Kief Studio

AI Use Cases by Business Size: What Small, Mid-Size and Large Firms Actually Use AI For

Automate

  • Published
  • Updated
  • 10 min read
Violet and gold ring of light with a small gap on black, cover for a note on AI use cases by business size

TL;DR: The best AI use cases depend less on company size than on how much repetitive, well-defined work a business has. Census Bureau data shows about one in five U.S. firms uses AI, mostly in sales and marketing, strategy and IT. Small firms should start with one task, mid-size firms with one workflow, large firms with governance.

We first published this note in 2024 as a broad tour of AI use cases for startups, mid-size companies and enterprises. A lot has changed since, and the U.S. Census Bureau now measures how businesses actually use AI, function by function. So I have rebuilt the note around that data. The tiers stay. The promises go, and what is left are AI use cases you can test against your own week.

Which AI use cases are businesses actually running?

Start with the measured picture, because it is smaller and more useful than the hype. The Census Bureau's Business Trends and Outlook Survey found that from December 2025 to May 2026 "overall AI usage hovered between 17% and 20%" of U.S. businesses, and that "between 20% and 23% of businesses expected to be using it in the next six months." (U.S. Census Bureau (opens in a new tab))

A Census working paper built on the survey's 2026 AI supplement goes a level deeper. During November 2025 to January 2026, "18% of firms used AI in a business function, rising to 32% on an employment-weighted basis." Among firms that use it, "57% of users integrate AI in three or fewer business functions, most commonly Sales and Marketing (52%), Strategy and Business Development (45%), and IT (41%)." (Census Bureau working paper CES-WP-26-25 (opens in a new tab))

The common AI use cases at the task level

The same paper names the tasks: "Writing, document analysis, and information search are the leading Generative AI use in tasks, though 65% of firms limit use to three or fewer tasks." (Census Bureau working paper CES-WP-26-25 (opens in a new tab))

That matches what I see when owners describe their AI use. It is drafting emails and proposals, summarizing long documents, answering "where is the policy on X" and cleaning up marketing copy. Those are real AI use cases. They are also the easy ones, and the gap between them and a system that runs part of the business is where most of the work lives.

AI helps people more often than it replaces them

The fear that AI means layoffs does not show up much in the data yet. "Most users (66%) rely on AI solely to augment tasks, while AI-related employment decreases are rare, occurring in only 2% of firms." (Census Bureau working paper CES-WP-26-25 (opens in a new tab)) The practical reading for an owner: plan AI use cases around the work your people dread, not around headcount.

Does company size change which AI use cases make sense?

Yes, but not in the way the old version of this note implied. Size changes how much repeat work you have, how clean your data is and how many people need to agree before something changes. It does not change what the technology can do.

The Census data shows the size gap clearly. "37% of firms with at least 250 employees reported using AI in their business operations," and 32% of firms with 100 to 249 employees did. Between December 2025 and May 2026, "AI use increased among firms with at least 20 employees but didn't change significantly among firms with fewer than 20 employees." (U.S. Census Bureau (opens in a new tab))

Industry matters as much as size. As of May 2026, use in the Information sector was 39.7% and in Finance and Insurance 33.9%, against a national rate of 19.8%, while retail sat around 14%. (U.S. Census Bureau (opens in a new tab)) A ten-person accounting office may be further along than a two-hundred-person retailer.

Why small firms say no

The working paper asked firms with no near-term plans why. The most common answer is that AI is "not applicable to the business (65% of firms, firm-weighted)." Lack of knowledge of what AI can do came second at 22%, followed by privacy and security concerns at 20%. (Census Bureau working paper, full text (opens in a new tab))

I read that first answer as an opening, not a verdict. Most owners who say AI does not apply are picturing a chatbot on the website. When we walk through their actual week, there is usually a copy-and-paste job somewhere that nobody enjoys.

What are good AI use cases for a small business or startup?

