The ledger

What happened after organisations cut roles for AI.

The cases

22 cases: 14 too early to tell, 4 holding, 2 strained, 1 partly reversed, 1 reversed.

Hidden work recorded as missing:

CompanyChangeReported outcomeCase
DNBNorway · Bank technologyAbout 400 roles cut as AI agents take on tasks.Too early to tellLatest news 6 October 2026.Read case on DNB
KonectaItaly · Customer service670 more roles to go in Italy as automation cuts volumes.Too early to tellLatest news 22 September 2026.Read case on Konecta
TikTokIreland · Content moderation300 net roles set to go as moderation work is automated.Too early to tellLatest news 26 August 2026.Read case on TikTok
Le ProgrèsFrance · Copy editing and layout60 posts to be cut as AI tools take on editing and layout.Too early to tellLatest news 16 October 2026.Read case on Le Progrès
Sparebanken NorgeNorway · Retail banking100 to 120 posts to go as AI takes over some tasks.Too early to tellLatest news 11 June 2026.Read case on Sparebanken Norge
Nielsen ItalyItaly · Audience measurement42 of 136 staff to be dismissed as AI takes data work.Too early to tellLatest news 11 April 2026.Read case on Nielsen Italy
INGNetherlands · Money-laundering checks1,250 jobs cut, many in anti-money-laundering checks.Too early to tellLatest news 20 March 2026.Read case on ING
TeleperformanceFrance · Customer service3,300 roles to go in an AI-linked restructuring.StrainedUnions denounce an 'overuse' of artificial intelligence at the Le Mans site.Read case on Teleperformance
IzziMexico · Customer service980 roles cut as AI takes over telephone customer service.Too early to tellLatest news 14 May 2026.Read case on Izzi
DineroDenmark · Customer support10 support roles not refilled as AI takes enquiries.Too early to tellLatest news 23 February 2026.Read case on Dinero
Mercado LibreArgentina · UX writing119 UX and content roles cut as designers use AI to write.StrainedUsers report trouble with simple problems until they reach a human.Read case on Mercado Libre
CCFFrance · Bank operations1,250 roles cut; CEO says AI removed back-office jobs.HoldingSavings of 141 million euros in 2025, targeting about 200 million by 2028.Read case on CCF
Allianz PartnersGermany · Customer service1,500 to 1,800 roles cut as AI takes on calls and claims.Too early to tellLatest news 8 July 2026.Read case on Allianz Partners
ABN AmroNetherlands · Bank support work5,200 net roles cut by 2028 as AI takes on some tasks.HoldingFTE decrease since end-2024 about 40% of the 2028 target.Read case on ABN Amro
SalesforceUnited States · Customer supportSupport headcount fell from 9,000 to about 5,000.Too early to tellLatest news 3 February 2026.Read case on Salesforce
AtlassianAustralia · Customer service150 support roles cut; AI self-service in contact form.HoldingA spokesperson says customer experience improvements reduced support needs.Read case on Atlassian
CBAAustralia · Customer service45 roles cut as AI voice bot takes calls; later reversed.ReversedCBA said its assessment was in error and the 45 roles were not redundant.Read case on CBA
IBMUnited States · Human resourcesHR partners fell from 1,500 to 350 as AskHR took over.HoldingNPS scores bumped up to +35 and now sit at +74.Read case on IBM
TelenorNorway · Customer serviceJust over 100 roles cut; Rørvik customer centre closed.Too early to tellLatest news 10 October 2024.Read case on Telenor
KlarnaSweden · Customer serviceChat assistant took on work of 700 employees.Partly reversedKlarna plans a recruitment drive so customers can reach a person.Read case on Klarna
DuolingoUnited States · Content and translationAbout 10% of contractors cut as GPT translates sentences.Too early to tellLatest news 9 June 2025.Read case on Duolingo
DukaanIndia · Customer service23 of 26 support roles cut as an AI chatbot took over.Too early to tellLatest news 14 July 2023.Read case on Dukaan

22 cases in the ledger. How the ledger is kept

What the ledger watches for

When an organisation announces cuts linked to AI, these are the signs, months later, that the hidden work went missing. The ledger looks for them each time it checks a case again. A case that holds up shows the opposite.

  1. Complaints rise, or move somewhere new.

    To regulators, ombudsmen, social media or the press, where the organisation counts them as something else.

  2. Customers come back about the same problem.

    Repeat contacts show answers that closed a conversation without solving it.

  3. The people who stayed report more pressure, or leave.

    The hardest cases have to land somewhere. Turnover among them is the cost arriving.

  4. The organisation hires again.

    For similar roles, offshore, through contractors, or under a new title.

  5. A regulator, ombudsman or union raises concerns.

    About service, about how decisions are made, or about how the change itself was made.

  6. The measure that justified the change goes quiet.

    It stops being reported, or is replaced by another one.

Saved now, lost later

An illustration of the pattern, not data. The saving is counted at once. The costs arrive later and are booked somewhere else, so they are rarely set against it.

WhatAt onceAfter 3 monthsAfter 6 monthsAfter 12 months
Counted in the business caseThe saving on the visible task
Not counted, booked elsewhereRepeat contactsHard cases pile upComplaints, elsewherePressure on those who stayedRehiringQuestions from a regulator

Statuses and follow-up

A case is a record that keeps changing. Each one is checked again on a schedule set by its status, and its page says when it was last checked, even when nothing new was found.

StatusMeaning
Too early to tellThe cut is recent, or nothing has been reported since
HoldingReports say the change works as intended
StrainedReports of problems: complaints, overtime, quality falling
Partly reversedSome roles or human options restored
ReversedThe company undid the cut

Too early to tell and strained cases are checked after three months. Holding cases after six months, then every year. Partly reversed cases after six months. Reversed cases once more after a year.

What is known beyond the ledger

The ledger records named cases with sources. Around it, two surveys and a forecast are widely quoted. Here is what they say, and where they come from.

Forrester’s Predictions 2026: The Future of Work, published in October 2025, forecast that about half of AI-attributed layoffs would be reversed through quiet rehiring, at lower pay or offshore.

Source: Resultsense, 30 October 2025 

A 2025 survey by Orgvue, which sells organisational design and planning software, asked more than 1,000 business leaders. About 4 in 10 said they had laid off employees as a result of deploying AI. Of those, 55 percent said they had made the wrong decisions about it.

Source: HR Dive, 29 April 2025 

A 2026 survey for SHL, which provides talent assessment to recruiters and employers, asked 2,036 employed adults in the United States and the United Kingdom in June and July 2026. Nearly one in five in the US (19 percent) and about one in eight in the UK (13 percent) said they had first-hand experience of being replaced by AI and later asked back. The release gives what workers would want in order to return, but does not say whether those shares are of everyone asked or only of those asked back.

Source: SHL, 29 October 2026 

Surveys and forecasts are not cases. A survey reports what people say, a forecast is a judgement about the future, and some come from companies that sell tools for the decision they describe. The ledger counts only what named organisations did, from sources anyone can check.

The data

Download the whole ledger as CSV or JSON, or follow new cases and updates with the RSS feed.

The cases are free to reuse under CC BY 4.0 , with credit to The Doorman Fallacy. To cite the idea itself, use the citation on the idea page.

For press and research questions, write to hello@thedoormanfallacy.com.