Choosing your first AI use cases: where companies use AI, and how to rank the options

A use case is one task that artificial intelligence (AI) will do or help with: drafting replies to customers, summarising contracts, sorting support tickets, checking invoices. The first ones a large company picks set what its staff learn early on, and which data and rules it deals with first.

This page shows where companies use AI today, by business function, and why some that considered it decided against it. It then gives what the research says about which tasks suit AI, and a way to rank your own candidates. It is part of the guide on enterprise AI adoption.

Where companies use AI today

The US Census Bureau's AI supplement, collected 17 November 2025 to 8 February 2026, asked firms in which business functions they had used AI in the last six months [1]. Because of Executive Order 14434, signed on 29 September 2026, the Census Bureau now calls AI "Super Intelligence" or "SI" in its files [2][3]. Its questions asked about "Artificial Intelligence (AI)" until the collection that began on 19 October 2026, and this supplement closed on 8 February 2026, so this page says AI [3].

Business function	All US firms	US firms with 250 or more employees
Sales and marketing	14.3%	19.0%
Strategy and business development	12.4%	18.2%
Information technology (IT)	11.4%	25.9%
Research and development (R&D)	11.2%	11.9%
Finance and accounting	6.6%	20.5%

The figures are shares of all firms in each group, from the supplement [1]. Large firms put IT first, and finance and accounting second. Among large firms that planned to use AI in the next six months, 65.4% expected to use it in IT and 60.1% in finance and accounting [1].

More firms used AI to support staff than to replace their tasks. Of the firms that used AI, 43.7% used it to "Supplement or enhance a task performed by an employee", and 10.1% to "Perform a task previously done by an employee" [1]. For large firms the shares were 54.1% and 11.5% [1]. Among firms whose staff used generative AI, which is AI that creates text, images or code, the most common tasks were writing or editing documents, emails or communications (85.4%), searching for information or technical help (49.9%) and interpreting, analysing, translating or summarising documents (44.6%) [1].

In the European Union (EU), Eurostat reports that "34.70% of enterprises using AI technologies used these software or systems for marketing or sales and 31.05% for the organisation of business administration processes or management" [4]. Logistics was the least common purpose, at 6.08% [4]. Size changes the order. Among EU enterprises using AI, 47.51% of large ones used it for security of information and communication technology (ICT), against 14.51% of small ones [4]. Sector changes it too. Retail firms using AI used it most for marketing or sales (48.18%), and electricity, gas, steam, air conditioning and water supply firms used it most for ICT security (32.02%) [4].

In the United Kingdom (UK), the Office for National Statistics (ONS) reports that improving business operations is the most common purpose businesses give for using AI [6].

Why some companies stop

Eurostat asks enterprises that considered AI and did not use it why. The database gives these shares for 2025, for enterprises with 10 or more employees [5].

Reason for not using AI	Share of EU enterprises that considered AI, 2025
Lack of relevant expertise	70.31%
Unclear legal consequences	53.61%
Concerns about data protection and privacy	52.72%
Availability or quality of data	43.51%
Incompatibility with existing equipment, software or systems	41.6%
Costs too high	38.37%
Ethical considerations	24.64%
AI not useful for the enterprise	17.79%

In the US, firms that did not plan to use AI in the next six months most often said "AI is not applicable to this business" (61.6%) [1]. Next came lack of knowledge of what AI can do (22.0%) and concerns about privacy or security (20.7%) [1]. Among large firms not planning AI, privacy and security concerns reached 29.7%, and 19.4% said AI was not mature enough [1]. In Canada, Statistics Canada found that cybersecurity or privacy concerns were the leading barrier, reported by 13.4% of businesses, followed by cost at 10.6% [7].

In the UK, earlier ONS analysis found that the most common barriers included "difficulty identifying business use cases", cost and lack of expertise; in June 2026, 41% of UK businesses with 10 or more employees reported no barriers [6]. A report by the OECD (the Organisation for Economic Co-operation and Development), Boston Consulting Group (BCG) and INSEAD says that the public bodies set up to help firms adopt digital technology "frequently highlight uncertainty over the return on investment as a critical obstacle for firms considering adopting AI" [11].

Not knowing what AI can do, finding no use case and doubting the return all concern the choice of work. In Reveneau's view, a first task that the company can test and measure helps with all three.

Which tasks suit AI: what the research says

AI's abilities are uneven

Fabrizio Dell'Acqua and co-authors ran a pre-registered experiment with 758 BCG consultants, published as a Harvard Business School working paper in September 2023. They found that "some tasks are easily done by AI, while others, though seemingly similar in difficulty level, are outside the current capability of AI" [8]. The authors call the line between the two a "jagged technological frontier". That means two tasks that look equally hard to a person can sit on different sides of what AI does well.