A small ground floor office in a triple decker at dusk, a desk and monitor lit purple through the front window
Small business AI use cases start at one desk: one task, done the same way every week.

A small team has little slack and little data. That points to AI use cases that save hours on work you already do, with a person checking the result. The original note listed scheduling, email, customer questions and financial modeling. The first three hold up. The last needs care, because a model that sounds confident about your numbers is not the same as a bookkeeper.

Small business AI use cases that tend to pay off

  • Drafting: first drafts of quotes, follow-up emails, job descriptions and social posts, edited by a person who knows the customer.
  • Summarizing: long vendor contracts, meeting notes and inspection reports turned into a one-page brief.
  • Answering the same questions: hours, service area, what a job includes. Our note on what an AI receptionist can and should not handle covers where the line sits.
  • Sorting the inbox: tagging new requests by type and urgency so the right person sees them first.

A startup has one advantage here: no old process to unwind. A founder can design the intake form, the CRM fields and the follow-up steps with automation in mind from the first customer. The original note called that agility. I would call it a chance to avoid building habits you will pay to automate later.

What to keep out of a small team's AI tools

Privacy and security were the third most common reason small firms held back, and that caution is reasonable. OWASP's 2025 list of the top risks for applications built on large language models includes "LLM02:2025 Sensitive Information Disclosure." (OWASP Top 10 for LLM Applications (opens in a new tab)) Decide what staff may paste into which tool before the habit forms. Brian Gagne, our CTO, wrote a plain guide on whether staff can put customer data into AI tools (opens in a new tab).

What AI use cases fit a mid-size company?

Mid-size firms are where the Census data shows growth, and where the work changes shape. Instead of one person with one tool, you have a process that crosses three departments and two pieces of software that do not talk.

The original note described AI for supply chain forecasting, inventory, quality control and personalized customer communication. Those are fair examples. The ones that ship are narrower than they sound.

From a task to a workflow

  • Order and inventory exceptions: the system flags the 5% of orders that do not match the usual pattern and routes them to a person, instead of trying to run purchasing on its own.
  • Customer service routing: AI answers the routine question, hands the complex one to a human with the history attached. The original note had this right.
  • Document intake: invoices, applications and work orders read, checked against the system of record and queued for approval.
  • Sales and marketing: the most common function in the Census data. Drafting campaigns, scoring leads and keeping the CRM current.

What separates a workflow from a pilot is the boring part: retries when a service fails, a log of what the system did, and a person who reviews the cases the model is unsure about. Our note on AI that works versus AI that sounds impressive goes further on that difference, and the AI implementation guide lays out the steps.

The data question comes first

Most mid-size AI use cases stall on data, not models. If customer records live in three places with three spellings of the same name, any AI built on top will repeat the confusion faster. Cleaning and connecting the data is often the first project, and our data engineering guide covers what that involves.

How do large enterprises approach AI use cases?

The original note said enterprises use AI for cybersecurity, forecasting, cross-department analytics and supply chains. All true, and all familiar. What is different at this size is that the hard problem is no longer finding AI use cases. It is managing dozens of them at once.

Governance before the next pilot

NIST's AI Risk Management Framework is the reference most large organizations reach for. NIST says it "is intended for voluntary use and to improve the ability to incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems." (NIST (opens in a new tab)) Its core describes "four specific functions" for managing AI risk in practice: "GOVERN, MAP, MEASURE, and MANAGE." (NIST AI RMF 1.0 (opens in a new tab))

In plain terms: decide who owns AI decisions, list where AI is used and what it touches, measure whether it does what you think, and fix or stop what does not. NIST added a profile for generative AI in July 2024 that "can help organizations identify unique risks posed by generative AI." (NIST (opens in a new tab))

Agents raise the stakes

AI agents, systems that take actions rather than just write text, raise the stakes most where many teams run them at once. OWASP describes the risk called excessive agency as the weakness "that enables damaging actions to be performed in response to unexpected, ambiguous or manipulated outputs from an LLM." Its root cause is "excessive functionality; excessive permissions; excessive autonomy." (OWASP (opens in a new tab)) Every agent needs the smallest set of tools and permissions its job requires. Our AI agent governance guide applies the same idea at any size.

