Named reversals, with dates and dollar amounts

Twelve documented reversals in five years. Two of them put humans back. Public embarrassment gets the coverage, cost and volume get the reversal.

The short answer

What does the research show about Named reversals, with dates and dollar amounts?

Twelve documented reversals in five years. Two of them put humans back. Public embarrassment gets the coverage, cost and volume get the reversal. Each case below was verified at a named news source or the company's own statement. Cases that could not be substantiated were dropped rather than softened.

Evidence

  • Klarna: the assistant was marketed as doing the work of 700 agents. By Q3 2025 customer service and operations cost had risen to $50M from $42M a year earlier, settling into a hybrid model with AI on about two-thirds of inquiries.
  • Air Canada: the tribunal rejected the argument that the chatbot was a separate legal entity and ordered C$812.02 in total.
  • Amazon: a recruiting engine built from 2014, 500 models and roughly 50,000 resume terms, penalized resumes containing women's and was disbanded by the start of 2017.

Supporting pages

Each case below was verified at a named news source or the company's own statement. Cases that could not be substantiated were dropped rather than softened.

These are documented cases, not a representative sample. The base rate of reversal is measured in Theme 3, not here.

What this page is

Twelve verified cases where a company deployed AI in a customer-facing or hiring role and then reversed the decision, with dates, dollar amounts, and a named source on each.

The argument

The reversals that get coverage are the embarrassing ones. The reversals that actually happen are driven by cost and volume, and the correction is usually a hybrid model rather than a retreat.

How to read it

  • This is a documented case list, not a representative sample. The base rate lives in the rollback theme, measured on 2,527 decision makers.
  • Cases that could not be substantiated at a named source were dropped rather than softened.
  • A reversal is not the same as a failure of the category. Two of the twelve put humans back. The rest changed scope.

Ten documented reversals in five years. Two put humans back. Three companies held the line.

Lanes separate the kind of event, because a shut-off drive-thru and a rehired support floor are not the same thing. The three that held are in the same frame on purpose.

Rolled back or shut off

  • Zillow OffersNov 2, 2021

    $421.6M segment loss before tax, ~25% workforce reduction

  • Chevrolet of WatsonvilleDec 2023

    Bot talked into a one dollar Tahoe, feature removed

  • DPDJan 2024

    Chat swore at a customer, post viewed 800,000 times in 24 hours

  • McDonald's with IBMBy Jul 26, 2024

    Order-accuracy failures across 100+ restaurants

  • Taco Bell, Yum BrandsAug 29, 2025

    Public rethink after an 18,000 water cup order

Replaced humans, then rehired

  • KlarnaMay 2025

    Cost became a too predominant factor, quality fell, agents rehired

  • Commonwealth Bank of AustraliaAug 21, 2025

    Call volumes rose, 45 redundancies reversed and apologized for

Announced, then walked back the message

  • DuolingoApr to May 2025

    AI-first memo, then I do not see AI as replacing what our employees do

Vendor repositioned

  • 11xMar 24, 2025

    Reported 70 to 80% churn, ARR counted 3-month break clauses as full year

  • ArtisanAug 2026

    Stop hiring humans narrative reversed, then made a splash hiring its first human BDR

Held the line

  • Coca-Cola2024 and 2025

    Backlash on generative advertising, approach repeated

  • IBMMay 2025

    ~200 HR roles to agents, total employment up, savings funded sellers

  • Salesforce2025

    Support from 9,000 to about 5,000, agents on 50% of Service Cloud interactions

What this does not say

This is a collected set of reported cases, not a sample. It cannot tell you what share of all deployments get reversed.

Publisher
Per-case verification at named publishers and company statements
Sample and method
Includes Klarna Q3 2025 earnings, the CBA reversal of August 21 2025, the Air Canada tribunal decision of February 16 2024 at C$812.02, and the TechCrunch investigation of 11x
Field dates
Not published by the source.

High confidence on the individual cases. Low as a base rate. This is a collected set of reported cases, not a sample.

High confidence

Public embarrassment gets the coverage. Cost and volume get the reversal.

n=10 documented reversals. Categories assigned by us from the reporting.

Documented cases in this set

  • Output was wrong in public4

    Air Canada, DPD, McDonald's, Chevrolet

  • Volume or cost went the wrong way2

    CBA, Klarna

  • Employees or customers rejected it2

    Duolingo, Coca-Cola

  • System broke under adversarial input2

    Taco Bell, Chevrolet

  • Model encoded bias1

    Amazon recruiting engine

What this does not say

The category counts describe what was disclosed, not what happened. Companies choose which failure they name.

