AI leverage: can you really multiply productivity by 10?
No. A 10x gain in the overall productivity of a person or a company is not a result documented by public studies. Measured gains range from 14 to 55% on specific tasks; a 1 to 10 ratio exists on some tightly scoped steps (a first draft, a sorting job), rarely on a whole job.
That does not mean AI leverage is imaginary. It is real, but it is built process by process, and it depends more on how people work than on the tool they pick. This article reviews the public figures, explains where "10x" promises come from, and describes how to create measurable leverage in an SME. For the big picture, see our complete guide to AI in business for SMEs.
Can you really multiply productivity by 10 with AI?
On a single task, sometimes; on a job or a company, no official source shows it.
The "10x" almost always comes from comparing the time of one step (writing a first draft, summarizing 40 pages) with the time it took before. Meeting minutes that took 45 minutes and now take 5 after review are indeed a 1 to 9 ratio on that step. But that step may be 3% of a project manager's week. The overall gain is then a few points.
The point to remember: a job's productivity is capped by the tasks AI does not touch (decisions, client relationships, coordination, checking). Speeding up one task tenfold does not multiply the final output by ten if the other steps stay the same. Economists call this a bottleneck: the slowest step sets the pace.
What do studies say about real gains?
Measured gains are meaningful but bounded: 14% to 55% depending on the task, and no statistically significant gain yet at company level.
The French Treasury (Direction générale du Trésor, Trésor-Éco no. 341, April 2024) surveyed the available research:
| Situation studied | Measured gain |
|---|---|
| Programmers with a coding assistant | 55% for writing code |
| Professionals on basic writing tasks (applications, summaries) | 37% on average |
| Customer advisers with a conversational agent | 14% on average, concentrated among the least experienced |
| Consultants on creative tasks | 43% for the least productive, 17% for the most productive (proposal quality) |
The same publication notes that effects on company productivity remain modest and not statistically significant at this stage. This research dates from 2023 and 2024: models have improved since, but no official French source publishes a 10x figure.
On French firms, the Banque de France (Bulletin no. 266, 3 September 2026) reports that 67% of companies with 20 or more employees use generative AI, mostly in a limited or experimental way. See our article on AI productivity in SMEs: 2026 key figures. Note the gap: usage is widespread, company-level gain is modest.
Why does individual gain not become company gain?
Because saved time dilutes: without explicit reallocation, an hour saved turns into messaging, not output.
Three mechanisms show up in the field:
- Review. A text generated in 30 seconds takes 10 minutes to check if it goes to a client. The net gain is the difference, not the generation time.
- Individual use. Everyone uses their own prompts and habits. Quality varies by person and nothing is capitalized.
- No baseline. Without a reference time, nobody can say what changed, and the leverage remains an impression.
Companies' own stated obstacles say the same: lack of skills and lack of use cases come before cost. Our article on mistakes to avoid when deploying AI in SMEs covers the most common traps.
Where is the leverage strongest?
On repetitive, well-bounded, easy-to-check tasks, and above all when they are chained in a workflow rather than handled one by one.
What we see in the SMEs we train (GrowthPerf field observation, not a published study): leverage is clearest in four families of tasks.
- First drafts of standard documents: quotes, minutes, replies to recurring requests.
- Summarizing and sorting: long documents, incoming requests.
- Preparation: meeting sheets, interview outlines, preliminary research.
- Automatic chaining: generative AI plugged into a process automation, which removes copy-paste between tools.
The last point changes the scale. AI used alone speeds up one person; AI embedded in a workflow speeds up a process and can run unattended on simple cases. That is where ratios of 1 to 5 or more become plausible on a full process, provided humans still validate risky cases.
How do you create real leverage in an SME?
By choosing one process, measuring its starting point, standardizing usage, then measuring again.
A five-step method:
- Pick a process that occurs at least weekly and involves several people.
- Measure for two weeks: duration per occurrence, volume, rework rate.
- Replace free use with a shared, tested prompt and examples of good outputs.
- Set the review rule: who checks, on which cases, with which grid.
- Measure again after four weeks and compute the net gain, review included.
To turn the result into euros, see our method for calculating AI ROI in SMEs and the definition of AI ROI. To pick indicators to track over time, see AI KPIs in business. An illustrative calculation, with assumptions to replace by your own: a team of 6 spends 3 hours a week each producing quotes; if the process drops to 1.5 hours after review, that frees 9 hours a week, about 400 hours a year over 44 weeks. It is not 10x, it is 2x on one specific process, and that is already a solid result.
What should you do with the time freed?
Assign it in advance, or it disappears.
Decide before rollout what the time saved will be used for: more sales meetings, faster response times, a backlog to clear. Without that decision, the gain is invisible in results, which partly explains the gap between declared use and measured gain. It also requires training teams for a framed use: see our guide to AI change management and the 90-day AI roadmap.
Frequently asked questions
Can AI multiply productivity by 10?
On a specific step, a 1 to 10 ratio is possible (a first draft, a sorting job). On a whole job or company, no public study shows it: measured gains range from 14 to 55% depending on the task (Trésor-Éco no. 341).
What productivity gain is realistic in an SME?
It depends on the process. The work reviewed by the Direction générale du Trésor shows 14% to 55% on individual tasks. On a full process, a 1.5x to 2x factor is a credible target, to be confirmed by a before-and-after measurement.
Why do my teams use AI without visible gains?
Often because use is individual, unmeasured, and the freed time is not reallocated. A targeted process, a shared prompt and a baseline measurement fix most of these cases.
Do you need to automate to get real leverage?
Yes, when the task is stable and repetitive. Embedding AI in a workflow removes manual handling and lets simple cases run without intervention. Keep human validation on sensitive cases.
How do you measure the leverage?
Pick one process, record its duration, volume and rework rate for two weeks, then compare after four weeks of framed use, including review time.
Does training really change the outcome?
Companies' stated obstacles are first skills and use cases. Training centered on your real tasks addresses them directly. It also meets the AI literacy obligation in Article 4 of the EU AI regulation (Regulation (EU) 2024/1689).
How GrowthPerf helps
GrowthPerf is a Qualiopi-certified training provider specialized in AI, no-code and automation for SMEs and nonprofits. We start from one of your processes, measure its baseline, train the people involved on that exact case, then measure again. The operational AI training follows this logic.
If you want to know which leverage is realistic on one of your processes before investing, contact GrowthPerf for a first conversation. To place your priorities, you can also revisit the complete guide to AI in business.