How do you pick your first AI use cases, at any size?

The original note covered eight themes per tier: operations, customer experience, strategy, workforce, innovation, global markets, sustainability and engagement. That is a lot of possibility and not much direction. Here is the short version I give owners.

  1. List the repeat work. Anything someone does the same way more than ten times a month.
  2. Write down what good looks like. If you cannot describe a correct output, a model cannot hit it either.
  3. Measure today. Time per task, error rate, how long a customer waits. Without a before, you cannot prove an after.
  4. Pick one. The task with the most volume and the lowest cost of a mistake.
  5. Keep a person in the loop. Review everything at first, then sample as trust builds.
  6. Decide where the data may go. Before anyone pastes a customer record into a tool.

Most proposed AI use cases fail that list at step two, and that is useful information. A vague task needs a clearer process first, and a clearer process sometimes solves the problem without AI at all.

Use cases worth skipping for now

Anything where a wrong answer reaches a customer or a regulator without a person seeing it first. Anything you cannot measure. And anything you are building mainly because a competitor announced it. The Census data says most firms that use AI keep it to three or fewer functions, and that is a sensible pace.

Frequently asked questions

What are the most common AI use cases in business?

In Census Bureau data, firms that use AI most often use it in sales and marketing, strategy and business development, and IT. At the task level, writing, document analysis and information search lead.

How many U.S. businesses use AI?

Between December 2025 and May 2026, about 17% to 20% of U.S. businesses reported using AI, and 20% to 23% expected to within six months, according to the Census Bureau.

Do small businesses use AI less than large ones?

Yes. 37% of firms with 250 or more employees reported AI use, and use grew among firms with at least 20 employees while it held steady among smaller firms.

Is AI replacing workers?

Rarely so far. In the Census working paper, 66% of firms that use AI use it only to help with tasks, and AI-related staff reductions showed up in 2% of firms.

What is a good first AI project for a small business?

A repetitive task you can describe clearly and check easily, such as drafting follow-up emails or summarizing documents. Measure it before and after, and keep a person reviewing the output.

Which framework should a company use to manage AI risk?

Many organizations use the NIST AI Risk Management Framework, a voluntary framework built around four functions: govern, map, measure and manage. NIST also publishes a profile for generative AI.

Where we fit

Our AI implementation work is for businesses that tried a pilot that never shipped, or that want AI to take repetitive work off their team. We pick the workflow, define what good output looks like and measure how it runs today. Then we build it with retries, error handling and quality gates on every output, keep a human path for anything the model is unsure about, and run it for you on LTFI, so nothing is installed on your side.

It is not for demo-only chatbots or projects meant to replace staff. We build systems that remove bottlenecks, and your people drive the work. We are based in Shrewsbury, Massachusetts.

First conversation is free. No commitment.

Tell us which task you would hand off first

Related articles

  1. Automate

    Small Business AI in Massachusetts: 51% Use a Platform. Few Can Name What It Changed.

    Small business AI in Massachusetts: 51% use a platform, but Census and bank data show far less in daily work. Why, and a four-question test for owners.

    Brian Gagne

  2. Automate

    The Receptionist, the Bookkeeper and the Marketing Intern: What AI Adoption Really Covers

    What small business AI adoption looks like in Census and Chamber data, which jobs AI covers (calls, bookkeeping, drafts) and how a small team should start.

    Amelia Gagne

  3. Automate

    AI That Works vs. AI That Sounds Impressive

    How to tell AI that works from a demo: what MIT, S&P Global, Gartner and RAND found about stalled pilots, and the vacation test for a production workflow.

    Brian Gagne