Voice AI and reception jobs: what the ILO, the OECD and France’s Treasury really say

In a research brief of May 2025, the International Labour Organization (ILO) updates its 2023 index of exposure to generative AI. One worker in four worldwide holds a job exposed to some degree, and the most exposed category (gradient 4) accounts for 3.3% of global employment. The ILO writes that clerical occupations continue to show the highest levels of exposure and that, in high-income countries, 34% of employment falls within one of the four gradients, compared with 11% in low-income countries.

The organisation insists on what the figure does not say: it measures potential exposure, a theoretical assessment of what the technology can do, not its actual application, and it represents a ceiling. It concludes that few jobs are currently at risk of full automation and that, because of the continuing need for human intervention, most will be transformed rather than eliminated, with the transition to be guided by social dialogue.

In its Employment Outlook 2023, the OECD estimates that the occupations most exposed to the risk of automation account for about 27% of employment on average in the sample of countries studied. The scope is precise: it is the top quarter of occupations ranked according to an expert survey covering 98 skills and abilities, applied to 2019 employment shares, and the risk measured covers AI but also other automation technologies such as robotics. The jobs most at risk are mainly low- or medium-skilled (construction, agriculture, production, transport), not office jobs specifically. The OECD also notes that three out of five workers surveyed in its surveys in finance and manufacturing fear losing their job to AI within the next ten years.

On observed effects, the same publication states that there is little evidence at this stage of a significant negative effect of AI on employment, while calling for caution: these data do not take into account recent progress in generative AI, which could extend the tasks that can be automated, and any negative effects could take time to materialise. The data predate the rise of generative AI: they should be read as a starting point, not a verdict.

Note no. 391 of the Direction générale du Trésor (DG Trésor, the French Treasury’s directorate), published in June 2026, lists estimates of jobs exposed to AI of between 5% and 60% depending on the studies, which illustrates the uncertainty of the measurement. It specifies that so-called threatened jobs are those in which a significant share of tasks can be substituted by AI, without presuming that they will disappear. It states that empirical studies cannot determine the total effect of AI on employment, for lack of hindsight and because adoption is still limited: about 20% of European Union companies had adopted it in 2025, according to the Global AI Adoption Index cited by the note.

The note also reports localised signals: since 2025, announcements of job cuts attributed to AI have multiplied in the United States and in France, even though AI sometimes serves as a pretext for other motives, and young people’s entry into the labour market appears to be slowing. Finally, it notes that 62% of French people see AI as a risk for employment, according to the 2025 digital barometer (baromètre du numérique), and cites experimental studies showing productivity gains ranging from +14% for customer service agents to +26% for developers, specifying that these results could be underestimated, most having been obtained with AI models that are now out of date.

These sources talk about tasks more than jobs. The ILO cites, among the tasks that benefit from generative AI, taking notes in meetings or scheduling appointments. Capturing a request, transcribing or summarising a call are tasks of the same nature, though these sources do not measure them precisely. What remains hard to automate is listening to a caller in difficulty, judging an unusual case, the relationship of trust with a regular customer.

The effects on employment will depend on how organisations use it: relieving reception of out-of-hours calls and repetitive requests to free up quality time, or cutting staff. The ILO stresses that job quality can improve or deteriorate depending on whether the technology is well integrated or is used to standardise work and increase surveillance. The Treasury note recalls that, in previous technological revolutions, the jobs created offset those that disappeared and the content of occupations was profoundly transformed, and it concludes that public support is needed, notably in training.

Which tasks do you want to delegate, which must stay with people, and how are the reception teams informed and involved? The ILO recommends social dialogue and consultation in the workplace: a rollout that is explained, with training for the new tasks (reviewing reports, handling complex cases), is more lasting than one that is imposed.

Alyas is designed to answer when nobody can, 24 hours a day, from the information you give it; it announces that it is an AI and that the call is recorded, captures the caller’s details, says when it does not know and offers to pass the request on, and transfers to a single number when the caller asks. Transcripts and summaries are placed in your workspace, so that the team can handle the cases that call for a person. Free 10-day trial, with no credit card required.