AI agents in business: real uses and benefits
Short answer Businesses use AI agents today mostly for customer questions, sales follow-up, invoices and bookkeeping, and marketing tasks. The best-measured gain is speed: one study found support staff resolved 14% more issues per hour with AI help. Adoption is still early, about 1 in 5 businesses in the US and Canada use any AI, and many agent projects fail.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callWhere do businesses use AI agents today?
In the jobs that are repetitive but need some judgment: answering customers, following up on leads, chasing invoices and keeping the books. An AI agent is software that takes a goal, picks its own next steps and works inside your tools (inbox, calendar, accounting app) until the job is done. A chatbot only answers; a rule only does what you set. For the full definition, see Agentic AI explained.
| Business job | What an agent does | Real example | What you keep |
|---|---|---|---|
| Customer questions | Answers chats and messages, hands hard cases to staff | Among Canadian businesses using AI, 28.2% use virtual agents or chatbots (Statistics Canada, 2026) | Refunds, complaints, anything off-script |
| Calls and booking | Answers the phone, books the slot | See AI receptionist for a small business | Pricing and diagnosis |
| Sales follow-up | Finds leads, drafts replies, suggests meetings | QuickBooks Customer Agent, launched July 2025 for US users (Intuit) | Which deals to chase, the final quote |
| Invoices and payments | Predicts late payers, sends invoices and reminders | QuickBooks Payments Agent (Intuit) | Payment terms, who gets extra time |
| Bookkeeping | Sorts transactions, helps reconcile | QuickBooks Accounting Agent (Intuit) | Review before filing anything |
| Marketing | Ads, search, content, reviews | See AI agents for small business marketing | Budget, offer, brand voice |
The pattern: the agent does the routine part, a person keeps the decisions that cost money or a customer. OpenAI, which sells agent technology, says agents fit work where fixed rules fall short: judgment calls, rulebooks too big to maintain, and messy text like emails and forms. Otherwise:
"Before committing to building an agent, validate that your use case can meet these criteria clearly. Otherwise, a deterministic solution may suffice." (OpenAI: A practical guide to building agents)
"Deterministic" here means a plain rule that always gives the same output, like an automatic receipt.
What benefits do AI agents bring?
Time back, mainly. The strongest independent evidence is one large study; most other numbers come from vendors.
| Claimed benefit | Source and what it measured | How much to trust it |
|---|---|---|
| 14% more support issues resolved per hour; 34% for new or less-skilled staff, little change for experts | NBER study of an AI assistant used by customer support workers (Brynjolfsson, Li, Raymond, 2023) | Independent research. An assistant that suggests replies, not a fully hands-off agent |
| Paid about 5 days faster on average | Intuit, for US beta users of AI-drafted invoice reminders, 2024 | Vendor figure from its own customers |
| 12 hours a month saved on bookkeeping | Intuit survey: 45% of users of its AI bank feed said so | Vendor survey; the other 55% saved less or didn't say |
| 80% of common customer service issues solved without a person | Gartner forecast for 2029 (March 2025) | A prediction, not today's results |
The study's finding matters for small teams: the biggest lift went to the least experienced workers. An agent can help a new hire answer like a veteran. It did little for people who already knew the job well.
Treat any "saves 30–50%" or "handles 80% of inquiries" claim as marketing until it names who was measured, doing what, and when.
How many businesses use AI agents?
Fewer than the headlines suggest. These figures cover any AI use, so agents are a smaller slice.
| Country | Share of businesses using AI | Source |
|---|---|---|
| US | About 18% of firms at the end of 2025 | Federal Reserve Board note on the Census Bureau's business survey, April 2026 |
| Canada | 19.2% in the 12 months before Q2 2026, up from 6.1% two years earlier | Statistics Canada, June 2026 |
In Canada, 40.0% of businesses said AI isn't relevant to them (Statistics Canada). In the US, the Fed note finds the largest firms adopt AI the most. The US survey broadened its AI question in November 2025, so older and newer US figures don't compare directly.
What goes wrong with AI agents?
Three things: projects that never pay off, agents that say the wrong thing, and agents that act without a check.
Projects get dropped. In June 2025, Gartner predicted that over 40% of agentic AI projects will be canceled by the end of 2027, citing rising costs, unclear value and weak risk controls. It also warned of "agent washing": vendors relabeling old chatbots and automation as agents.
You own what the agent says. In 2024, a British Columbia tribunal ordered Air Canada to pay a customer after its website chatbot told him he could claim a bereavement discount after travel, which the airline's policy didn't allow. The airline argued the chatbot was a separate entity responsible for its own words. The tribunal disagreed:
"It should be obvious to Air Canada that it is responsible for all the information on its website." (BC Civil Resolution Tribunal: Moffatt v. Air Canada, 2024 BCCRT 149)
That is a Canadian small claims ruling, not a national rule, but it shows how a court can treat a bot's answer.
It acts on money without a check. OpenAI's guide names the actions that should keep a person in the loop:
"Actions that are sensitive, irreversible, or have high stakes should trigger human oversight until confidence in the agent's reliability grows." (OpenAI: A practical guide to building agents)
Its examples: canceling orders, approving large refunds and making payments.
How do you start with an AI agent?
Pick one job, measure it, and keep approval on anything that moves money.
- Choose one repetitive job that eats hours each week, such as invoice reminders or after-hours inquiries.
- Ask if a rule would do. If the job never needs judgment, a simple automation in the tool you already pay for is cheaper.
- Check the agent lives where your data is. Many accounting, booking and ad tools now include agents, so start there before buying a separate product.
- Set limits. Approval for refunds, payments, price quotes and anything sent to all customers.
- Measure before and after: hours spent, days to get paid, replies missed. Keep it if the number moves.
Example
Illustrative numbers, not real clients.
A cleaning company chasing invoices
The owner spends 4 hours a month chasing 30 overdue invoices. She turns on the reminder agent in her accounting app and approves each reminder for the first month. Average days to get paid falls from 35 to 31, and chasing takes 1 hour. She stops approving routine reminders and keeps approval on anything over $1,000.
A dental office answering after hours
About 20 messages a week arrive after closing, and half wait until morning. A chat agent answers hours, insurance accepted and open slots, and hands anything clinical to staff. The office reviews a week of transcripts before letting it book on its own.
FAQ
Are AI agents the same as chatbots?
No. A chatbot answers questions. An agent also takes steps in your tools, like sending an invoice or booking a slot, then checks the result.
Will AI agents replace my staff?
The evidence so far shows help, not replacement. The NBER study found the biggest gains for newer workers using AI alongside their own judgment.
Do I need a developer to use an AI agent?
Usually not. Many agents now come built into accounting, booking, phone and ad software a small business already uses.
Who is responsible if an AI agent gives a customer wrong information?
In the Air Canada case, the business was. Review what your agent says and keep a person on refunds, prices and promises.
Different for you?
- Your main goal is more leads from marketing: see AI agents for small business marketing or the Conversion agent.
- You miss calls: see AI receptionist for a small business.
- You want to know what's safe to run hands-off: see AI marketing automation.
- You're unsure what counts as an agent: see Agentic AI explained.