Which AI use cases really pay off for owner-managed and mid-sized businesses?The ranking, with the evidence

For an owner-managed or mid-sized business, the AI use case with the best-established return is the customer service assistant. In the study by Brynjolfsson, Li and Raymond, published in 2025 in the Quarterly Journal of Economics, the tool increased the number of enquiries handled per hour by 15% on average across 5,172 advisers, and by around 30% among the least experienced. The measurement comes from a single company, a large software publisher, and concerns one specific task. Even with that limitation, it remains one of the more robust measurements available to a chief executive today.

I am publishing this ranking because the one you find by searching for “best AI use cases” is almost always a league table without sources: return‑on-investment percentages set side by side, with no indication of who measured them or how. An executive committee should not commit a budget on that basis. For many years I ran a vocational training group of around a thousand employees, and I now operate my own AI agents. In both cases I learnt not to commit money on a figure whose method I did not know, and that spared me several costly mistakes.

Why a sourced ranking now?

French small and mid-sized businesses are only now beginning to adopt AI, some way behind large companies. According to Insee (France’s national statistics office), 18% of businesses with 10 or more employees used at least one AI technology in 2025, against 6% in 2023. The difference by size is marked: 15% between 10 and 49 employees, 31% between 50 and 249, 58% above that (Insee Première no. 2120, July 2026).

The delay has one advantage: choices can now be made on the basis of published results. They remain scarce. In McKinsey’s global survey of 1,719 respondents, carried out in May and June 2026, 37% attribute at least some positive effect on their operating profit to AI, and only 6% attribute at least 5% of their operating profit to it while judging the value created significant (McKinsey, The State of AI, August 2026). For a smaller business, the task is to choose a first use with a genuine chance of appearing in the profit and loss account.

How is this ranking constructed?

Each line has been tested against three rules.

A measured result. Only uses whose gain was measured against a group without access to the tool are included: a controlled trial or a field experiment, published with its method. Self-reported surveys and suppliers’ figures provide context, never a score.

A line in the profit and loss account. Each use is linked to the line it affects: revenue, staff costs, purchasing. Time saved counts only if it translates into additional capacity or revenue.

The cost of entry for a smaller business. For a comparable gain, a use that starts with a subscription ranks ahead of one that requires clean historical data and a project of several months.

Rank	Use case	Line affected	Measured gain	Evidence	Cost of entry
1	Customer service assistant	Staff costs	+15% enquiries handled per hour, around +30% for the least experienced	Phased roll-out observed in real conditions, 5,172 advisers	Low
2	Drafting routine business documents	Staff costs	-40% time, +18% assessed quality	Controlled trial, 453 professionals	Very low
3	Demand forecasting and pricing	Revenue	+9.7% revenue in the test group	Field experiment, one online retailer	High
4	Software development assistant	Staff costs	+26% tasks completed, but -19% speed for experienced developers on their own code	Three field experiments, 4,867 developers; one controlled trial with the opposite result	Low
5	Analysis and preparation of deliverables	Staff costs	+25% speed and around +32% quality on suitable tasks, lower accuracy on a task outside the tool’s range	Controlled trial, 758 consultants	Low

Why does customer service come first?

It combines what the others have only in part: a measurement taken in real conditions, over several months, and a gain that can be read directly as capacity. The most useful result for a smaller business lies in the detail: the tool mainly benefits recently recruited advisers, and does very little for the most experienced (Brynjolfsson, Li and Raymond, NBER and then Quarterly Journal of Economics, 2025). In practice, the tool passes on to newcomers part of the know-how of longer-serving staff.

Smaller businesses know the problem well: each departure from customer service opens a period in which the replacement answers more slowly and less accurately. For a team of four, a 15% gain represents a little over half a person’s capacity. This is an indicative calculation, which assumes that the volume of enquiries justifies that capacity. It also requires an up-to-date base of answers, since the assistant relies on what the business has already written.

