The short answer
How to use AI in business: pick one repeated task in one department, check every output in under a minute, and tie it to a number you already track.
In a study of 5,179 customer support agents, an AI assistant raised issues resolved per hour by 14 percent on average and by 34 percent for novice workers.
BCG found 60 percent of companies report minimal gains (1,250 executives, 2025). Choose the first use with the three-question test below.
In this article 8 sections
The five lessons
Read these if you read nothing else.
- AI pays on a repeated task that a person can check fast. It does not pay on a task you do twice a year.
- The best measured result comes from support work: 14 percent more issues resolved per hour, 34 percent for novices.
- Most companies see little so far (60 percent, BCG 2025). Starting small and measuring beats starting wide.
- Pick one use, in one department, with one owner and one number you already track.
- A person keeps the call on money, hiring and anything said to a customer in your name.
What does the research say about using AI in business?
Newest first. The two adoption figures come from different groups. Census samples all US businesses. The NBER paper surveys executives in seven countries.
- 14%more issues resolved per hour, 5,179 support agentsNBER, 2023
- 34%gain for novice and low-skilled workersNBER, 2023
- 60%of companies report minimal gains from AIBCG, September 2025
- 19.8%of US businesses use AI, May 3, 2026US Census Bureau, May 2026
See the samples behind these numbers
| Finding | Sample | Source |
|---|---|---|
| 78 percent of firms use AI. Executives report AI raised annual labor productivity growth by 0.2 percent over three years and expect 0.6 percent a year ahead. Self-reported, not measured# | Nearly 9,000 CFOs, CEOs and senior executives, seven countries | NBER Working Paper 34836, 2026 |
| AI use hovered between 17 and 20 percent of US businesses (19.8 percent on May 3, 2026). 37 percent of firms with 250 or more employees use it. Under 20 percent of firms with four or fewer# | Census Business Trends and Outlook Survey, nationally representative, December 14, 2025 to May 3, 2026 | US Census Bureau, May 2026 |
| 60 percent of companies report minimal revenue and cost gains. 70 percent of AI's potential value sits in core business functions, 13 percent in IT# | 1,250 senior executives and AI decision makers, nine industries | BCG, September 2025 |
| An AI assistant raised issues resolved per hour by 14 percent on average and by 34 percent for novice and low-skilled workers, with minimal impact on experienced workers# | 5,179 customer support agents, working paper April 2023, later published in the Quarterly Journal of Economics | NBER Working Paper 31161 |
- The support study is the clearest measured result here. It tracked a real task, issues resolved per hour, for thousands of agents. That is the kind of number I would ask of any first use.
- BCG places most of the value in core business functions. That points the first use at the work that earns or keeps revenue, not at a side tool.
Where does AI pay first in each department?
These cards are our operating method. They are judgment from running this work, not a survey result. Each card gives a first use small enough to test in a month.
Sales
- First use
- Draft the follow-up on every open quote from the CRM notes. The rep sends it
- What it replaces
- Follow-ups written from scratch, or skipped
- Number that proves it
- Share of quotes followed up inside 2 days, then quote-to-close rate
- Risk to check
- A wrong price or promise in a draft the rep did not read
Customer service
- First use
- Draft replies to the 10 most common questions. The agent edits and sends
- What it replaces
- Retyping the same answer
- Number that proves it
- Issues resolved per hour, the measure in the support study
- Risk to check
- A confident wrong answer on refunds or warranty
Operations
- First use
- Turn intake notes and emails into structured work orders
- What it replaces
- Retyping between systems
- Number that proves it
- Minutes per work order and rework rate
- Risk to check
- Bad input data passes quietly into the schedule
Finance
- First use
- Match invoices to purchase orders and list the exceptions. A person approves payment
- What it replaces
- Line-by-line matching by hand
- Number that proves it
- Days to close the month and exceptions caught
- Risk to check
- A payment leaves without approval. Ask your accountant about tax and audit rules
HR
- First use
- Draft job posts and answer routine benefits questions from your own policy
- What it replaces
- Rewriting posts and repeating answers
- Number that proves it
- Hours spent per hire and time to answer an employee
- Risk to check
- Screening bias and employment law. Keep candidate decisions with a person and ask counsel before screening
Marketing
- First use
- Reply to web leads at once and text back missed calls
- What it replaces
- Leads that wait hours for an answer
- Number that proves it
- Minutes to first reply and booked appointments
- Risk to check
- A text sent outside client hours or without a human fallback
How do you choose the one first use?