On a task chosen to be outside that line, "consultants using AI were 19 percentage points less likely to produce correct solutions compared to those without AI" [8]. On 18 tasks inside it, consultants with AI finished 12.2% more tasks and worked 25.1% faster [8]. BCG staff are among the authors, and the paper is a working paper.

So how hard a task looks to a person is a poor guide to whether AI will help with it. In Reveneau's view, each candidate needs a test on the company's own work before anyone relies on it.

Gains are largest where staff lack experience

Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied an AI assistant for customer support agents, in a paper published in the Quarterly Journal of Economics in February 2025. They found that "the gains from AI adoption are largest for moderately rare problems, where human agents have less baseline experience but the system still has adequate training data" [9]. The best first uses can be the ones that are hard for new staff and common enough that the company has many past examples.

How much work AI could speed up

Tyna Eloundou and three co-authors estimated in 2023 that with access to a large language model, 15% of all worker tasks in the US could be completed faster at the same level of quality [10]. A large language model (LLM) is an AI model trained on large amounts of text. With software built on top of LLMs, their estimate rises to between 47% and 56% of tasks [10]. They reached these figures by rating occupations against what LLMs can do, using human raters and GPT-4, so the figures describe what could be possible. The authors say they "do not make predictions about the development or adoption timeline" [10].

How to rank your candidates

The US National Institute of Standards and Technology (NIST) AI Risk Management Framework asks a company to write down three things about each AI system. Item MAP 1.1 asks that the "Intended purposes, potentially beneficial uses, context-specific laws, norms and expectations, and prospective settings in which the AI system will be deployed are understood and documented" [12]. Item MAP 3.1 asks that "Potential benefits of intended AI system functionality and performance are examined and documented" [12]. Item MAP 3.2 asks the same for the costs of errors [12]. Those three notes are the start of a ranking.

The rules also shape the ranking. Under the EU AI Act, AI used to evaluate the creditworthiness of people (AI that detects financial fraud is excepted), or for "risk assessment and pricing in relation to natural persons in the case of life and health insurance", is high-risk under Annex III point 5 [13]. High-risk duties for Annex III systems apply from 2 December 2027, under the 2026 amendment [14]. A high-risk use carries more duties, so it is a slower first project.

The table below is Reveneau's scoring sheet, built from these sources. Score each candidate from 1 to 3 on each line.

Question	Score 3	Score 1	Source behind it
How often does the task happen?	Many times a day	A few times a year	Reveneau's view: a rare task returns little
What does an error cost?	Cheap and easy to catch	Harm to a customer or a legal breach	NIST MAP 3.2 [12]
Can you write tests for it?	Past examples with known right answers	No agreed right answer	Dell'Acqua et al. [8]
Is the data ready?	Clean, in one place, allowed for this use	Scattered, or limited by consent	Eurostat reasons [5]
Which rules apply?	No high-risk category	Annex III high-risk, such as credit scoring	EU AI Act [13]
Does it need an older system?	No, or through an existing interface	Yes, with no interface	Eurostat reasons [5]
Is there an owner?	A named manager who wants it	Nobody asked for it	NIST MAP 1.1 [12]

A test, in this sense, is an eval: a written, repeatable check of the AI system's answers against cases the company chooses. The guide on why AI evals matter explains them. If a task cannot be tested, the company cannot tell whether AI helps with it.

Reveneau's view on a first set

Reveneau's view is that a large company should start with two or three uses, picked from the top of the scoring sheet. Good first uses happen often, an error is cheap to catch, and past examples with known answers exist to build tests from. Writing, search and summaries inside the company, and work in IT and finance, can fit that pattern. Writing and search are the most common generative AI tasks in the Census supplement, and IT and finance are the functions where the most large US firms use AI [1]. High-risk uses and uses that need large changes to old systems come later, after the company has learned on simpler work.

Each first use then needs a build decision, covered in build, buy or partner, a data check, covered in getting company data ready for AI, and a measure of success, covered in measuring what AI returns. If a candidate use takes actions on its own, such as sending emails or changing records, read AI agents in production first.

Reveneau is an AI software development consultancy with offices in San Francisco and Vancouver. Reveneau says AI writes all of its code and every change must pass an eval suite written from the specification, the written description of what the software must do. Reveneau says it uses AI instead of hiring more engineers and puts that saving into clients' prices. To talk about a first use case, see Reveneau's services or contact Reveneau.