How much does a French SME lose on automatable tasks in 2026?#

By Laurent Perello, founder of Perello Consulting -- web pioneer for over 25 years, AI operator in production since 2024. Last updated: 13 April 2026.

A French SME with fifteen employees in business services loses, in 2026, between 5,500 and 7,300 euros per month on automatable tasks -- that is 66,000 to 88,000 euros per year. This range does not depend on any American benchmark. It is calculated function by function, from three public ingredients: the hours spent on repetitive tasks, the fully loaded hourly cost of each role (INSEE¹²), and the automatable share of the role (France Strategie³⁴, OECD⁵, Goldman Sachs⁶). This article publishes the formula. You can redo the calculation for your own organisation, with your own figures and your own sources.

Why most executives miss this question#

You know your revenue, your gross margin, your payroll, probably your customer acquisition cost. You almost never know the monthly cost of automatable tasks in your organisation. This is not an oversight on your part. It is the result of three structural biases that make the question invisible to the market.

The first is the opacity of consulting firms. Large integrators speak in aggregate "productivity gains", in "transformation" percentages, in multi-year "roadmaps". None of these units land on a table. The Bpifrance Diag Data IA⁷ sets a public framework for the audit, but the calculation detail remains proprietary to each provider. The result is a quote, not a method.

The second is the average benchmark. A McKinsey⁸ or BCG⁹ study announces global ranges, useful for the press, useless for your decision. If you read "20% to 30% potential gains", you do not know whether that applies to your fifteen-person Paris consulting firm or to a metalworking shop in Champagne. The average hides the variance.

The third is the absence of a public method in French. France Strategie reports³⁴, INSEE indicators¹², CNIL guidance¹⁰ all exist. Nobody has yet combined them into a formula usable by a non-technical executive. This article fills that gap, so that you can redo the calculation at home. The method is public, the formula is explicit, your sources are clickable at the end.

The public calculation formula#

The monthly loss of an SME on automatable tasks is calculated as follows: for each function, multiply the monthly hours spent on repetitive tasks by the fully loaded hourly cost of the typical employee, then by the automatable share of the function. Sum across all functions. Apply a prudential adjustment of thirty to fifty percent. The result is a range.

Formally:

monthly_loss_EUR = SUM_f ( hours_month[f] x hourly_cost[f] x auto_rate[f] ) x (1 - prudential_adjustment)

Each term has a public source.

Monthly hours per function are derived from a self-declaration by the executive, multiplied by 4.33 to convert from week to month. The hour bands used in the Perello Consulting audit form follow the nomenclature of DARES working-conditions surveys¹¹ and France Strategie work on occupations in 2030³.

The fully loaded hourly cost is calculated from the payroll and headcount per NAF sector published in the INSEE ESANE database¹, divided by 1,600 effective annual working hours, with application of an employer loading coefficient (approximately 1.42 on average in the private sector, URSSAF framework¹²). The INSEE ICHTrev-TS index² provides the quarterly control point.

The automatable rate for each function is a triangulation. On the French side, France Strategie³⁴ on AI-driven job transformation. Internationally, the OECD Employment Outlook⁵, the Goldman Sachs report on AI effects on growth ("between 25% and 50% of workload could be replaced"⁶), and the Anthropic Economic Index¹³ which empirically measures the penetration of generative AI in professional tasks.

The prudential adjustment of 30 to 50% absorbs what is not recovered in the first year: unready data, resistance to change, learning costs, necessary human oversight. In the worked example below, we use 40%. This value is not a commercial concession. It is methodological honesty: no deployment achieves 100% of theoretical potential. Ever.

[UNIQUE INSIGHT] Honesty about the prudential adjustment is what distinguishes a publishable figure from a sales slide. Most firms announce the theoretical potential. They then present a quote calibrated on that potential. You discover the gap six months later, when your budget line no longer holds.

The case of a 15-person SME, function by function#

Take a reference SME close to yours: fifteen full-time equivalents, business services sector (consulting), split into one executive, one sales director, two salespeople, two marketing, three delivery, two admin-finance, one HR, one support, two production operations. Average fully loaded hourly cost: EUR 75/h, consistent with the INSEE ESANE¹ range for your sector. The hours reported are per-band medians from the Perello audit; replace them with your actual measurements to refine.