Score each candidate from the cards with three questions. This is our operating method. Answer yes or no, and pick the use with three yeses.
- Does the task repeat at least weekly? Volume gives you a result in a month, not a year.
- Can a person check each output in under a minute and undo it? If a mistake cannot be undone, it is not a first use.
- Is there a number you already track that moves in dollars? If you must build the measure first, the project starts late.
What does a first use look like in numbers?
Example: a service company answers 600 customer emails a month, and each takes 6 minutes. That is 3,600 minutes, or 60 hours. If a draft saves 2 minutes per email, the saving is 1,200 minutes, or 20 hours a month. The numbers are round and made up, but the method is real: count the minutes, set the target, then compare after four weeks.
The support study counted issues resolved per hour, not time saved on a draft. I prefer that measure, because it ties the tool to finished work. For a phased plan that goes beyond the first use, see the AI implementation roadmap. For the first purchase, read best AI tools for business.
Who should wait?
If no task passes all three questions, wait and fix the process first. A broken handoff does not improve when a tool sits on top of it. This guide is also not for a company that wants AI to make hiring, pricing or legal calls alone.
The Census figure shows how early this is: under 20 percent of US businesses use AI today. Starting with one measured task is a normal place to be.
What to do this week
Five steps, each with an owner.
- CEO: name the one department where a slow or repeated task costs you the most.
- Department head: list the three most repeated tasks and score each with the three questions.
- Department head: write the baseline for the winning task, such as minutes per item, over the last 30 days.
- Operations lead: name the person who checks every output, and write down what they may never approve.
- CEO: set the review date at 30 days and the stop rule in advance. If the number does not move, the use ends.
What is the next step?
If you want a second pair of eyes on the choice, see AI implementation consulting. We start from your own data, people, process and decisions, and measure progress at each step.
Questions leaders ask
How do I start using AI in my business?
Pick one repeated task in one department, such as drafting replies to common customer questions. Name the person who checks each output, record the baseline for 30 days, then compare. Add a second use only after the first moves a number you already track.
How can AI help my business?
It helps most on repeated work that a person can check fast. In a study of 5,179 customer support agents, an AI assistant raised issues resolved per hour by 14 percent on average. Support, follow-up, intake and invoice matching are common first uses.
What are the best AI use cases in business?
Customer service replies, quote follow-up, work order intake, invoice matching, routine HR answers and instant replies to web leads. I would rank them by the three-question test: repeats weekly, checked in a minute, tied to a number in dollars.
How many businesses use AI?
It depends on who you ask. The US Census Bureau found 19.8 percent of businesses used AI on May 3, 2026, and 37 percent of firms with 250 or more employees. An NBER survey of nearly 9,000 executives in seven countries reported 78 percent of firms.
Does AI really pay off for companies?
Sometimes. BCG reports 60 percent of companies see minimal revenue and cost gains (1,250 executives, 2025). The measured support study showed 14 percent more issues resolved per hour. The gain follows the choice of task and the checking.
SOURCES
- NBER Working Paper 34836 (2026): nearly 9,000 executives, seven countries; 78 percent of firms use AI; executive-reported productivity effects
- US Census Bureau, Business Trends and Outlook Survey (May 2026): 19.8 percent of businesses use AI; 37 percent at 250 or more employees
- BCG, AI Leaders Outpace Laggards (September 2025): 1,250 senior executives; 60 percent report minimal gains
- NBER Working Paper 31161, Generative AI at Work (2023): 5,179 customer support agents; 14 percent and 34 percent gains
Where’s the money leaking?
Pick one. Be honest.
Cite this article
Grigorchuk, T. (2026, November 10). How to Use AI in Business: Where It Pays First, by Department. Megawebvision. https://megawebvision.com/insights/how-to-use-ai-in-business