Productivity and AI
It is the most consistent finding of recent surveys, and the most uncomfortable one: 80% of the respondents to McKinsey's 2026 global survey report that AI has improved their individual productivity, but only 37% attribute an effect on their company's earnings to it, a share unchanged over a year. The time people save therefore does not automatically become value for the organisation. This page explains what separates the two, with sourced figures and the legal framework applicable in France today.
18% of French companies with 10 employees or more were using at least one AI technology at the start of 2025, against 6% two years earlier (INSEE, July 2026). But the gap widens with size: 15% among companies with 10 to 49 employees, 58% above 250. On the scope of very small companies and SMEs, the France Num 2025 barometer measures 26% of users, a figure that has doubled in a year.
OpenAI measures 40 to 60 minutes saved per day of use among the users of its enterprise offering, a self-reported figure collected by the vendor from its own customers. An academic measurement published in the Quarterly Journal of Economics in 2025, covering 5,172 customer support agents, establishes an average gain of 15% in cases handled per hour, heavily concentrated among the least experienced people.
Because plugging AI into an unchanged process produces individual comfort, not a measurable result. In McKinsey's 2026 survey, only 6% of respondents belong to the organisations it ranks as high performers on AI. What sets those apart is identified: nearly three-quarters of them have fundamentally redesigned their work processes, against one-quarter of the others.
The main French obstacle is neither price nor technology: 71% of the companies that do not use AI simply say they see no use for it (INSEE, 2025). The method therefore starts by listing the frequent, time-consuming and repetitive tasks, then by costing them in hours per month.
A tool that needs the case explained to it again at every request eats part of the time it saves. Syniaps is plugged into your documents, your emails and your calendar, and that memory belongs to the company: it serves everyone, and it outlives the departure of any one person.
This is the factor McKinsey identifies as the most decisive, ahead of twenty-four others tested. Automating the writing of a follow-up changes nothing if the follow-up process stays manual from end to end. What changes the accounts is handing over the whole mission, from spotting the unpaid invoices to the approved send.
The CNIL points out that it is the using organisation that bears legal liability if its staff misuse AI. In Syniaps, every action that commits the company goes through an explicit approval in the conversation, and every step stays traced.
In practice
Take one precise task, not the whole company. Count the number of people involved, the time they spend on it each week, and the full hourly cost. A team of ten people who each devote three hours a week to searching, summarising and following up represents about 130 hours a month. Then measure the real time on that single task for a month, before and after. That is the only figure that belongs to you, and the only one that counts: published averages help decide where to start, never to justify a budget.
List twenty to thirty repetitive tasks and cost their hours. Pick three of them, including one simple task and one that touches your internal data. Give the AI the context it needs rather than pasting it back every time. Redesign the whole process, not the writing step alone. Measure before and after on the chosen task. Then extend, once the proof is made, and not before.
The last two editions of the McKinsey survey point to the same cause. In 2025, out of twenty-five attributes tested, the redesign of processes was the one that weighed most on the ability to see an effect on earnings. In 2026, nearly three-quarters of the organisations that succeed had fundamentally redesigned their processes, against one-quarter of the others, and the share of high-performing organisations stays flat, at around 6%. The subject is therefore not the power of the model, it lies in the organisation of work around it.
The transparency rules of the European AI regulation have applied since 2 August 2026: a conversational system must tell the person it is an AI, generated content must be identifiable and deepfakes labelled. Eight practices have been prohibited since February 2025, among them emotion recognition in the workplace, which concerns an employer directly. The reinforced obligations applying to so-called high-risk uses, among which the tools for sorting applications and managing employees, will come into force on 2 December 2027. These dates are the ones published by the European Commission, and they have moved: do not rely on third-party summaries.
The CNIL describes them for any company that uses a generative AI system. Frame the use with an internal charter that names the authorised uses and the prohibited ones. Submit only information you are allowed to share, never confidential or personal data in a consumer service. Check the results systematically and never take them as they are. Question the provider about its role, about transfers outside the European Union, and object to the reuse of usage data. Finally, appoint a point of contact and train the users.
Yes, and it is well documented at the individual level. A study published in the Quarterly Journal of Economics in 2025, covering 5,172 customer support agents, measures 15% more cases handled per hour. OpenAI reports 40 to 60 minutes saved per day of use among its enterprise customers, a figure collected by the vendor itself. Be careful about the scope, though: the first measurement covers a single task, and its gain is concentrated among the least experienced people.
Because the time saved scatters if it is not reinvested in a redesigned process. McKinsey's 2026 survey is clear: 80% of respondents observe an individual productivity gain, only 37% an effect on their company's earnings, and that share has not moved in a year. The organisations that manage it, which 6% of respondents belong to, stand apart through the redesign of their work processes.
With the tasks, not with the tools. That is exactly the French sticking point: 71% of the companies that do not use AI say they see no use for it and 54% cite a lack of expertise (INSEE, 2025). List the frequent, repetitive and digital tasks, cost their monthly hours, and start with the one whose result is the easiest to measure.
Since 2 August 2026, the transparency rules of the European AI regulation apply: tell the person they are talking to an AI, make generated content identifiable, label deepfakes. Eight practices have been prohibited since February 2025, among them emotion recognition at work. The obligations on high-risk uses, including the sorting of applications, will apply on 2 December 2027.
The CNIL recommends clearly defining the authorised uses and the prohibited ones, stating that nothing confidential or personal should be submitted to a consumer service, requiring the results produced to be checked, and naming a point of contact. The reason is simple: it is the using company that bears legal liability if its staff misuse AI.
18% of companies with 10 employees or more at the start of 2025, against 6% two years earlier, according to INSEE, with a marked gap by size: 15% between 10 and 49 employees, 58% from 250 upwards. On the broader scope of very small companies and SMEs, the France Num 2025 barometer measures 26% of users, a figure that has doubled in a year. The uses remain mostly conversational: only 5% of very small companies and SMEs report automating tasks.
No, and it is even the opposite of a good method. The first useful commitment is a measurement: the time actually spent on one precise task, before and after, over a month. At Syniaps, plans start from 165 € excl. VAT per month with a free 30-day trial and no commitment, which is amply enough to obtain that measurement before extending anything.
Free 30-day trial, no commitment. Nothing to install, nothing to learn.
Start the free trial