Which sales and operations jobs AI has proven it can do

Customer support is where AI automation for small business has the most proof. A peer-reviewed study of 5,172 support agents found AI help raised issues resolved per hour by 15%¹, and vendor case studies say AI agents now close a large share of routine tickets at smaller firms²³. Invoice reminders and lead follow-up have some evidence. We found no measured small-firm result for proposals, meeting notes or weekly reports.

AI workflow examples ranked by strength of proof

We ranked each job by who measured the result. An independent study counts most, then a company comparing its own customers, then a customer quote on a vendor's site. Nearly every small-firm number we found is the last kind.

AI customer service automation has the most proof

The strongest evidence is a study of chat support at a Fortune 500 software firm, published in the Quarterly Journal of Economics in 2025¹. The AI suggested replies and agents could edit or ignore them. Newer and less-skilled agents gained the most, while the most experienced gained a little speed and lost a little quality¹.

That study measured AI helping staff. The cases below measure an agent answering customers on its own, which is harder, and every one comes from the vendor. What the productivity research shows covers the wider trial evidence.

How Sticos set up its agent

Sticos, a Norwegian compliance-software firm in the Visma group, gave the most detailed setup story we found². It launched HubSpot's support agent on in-product chat first, then moved to email with the question types it trusted most. Tickets that need empathy or complex problem-solving go to people. Staff tagged each ticket they answered with whether the agent could have handled it and whether an article was missing. HubSpot reports the agent now handles 41% of incoming support, against a 40% goal².

The same agent gets very different numbers at different firms. Kaplan and Canibuild both point to their help content when they explain theirs³. Company AI brains that worked also answered from one curated library, and how AI knowledge assistants work shows how a missing article turns into a wrong answer.

What are AI Agents?IBM Technology · 12 min · Plays here

AI invoice reminders and sales follow-up have some proof

Invoice reminders have the cleanest comparison here. Intuit reports that U.S. beta customers using AI-drafted reminders were paid an average of 5 days faster than when standard reminders went to the same customers, from January to August 2024⁴. It's Intuit's own opt-in data, but holding the customers fixed makes it stronger than most vendor numbers. We found no Canadian figure.

Sales follow-up rests on vendor quotes. RevenueWell says meetings booked rose 28% after it replaced its email sequences with HubSpot's prospecting agent³. The older evidence is about speed: in a 2011 study of 1.25 million leads, firms that tried to reach a lead within an hour were nearly seven times as likely to qualify it as firms that tried an hour later⁵. An AI can reply in minutes, but we found no study of whether its replies qualify leads as well as a salesperson does.

Old leads, proposals, meeting notes and reports have the least proof

For dormant leads, HubSpot says Ving booked meetings with leads quiet for over a year, and gives no counts³. For proposals, meeting notes and weekly reports, we searched the customer stories of HubSpot, Salesforce, Google, Microsoft and Intercom and found no small firm with measured results.

The closest evidence for drafting is an experiment published in Science. Professionals randomly given ChatGPT for writing tasks in their own fields took 40% less time, and graders rated their work 18% higher⁶. Those tasks had none of a firm's own prices or client history, so treat the gains as a ceiling and have a person check every figure in a proposal.

AI chatbot liability falls on the business

Klarna automated support at scale, then brought people back. In February 2024 it said its AI assistant handled two-thirds of its customer service chats in its first month⁷. In May 2025 its CEO said the cheaper service had delivered lower quality, and Klarna began hiring human agents again⁸. It kept the AI, and in November 2025 said the assistant did the work of more than 853 agents⁹.

In Canada, a tribunal has already made a business pay for its bot. In February 2024, British Columbia's Civil Resolution Tribunal ordered Air Canada to pay C$812.02 after its chatbot told a grieving customer they could claim a bereavement fare after travelling, which the airline's policy didn't allow¹⁰. The tribunal found the airline "is responsible for all the information on its website," chatbot included¹⁰. Until you've measured how often your AI is wrong, have a person review what it sends and give customers an easy route to a human.

Check consent rules and vendor claims before outreach

• Under Canada's anti-spam law (CASL), implied consent from a purchase lasts two years and from an inquiry six months¹¹. Check consent dates before an AI emails old leads.
• In the U.S., the FCC ruled in 2024 that AI-generated voices count as artificial voices under the Telephone Consumer Protection Act, so AI voice calls need prior express consent unless an exemption applies¹².
• A class action alleges Otter.ai's notetaker recorded calls without everyone's consent, and in August 2026 a federal judge let several claims proceed, with no finding of liability yet¹³. Ask everyone before a notetaker records a call.
• In 2025 the FTC sued Air AI, which said its product could replace human service reps, over false earnings and refund promises¹⁴. In 2026 it agreed to a proposed order banning it from marketing business opportunities¹⁵. Ask any vendor for customers you can call and the baseline behind each number.

Where to start with AI automation for small business

Start with one task you can measure. The calculator below gives a rough sense of what one task is worth, using your own numbers.

1. Pick support or invoice reminders, the two jobs with measured results¹⁴.
2. Record a few weeks of volume, time to close or days to payment before you switch anything on.
3. Fix the help content, starting with articles that contradict each other. Air Canada's chatbot gave a rule its own bereavement page contradicted¹⁰.
4. Launch on one channel with a numeric target and a route to a person, as Sticos did². Expect months of tuning, since Canibuild's rate climbed over almost a year³.
5. Have a person approve AI-drafted messages until you know the error rate, and check consent dates before any list goes out¹¹.
6. Train the staff who'll supervise it. Why staff training decides the result covers what to teach, and how the audit works shows how we'd scope the first workflow.