Does routine drafting really merit second place?

For reasons of cost. In the trial by Noy and Zhang published in Science in July 2023, 453 managers and professionals drafted letters, memos and analysis plans. Those with access to the tool took 40% less time, with quality judged 18% higher by independent assessors (MIT, summary of the study).

The measured time saving is the highest in the ranking, and the cost of entry is a subscription. What moves it down a place: the measurement comes from an online trial, on tasks of around half an hour, not from a business in operation. And the time saved only appears in the profit and loss account if someone decides how it is to be redeployed.

Why is the use that generates the most revenue only third?

It is the only one on the list that directly affects the top line. At Rue La La, an American online flash-sale fashion retailer, a demand forecasting tool combined with price optimisation increased revenue in the test group by around 9.7%, with a 90% confidence interval of 2.3% to 17.8% (Ferreira, Lee and Simchi-Levi, Manufacturing & Service Operations Management, 2016).

But it requires what many smaller businesses do not yet have: a clean sales history, a catalogue broad enough to give the forecast something to work on, and prices that can be changed frequently. The result comes from a single retailer, with a very particular sales model. It is probably the most powerful lever on the list for a wholesaler, an online retailer or a manufacturer with a catalogue, and considerably less so for a services business working on fixed fees.

Why do two studies of software development contradict each other?

Three field experiments at Microsoft, Accenture and a Fortune 100 company, covering 4,867 developers, measure 26% more tasks completed with a coding assistant, with the gain going mainly to the least experienced (Cui et al., Management Science, 2026). Conversely, the METR controlled trial published in July 2025, involving 16 experienced developers and 246 tasks on their own projects, measures a slowdown of 19%. The same developers expected to save 24% of their time and, once the work was finished, still believed they had saved 20% (METR, 2025).

The two results are not necessarily contradictory. One plausible explanation: on code they know by heart, developers lose time reviewing the tool’s suggestions, whereas on less familiar code they would gain time. For a business with one or two developers, it is better to decide task by task where the tool is of use.

And analysis, presentations and briefing notes?

This is the experiment conducted in 2023 with 758 consultants at Boston Consulting Group, published in 2026 in Organization Science (Dell’Acqua et al., Harvard Business School). On the tasks for which the tool was suited, the consultants completed more than 12% more tasks, more than 25% faster, with quality judged around 32% higher.

On a task chosen to exceed what the tool can do, the proportion of correct answers fell from 84.5% without AI to 70.6% with it, and to 60% among those who had also received an introduction to its use. For me, this is the principal finding: the tool does not indicate when it has gone beyond its range, and it falls to the business to identify those tasks and keep human oversight of them. Hence fifth place, despite a high gain.

What is not ranked, and why?

Predictive maintenance, visual quality control, fraud detection, autonomous agents that carry out sequences of tasks from start to finish. These uses may well be helpful, but I have not so far found a published study with a control group whose result can be applied to a business of 20 to 200 employees. The figures in circulation come from suppliers or consultancies, often without a published method. They will be added if such a measurement is published.

The McKinsey survey at least gives a direction: cost reductions are most often reported in the supply chain, service operations and production, and revenue increases in marketing and sales. These are self-reported, without controlled measurement, but they indicate which functions to examine first.

Where to begin?

With the first rank if you have a customer service team, the second if not: these are the two least expensive and best-evidenced points of entry. Then do what the studies in this ranking do, on your own scale. Equip half the team for six to eight weeks, keep the other half as a control group, and measure the same indicator on both sides. For instance, enquiries handled per hour, response time, the number of documents produced or the rate of rework.

You will then know whether the tool pays in your business, and not only in a study. We design tests of this kind with the chief executives we work with, starting from the decision maturity diagnostic. Definitions of the terms used here are in the glossary of the economics of decision-making.

Frequently asked questions

Peter Guillon, founder of Exelvys. The ranking is revised with each new published measurement.