banks and credit unions follow their own regulators' parallel rules.

Four questions to ask any voice AI vendor, including us

  1. Do you use our call audio or transcripts to train any model, and what does the contract say?
  2. How are the AI disclosure and the recording notice played, and can we change the wording by state?
  3. Where is call data stored, how long is it kept, and can we delete it on request?
  4. Show me the log of every record the AI read or changed on a call, with times, for audit.
10 · Where it plugs in

The systems the AI reads from, writes to, and hands over to.

VoiceGenie connects to CRMs and calendars directly, and to other systems through the open API, post-call webhooks or Zapier. These are the systems usually involved at a bank, credit union or lender.

Core and digital banking

  • FIFiserv
  • JHJack Henry
  • FIFIS
  • SYSymitar
  • COCorelation
  • Q2Q2
  • ALAlkami

Loan origination and servicing

  • ENEncompass
  • MEMeridianLink
  • NCnCino
  • BLBlend
  • IMICE MSP
  • SASagent
  • LOLoanPro

Cards, fraud and disputes

  • CPCard processor consoles
  • FFFICO Falcon
  • DCDispute case management
  • CTCore transaction history

Phones and CRM

  • EPExisting PBX or IVR
  • TWTwilio
  • TETelnyx
  • PLPlivo
  • SASalesforce
  • HUHubSpot
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, for the AI and your staff

A bank does not need one vendor for the dialer, another for the AI agent, another for the agent desk and another for staff tools. VoiceGenie is one platform. You connect your systems and load your knowledge once. Every input then 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, loan system and card tools once. The AI reads status and staff see the same record.

One knowledge base

Write approved answers on hours, due dates and documents once. The AI speaks them and AI Assist shows them to staff.

One recording and QA view

Every AI and human call is scored in one place. A corrected answer or QA finding improves both.

12 · What goes wrong

Five ways bank and lender AI projects go wrong

Letting the AI quote rates, pre-qualify or explain a decline
Keep the AI to facts and status. Rates, eligibility and reasons for a decision come from licensed staff and your decision system, which carries the Regulation B reasons.
Verifying identity with questions a scammer can look up
Use a one-time code to the number on file, or a call-back, for anything beyond public facts. Treat knowledge questions alone as too weak for resets and money.
Not time-stamping the first report of a disputed charge
Record the minute of the caller's notice in the case and read it back. A late or missing stamp is a Regulation E problem you created for yourself.
Cutting staff on projected containment before measuring it
Hold headcount flat for the pilot. In 2025 one large Australian bank reversed 45 call center job cuts after the AI did not reduce the need for people [s15]. Decide staffing only after eight weeks of resolution data.
Building a doom loop with no easy way to a person
Offer a person on the first request and after two misses. The CFPB documented customers stuck in repeating bot loops [s8]. Measure repeat contact in 3 days, not containment.
13 · Questions

Questions people ask before they start

Can AI answer phone calls for a bank or credit union?

Yes, for the right calls. An AI agent can answer on the first ring, check identity, read status from your systems, send a text link and hand a call to a person with the full record. It should not make credit decisions, give advice or handle a scam in progress. Those go straight to your staff.

Which bank calls should stay with a person?

Credit decisions and the reasons behind them, scams in progress, hardship and loss mitigation, estates and powers of attorney, complaints, and anything where the caller is distressed. The AI can still answer, identify the caller and pass along a summary, so your staff start informed. They just make the call themselves.

How do you verify a caller's identity on an AI call?

Avoid relying on questions a scammer can answer. Federal regulators say reliable verification generally does not depend only on knowledge-based questions. Use a one-time code to the phone on file, a call-back to a known number or a person for high-risk actions. Public facts, such as branch hours, need no check.

Does a phone call about a disputed charge start the Regulation E clock?

For debit and other electronic transfers, yes. A notice of error can be oral, and the institution may ask for written confirmation within 10 business days. Credit card billing errors under Regulation Z need a written notice. Build the AI script to time-stamp every report and send the right form. Confirm the wording with counsel.

Will callers hate talking to an AI about their money?

Some will, so design for them. Gartner reported in 2026 that 87% of customers say access to a human is essential when companies use generative AI, and a 3,566-customer survey found only 27% would try a chatbot again after a bad experience. Say it is an AI, keep the call short and make a person one request away.

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

Do both in the first sentence. AI disclosure rules vary by state and are changing, and recording consent differs across states, so one opening that says both is the safest design. Outbound AI voice calls are also treated as artificial-voice calls under FCC rulings, so consent and opt-out handling matter.

How much does it cost, and how long does it take to start?

The AI is $0.14 per minute, billed per second, with the phone line extra and no minimum commitment. Human seats run $29 to $99 per person per month on annual billing. A careful pilot takes about 12 weeks, because most of the time goes to identity checks, compliance review and testing, not the technology.

Does it work with our core system and loan origination software?

VoiceGenie connects to HubSpot, Salesforce, Zoho and Google directly. Core banking, loan origination and card systems connect through the open API, post-call webhooks or Zapier. Start read-only, such as application status, and add actions like card locks only after your security team has reviewed the permissions.

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, US Contact Center Decision-Makers' Guide, 17th edition (2025).
  2. First Orion, Branding Financial Calls: 2022 Financial Services Customer Communications Report (2022).
  3. Federal Trade Commission, FTC data on imposter scams, 2025 losses (2026).
  4. Deloitte Insights, AI-assisted customer service in banks (2026 Global Contact Center Survey) (2026).
  5. SQM Group, What Is a Good FCR Rate? (2022).
  6. US Bureau of Labor Statistics, Customer Service Representatives, Occupational Outlook Handbook (2025).
  7. Federal Reserve, Consumer Compliance Outlook, Error Resolution Procedures Under the Electronic Fund Transfer Act and Regulation E (2025).
  8. Consumer Financial Protection Bureau, 12 CFR 1026.13 Billing error resolution (Regulation Z) (2026).
  9. FFIEC via Federal Reserve, Authentication and Access to Financial Institution Services and Systems (2021).
  10. Smith Debnam Narron Drake Saintsing & Myers (law firm summary of the FCC 2015 order), The FCC TCPA ruling: what financial institutions need to know (2015).
  11. Cornell Legal Information Institute, 12 CFR 1002.9 Notifications (Regulation B) (2026).
  12. Federal Trade Commission, FTC Safeguards Rule: What Your Business Needs to Know (2026).
  13. Consumer Financial Protection Bureau, Chatbots in consumer finance (2023).
  14. Gartner, Gartner Survey Finds Only 27% of Customers Would Try a Chatbot Again After a Negative Experience (2026).
  15. BankInfoSecurity, Australian Bank Backtracks on AI-Led Job Cuts (2025).
  16. Cooley LLP, FCC: AI-generated robocalls illegal under the TCPA (analysis of FCC 24-17 declaratory ruling) (2024).