Publisher
The Revenue AI Report categorization of verified cases
Sample and method
n=10 documented reversals plus 3 companies that held the line, 2021 to 2026. Categories are ours, sources are per case
Field dates
Not published by the source.
Low confidence
Public embarrassment gets the coverage. Cost and volume get the reversal.

n=10 documented reversals. Categories assigned by us from the reporting.

Public embarrassment gets the coverage. Cost and volume get the reversal.
Documented cases in this setValueNote
Output was wrong in public4Air Canada, DPD, McDonald's, Chevrolet
Volume or cost went the wrong way2CBA, Klarna
Employees or customers rejected it2Duolingo, Coca-Cola
System broke under adversarial input2Taco Bell, Chevrolet
Model encoded bias1Amazon recruiting engine

Source: The Revenue AI Report categorization of verified cases. n=10 documented reversals plus 3 companies that held the line, 2021 to 2026. Categories are ours, sources are per case Confidence: Low.

What this does not say: The category counts describe what was disclosed, not what happened. Companies choose which failure they name.

Also in the record

Figures that sit alongside these charts.

  • Klarna: the assistant was marketed as doing the work of 700 agents. By Q3 2025 customer service and operations cost had risen to $50M from $42M a year earlier, settling into a hybrid model with AI on about two-thirds of inquiries.
  • Air Canada: the tribunal rejected the argument that the chatbot was a separate legal entity and ordered C$812.02 in total.
  • Amazon: a recruiting engine built from 2014, 500 models and roughly 50,000 resume terms, penalized resumes containing women's and was disbanded by the start of 2017.
  • Not every disclosure is a reversal. IBM and Salesforce both held their positions and reported redeployment rather than retreat.

The brief

What is going on here, and why it matters.

The charts above are the evidence. This is the read: what the data describes, the mechanism behind it, where the argument could be wrong, and what a revenue team does about it.

01

What the cases have in common

Klarna is the reference case because it has numbers on both sides. The assistant was marketed as doing the work of 700 agents. By the third quarter of 2025, customer service and operations cost had risen to $50 million from $42 million a year earlier, and the company settled into a hybrid model with AI handling roughly two-thirds of inquiries. The reversal was partial and it was driven by economics, not by outrage.

Air Canada is the reference case for liability. The tribunal rejected the argument that the chatbot was a separate legal entity and ordered C$812.02. The dollar figure is trivial and the precedent is not: the output of your agent is your statement.

02

The trigger nobody plans for

Public embarrassment gets the coverage and cost gets the reversal. That ordering matters for how you brief a board. A risk register built around brand incidents will miss the failure mode that is far more likely to end the deployment, which is unit economics that stop working once volume arrives and the escalation path fills with the hard cases the agent could not close.

Amazon's recruiting engine is the exception that names a third trigger. Built from 2014 with roughly 500 models and about 50,000 resume terms, it penalized resumes containing the word women's and was disbanded by the start of 2017. That reversal came from an audit, which is the only trigger that fires before customers feel it.

03

What the non-reversals say

Not every disclosure is a reversal. IBM and Salesforce both held their positions and reported redeployment rather than retreat. Counting those as reversals would inflate the pattern, and the ledger keeps them separate on purpose.

The honest read is that AI in customer-facing work is settling into a scope question rather than a yes-or-no question. Twelve documented reversals in five years, two of which put humans back, describes a market recalibrating scope, not a market backing out.

What to do with it

The move, by seat.

CX and support leaders
Model the escalation path at full volume before launch. The reversal usually starts when the residual cases, not the easy ones, become the whole queue.
Finance
Track total cost to serve, not deflection rate. Klarna's deflection worked and the cost line still rose.
Legal
Assume the agent's output is a company statement. Air Canada already settled that question.

Questions this page answers

What the data says, in plain language.