The master table#

Line-by-line reading#

Administration and accounting hold the largest pocket. Invoice processing, follow-ups, bank reconciliations, expense reports, social declarations: one hundred monthly hours at roughly EUR 55/h loaded, half of which can be handed to AI workflows. France Strategie⁴ confirms that administrative functions are among the most exposed to generative AI transformation.

Next comes your sales team. Lead qualification, proposal drafting, CRM updates, meeting notes, post-meeting follow-ups. The hourly cost is higher (EUR 75/h), the automatable rate slightly lower (40%) because your final negotiation stays human. The Anthropic Economic Index¹³ documents the growing penetration of AI on preparatory commercial tasks.

Marketing combines content creation, scheduling, performance analysis, reporting. The automatable rate is high (50%); the hour volume is more modest in a fifteen-person SME. Management, though low in hours, weighs heavily because the hourly cost of the executive and senior staff is high (EUR 130/h loaded).

The retained range#

After applying the 40% prudential adjustment, the recoverable monthly loss sits between EUR 5,544 and EUR 7,308, rounded to EUR 5,500-7,300 per month, or EUR 66,000-88,000 per year, before the cost of the AI solution itself. This figure serves as the reference for sizing an automation project for this type of organisation.

The three main measurement errors#

Over-estimation: the "McKinsey benchmark" effect#

The first error is importing a global figure from an international study and applying it as-is. A McKinsey⁸ or BCG⁹ report announces a productivity gain potential. You multiply by your payroll. You get an impressive number. That number is not wrong in the abstract. It is simply unusable: it is not calibrated to your employment structure, your loaded hourly cost, your data maturity. The slide inflates expectations. The deployment disappoints.

Under-estimation: the blind spot of invisible tasks#

The second error is the opposite. You think of your "visible" functions (admin, sales) and forget the dispersed hours: weekly reporting, meeting notes, monitoring, expense management, internal follow-ups, document updates. These micro-tasks, taken one by one, seem negligible. Aggregated across your fifteen employees, they often weigh as much as official administration. The DARES working-conditions survey¹¹ documents this fragmentation of time on non-core repetitive tasks.

Gross-to-net confusion: the cost of automation#

The third error is announcing the gross gain as if it were a net result. The EUR 5,500-7,300 per month range is a gross gain: the value of time you stop spending. To get to net, subtract the cost of AI licences (EUR 20 to 100 per user per month depending on tools), the initial deployment cost (EUR 5,000 to 50,000 depending on scope, Bpifrance Diag Data IA framework⁷), and the cost of human support. These costs are partly absorbed by the 40% prudential adjustment; they do not disappear from your P&L for all that.

How these figures vary by size#

The multipliers are not linear. A fifty-person SME does not lose ten times more than a five-person micro-enterprise. Support functions (HR, management, IT) grow slower than operational functions. Some automatable gains have a structural ceiling: an SME has already automated payroll, basic accounting, recurring invoicing.

Two useful readings. First, if you run a micro-enterprise, do not try to match mid-cap figures: your cost structure is different, your automation ROI is too. Second, between the 15-person and 50-person SME, the per-head gain increases because your support functions start to justify dedicated tools. This is the bracket where your automation budget becomes structurally profitable.

[ORIGINAL DATA] The cadence observed on our own internal infrastructure (seven AI orchestrators, one human) shows that an organisation designed for automation from the root can hold, at equivalent scope, the figures of a fifty-person SME with the human headcount of a micro-enterprise. This is not a sales argument: it is the reason the firm applies its method to itself before delivering it.

How these figures vary by sector#

The fully loaded hourly cost varies by a factor of three between catering (around EUR 32/h) and consulting (around EUR 110/h), according to INSEE¹ and URSSAF¹² data. For the same recovered hour, the economic value is not the same. The automatable rate also differs, because the nature of tasks changes.

Consulting and tech occupy the top of the range for two converging reasons. First, hourly cost: a senior consultant costs their firm significantly more than a production operator. Second, the nature of tasks: document production, analysis, reporting, proposal writing concentrate precisely what generative AI handles best, with exposure rates confirmed by the Anthropic Economic Index¹³.

Retail and manufacturing are lower in absolute value. Retail because the average hourly cost is modest. Manufacturing because physical tasks remain little affected by generative AI; industrial automation runs through other levers (robotics, MES, sensors) documented separately by the DGE¹⁴ and OECD reports¹⁵.

Why a range rather than a single figure#

Giving you a single figure would be false precision. Four parameters in the calculation carry a margin of uncertainty, and this uncertainty must be made explicit so that the figure remains usable by your auditor, your CFO, your shareholders.

Declared hours carry a margin of approximately +/-25%. When you answer the diagnostic, you estimate, you do not measure. This uncertainty reduces within one to two weeks through spot time-tracking on your relevant functions.

The fully loaded hourly cost carries approximately +/-15% margin, depending on your collective agreement, your seniority structure, your variables, your actual charges. This uncertainty reduces with your actual payroll data, cross-referenced with the INSEE¹² and URSSAF¹² framework.

The automatable rate carries +/-20%. The gap between 40% and 50% on a sales function depends on CRM data quality, call volume, process stability. A 90-day pilot on one function resolves this uncertainty better than any benchmark.

The prudential adjustment carries +/-10 points, depending on starting digital maturity. An already-tooled organisation recovers faster (-30%); a less mature organisation recovers less (-50%). This value stabilises after six months of actual deployment. Before that, any additional precision would be an artefact.

This honesty about uncertainty is what you are entitled to demand from serious costing. It lets you reason on the low bound to commit your budget and on the high bound to aim your roadmap.

What these figures let you decide#

The calculation is not meant to impress. It is meant to arbitrate. Three concrete decisions flow directly from the range.

Prioritise the projects#

You rank your functions by decreasing monthly loss, weighted by data maturity and implementation risk. In the 15-person SME case, administration and sales come out on top. Marketing and management follow. HR and production close the list. This hierarchy is not universal: it changes with your sector, your employment structure, your starting digitalisation. The sorting method stays the same.

Arbitrate the annual budget#

The expected monthly ROI of a project serves as the bound for your deployment budget. A project promising EUR 2,200/month gross gain can reasonably absorb EUR 15,000 to 25,000 in deployment, amortised over twelve months, with recurring costs absorbed in your prudential adjustment. A project promising EUR 220/month justifies only a fraction of that budget. Your arbitrage happens line by line, not in bulk.

Pace the deployment#

You do not automate everything at once. The pacing we recommend follows three stages: one priority project over 90 days (pilot supervised by your teams), extension to two complementary projects over the next 90 days, generalisation over the remaining six months. This sequence minimises your risk of poorly calibrated automation, and it respects the CNIL framework¹⁰ and the European AI regulation¹⁶ which require documented supervision of AI systems deployed in production.

For more on the operational sequence, the Perello Consulting public methodology details the five stages of the flywheel (build, prove, systematise, duplicate, transmit). The daily public log perfectaiagent.xyz records the dated proofs, and the coordination repository VantagePeers publishes the source code under MIT licence.

Frequently asked questions#

Why a range and not a single figure?#

Because four parameters in the calculation carry uncertainty: declared hours (+/-25%), fully loaded hourly cost (+/-15%), the automatable rate per function (+/-20%), and the share of potential actually recovered in the first year (between 50% and 70%). Giving a single figure would be false precision. An honest range is more useful: it lets you reason on the low bound to commit and the high bound to aim. It is also the low bound that feeds the minimum guaranteed ROI calculation.

How do I calculate the figure for my own SME?#

Three steps. First: for each function (admin, sales, marketing, support, HR, production, management), estimate the weekly hours on repetitive tasks, multiply by 4.33. Second: apply the fully loaded hourly cost for each function, from your actual payroll or the INSEE sectoral table¹. Third: multiply by the automatable rate (between 30% and 50% depending on function), sum, subtract a 40% prudential adjustment. The /free-ai-audit form automates these calculations in ten minutes and delivers the report.