What does the research show about Named reversals, with dates and dollar amounts?
Twelve documented reversals in five years. Two of them put humans back. Public embarrassment gets the coverage, cost and volume get the reversal. Each case below was verified at a named news source or the company's own statement. Cases that could not be substantiated were dropped rather than softened.
What does the figure "Ten documented reversals in five years. Two put humans back. Three companies held the line" show?
Lanes separate the kind of event, because a shut-off drive-thru and a rehired support floor are not the same thing. The three that held are in the same frame on purpose. Source: Per-case verification at named publishers and company statements. Includes Klarna Q3 2025 earnings, the CBA reversal of August 21 2025, the Air Canada tribunal decision of February 16 2024 at C$812.02, and the TechCrunch investigation of 11x Confidence: High. Caveat: High confidence on the individual cases. Low as a base rate. This is a collected set of reported cases, not a sample.
What does the figure "Public embarrassment gets the coverage. Cost and volume get the reversal" show?
n=10 documented reversals. Categories assigned by us from the reporting. Source: The Revenue AI Report categorization of verified cases. n=10 documented reversals plus 3 companies that held the line, 2021 to 2026. Categories are ours, sources are per case Confidence: Low.
What else sits alongside these figures?
Klarna: the assistant was marketed as doing the work of 700 agents. By Q3 2025 customer service and operations cost had risen to $50M from $42M a year earlier, settling into a hybrid model with AI on about two-thirds of inquiries. Air Canada: the tribunal rejected the argument that the chatbot was a separate legal entity and ordered C$812.02 in total. Amazon: a recruiting engine built from 2014, 500 models and roughly 50,000 resume terms, penalized resumes containing women's and was disbanded by the start of 2017. Not every disclosure is a reversal. IBM and Salesforce both held their positions and reported redeployment rather than retreat.
Where does this data come from?
Every figure is reproduced from a named publisher: Klarna, reversal coverage and Q3 2025 earnings, Air Canada tribunal decision, TechCrunch investigation of 11x, Zillow Offers wind-down. Sample, field date, and confidence are shown on each chart. Sources marked as vendor research are labelled on the page.

Cite this page

Permanent URL and suggested citation.

https://www.therevenueaireport.com/research/named-reversals

Kvarfordt, Jonathan. "Named reversals, with dates and dollar amounts." The Revenue AI Report, Research Library. https://www.therevenueaireport.com/research/named-reversals

Figures on this page are reproduced from the publishers listed below. Cite the original publisher for the underlying data, and this page for the compilation and framing.

Sources

Every publisher used on this page.

If a metric, model term, or method on this page is unfamiliar, every one of them is defined in The AI and Revenue Dictionary. Sample size, field date, and confidence tags are explained there too.

Could not confirm

What we looked for and did not find.

Claims found during research and not charted

Nothing on this theme was dropped for sourcing. Every claim we found that met the standards on the Research hub is charted above, and anything that failed them would be listed here by name.

The live record

Every case on this page also lives in The Reversal Ledger, a filterable row-per-case record with a permanent anchor on each reversal so a single case can be cited on its own.

Read the analysis

Issues built on this theme.

Research on this site is the evidence layer. These essays take the numbers above and apply them to real decisions, so you can see how the data reads in practice.

  • 11x After the Walk-Back: What Buyers Should Ask Now

    The autonomous SDR category had a public credibility event and the vendor is still selling. Here is the documented timeline, what changed, and the reference-check script to run before any AI SDR contract reaches signature.

How to cite this research

Written by Jonathan Kvarfordt, Founder and Principal Analyst, The Revenue AI Report. Published under CC BY 4.0.

APA

Kvarfordt, J. (2026). Named reversals, with dates and dollar amounts. The Revenue AI Report. Retrieved from https://www.therevenueaireport.com/research/named-reversals

MLA

Kvarfordt, Jonathan. "Named reversals, with dates and dollar amounts." The Revenue AI Report, 31 Aug. 2026, www.therevenueaireport.com/research/named-reversals.

BibTeX

@misc{kvarfordt2026namedreversals,
  author = {Kvarfordt, Jonathan},
  title = {Named reversals, with dates and dollar amounts},
  year = {2026},
  publisher = {The Revenue AI Report},
  url = {https://www.therevenueaireport.com/research/named-reversals}
}

Share this research

Posting to Instagram or TikTok? Copy the link, it carries the title, summary and share image.

Next theme

What separates the deployments that work

The largest gap between AI leaders and everyone else is not technology. It is having decided what to build.

Subscribe

Get the weekly issue built on this data.

Arrives weekly by email. Free. Unsubscribe anytime. By subscribing you agree to our Privacy policy and Terms. We never sell or share the list.