What sources do you use and are they public?#

All sources are public and clickable. For hours and employment structure: INSEE ESANE database¹ and DARES surveys¹¹. For fully loaded hourly cost: INSEE², URSSAF¹². For the automatable rate: France Strategie³⁴, OECD Employment Outlook⁵, Goldman Sachs Research⁶, Anthropic Economic Index¹³. For the legal framework: European regulation 2024/1689¹⁶, CNIL recommendations¹⁰ and ANSSI¹⁷. The full bibliography is at the end of this article, with consultation date for each source.

Does the hourly cost vary by sector?#

Yes, significantly. The fully loaded hourly cost ranges from approximately EUR 32/h in catering to approximately EUR 110/h in consulting for a mid-sized SME, according to the INSEE ESANE database¹ and the URSSAF loading coefficient¹². This variation explains why two fifteen-person SMEs, one in retail and the other in consulting, can lose very different amounts on the same hours of repetitive tasks. The sectoral table presented above gives you an order of magnitude by sector.

Why are some tasks more automatable than others?#

A task is automatable when it is repetitive, relies on structured or semi-structured data, follows a reproducible pattern, and does not require contextual judgement with high human stakes. Administration, sales prospection and marketing concentrate these characteristics, with rates of 40% to 50%. Operational production, people management and strategic decisions mix the repetitive and the singular, with lower rates (30% to 35%). These rates come from France Strategie⁴, OECD⁵ and Goldman Sachs⁶ work.

Do these figures include the cost of AI itself?#

No. The range presented is a gross gain -- the value of time you stop spending. To get to net gain, subtract AI licence costs (EUR 20 to 100 per user per month), initial deployment cost (EUR 5,000 to 50,000 depending on scope, Bpifrance Diag Data IA framework⁷), and support costs (training, change management, human oversight). The 40% prudential adjustment applied in the worked example absorbs part of this friction, not all of it.

What happens if I automate badly?#

Poor automation often costs more than no automation. Three typical mistakes: automating a task whose data is not clean (the output degrades quality), automating without a human supervision phase at the start of deployment (errors accumulate silently), automating without GDPR or AI Act compliance (CNIL sanctions¹⁰ possible up to 4% of worldwide turnover under the European regulation¹⁶). A preliminary audit, a pilot on one function, and measured follow-up limit these risks.

Where do I start concretely?#

By measuring. As long as hours and costs are not laid on a table, every automation decision is a hunch. The recommended sequence is four steps: fill in the /free-ai-audit diagnostic (ten minutes, free), receive the costed report (one week), choose the first function to automate based on monthly ROI weighted by data maturity, deploy a 90-day pilot before any generalisation. Who operates it next -- you can meet them on the team page.

Calculate your own figure#

The EUR 5,500-7,300 per month range applies to a fifteen-person SME in business services. Your employment structure, your sector, your digital maturity will change this figure by significant amounts. The only way to get an actionable value is to redo the calculation on your own organisation.

The /free-ai-audit diagnostic takes the formula published above and populates it automatically from your answers. The report is delivered within one week, structured by intervention priority, deployment complexity and costed impact. It is actionable from the day after the meeting, with or without continuing the relationship. To check beforehand that the firm actually operates its method on itself, you can browse the public log https://perfectaiagent.xyz.

The full methodology details what keeps this costing in production: three-tier architecture, dated public log, non-captive source code. The team page presents the ten orchestrators who operate the method.

Request your AI audit -> -- free, one week, costed report, no strings attached.

Read also#

• Where to start with AI automation when you run an SME in 2026?
• What are the real risks of poor AI use in a French SME?

Sources and methodology#

This article is dated 13 April 2026. Sources were consulted on that date. A revision is scheduled every six months to integrate the most recent editions of the cited reports. All URLs below are clickable and public. Specific figures used in the body of the article refer by superscript to the corresponding entry.

Main French sources

International sources

About the author#

Laurent Perello runs Perello Consulting, an independent AI automation firm for French SMEs. After 25 years building products for the web, he now orchestrates ten AI agents that he pilots alone, with a production log published daily at perfectaiagent.xyz. He publishes his methodologies and pricing online so that every executive can decide with full information.

Orchestrator: Alpha -- Perello Consulting | 2026-04-17