AI automation for small business works best one task at a time. This lesson gives you a quick test for any task, and shows what our data says about the tasks in common small-business roles.
In this lesson
- A five-question test you can run on any task in your business.
- The four outcomes: hand it over, AI drafts and a person checks, AI helps a person, or keep it human.
- Typical candidates in eight business functions, each tagged with an outcome.
- How to read our task ratings for any role on your payroll, live from our data.
AI automation for small business starts with tasks, not tools
The most common mistake is to start with a tool (“we bought an AI assistant, now what?”) or a job title (“can AI replace our receptionist?”). Both are the wrong size. A job is a bundle of tasks, and AI is good at some of them and poor at others, often within the same morning’s work. A receptionist answers routine calls, books appointments, calms an upset visitor and signs for deliveries. AI can take some of that, help with some, and has nothing to offer for the rest.
So the unit of this lesson is the task: one piece of work with a clear start and end, such as “send appointment reminders”, “reconcile the bank account”, “write the weekly social post” or “decide whether to give a refund”.
The task test
For any task, ask five questions.
- 1Is it done often, the same way?YesNoSometimes
- 2Is it mostly words, numbers or data on a screen?YesNoSome hands-on work
- 3Can someone check the result quickly?YesNoOnly an expert
- 4What does a mistake cost, and who would notice?
- 5Does a customer, regulator or law expect a person to do or sign it?YesNoNot sure: check
- Is it done often, the same way? Volume and repetition are where AI saves the most time, and where it is easiest to set up and check. A task done twice a year is rarely worth automating.
- Is it mostly words, numbers or data on a screen? AI works with text, figures, images and sound. If the task is physical or happens in person, AI can at most help with the paperwork around it.
- Can someone check the result quickly? If a person can tell in a few seconds whether the output is right, such as a meeting summary they attended or an email reply they can read, AI is safe to try. If checking takes as long as doing, the time saved disappears.
- What does a mistake cost, and who would notice? A clumsy social post costs little and you would see it. A wrong figure on a tax return, a wrong dose in a care plan or a wrong promise to a customer costs a lot, and may not be spotted until too late.
- Does a customer, regulator or law expect a person to do or sign it? Some work must be done or signed off by a named, qualified person. Others, like hiring decisions, carry legal duties about fairness. AI can help prepare that work, but the person stays responsible. Lesson 10 covers the rules.
You do not need a score. Two or three honest answers usually make the outcome obvious.
The four outcomes
The answers place each task in one of four outcomes. The two questions that matter most are how often the task happens and how easy it is to check, with the cost of a mistake weighing on both.
A two by two grid. Across: how easy the result is to check, hard on the left, easy on the right. Up: how often the task happens, rare at the bottom, frequent at the top. Frequent and easy to check: hand it over, with spot checks (tag incoming invoices, book appointments). Frequent and hard to check: AI drafts, a person checks every one (reply to complaints, quote prices for custom jobs). Rare and easy to check: AI helps when needed (draft a job ad, plan a holiday rota). Rare and hard to check or high stakes: keep it human (decide a pay rise, answer a legal letter). Dot size shows the cost of a mistake.
- Hand it over, with spot checks. Frequent, easy to check, low cost if wrong. AI does the task end to end; a named person samples the output each week. Examples: appointment reminders, sorting incoming email into folders, transcribing voicemails.
- AI drafts, a person checks every one. Frequent, but harder to check or more costly if wrong. AI produces a first draft; a person reads, corrects and sends every one. Examples: replies to customer complaints, quotes from site notes, job adverts.
- AI helps when needed. Occasional tasks that are easy to check. A person does the work and asks AI for help with parts of it. Examples: researching a new supplier, drafting an unusual letter, making sense of a long regulation.
- Keep it human. Rare, hard to check or high stakes, or where a person is expected by law or by the customer. AI may help with background reading, but the work and the decision stay with a person. Examples: deciding who to hire or let go, signing off accounts, handling a bereaved customer, anything physical or on site.
Notice that even “hand it over” names a person. In this course, every task that goes to AI has someone who checks it, and Lesson 5 shows how to put those names on an org chart.
Typical candidates by function
The cards below are starting points, not rules. Run the task test on your own version of each task: a small business with three customer emails a day is in a different place from one with three hundred.
Admin and scheduling
- Booking and rescheduling appointments: hand it over, with spot checks
- Meeting notes and action lists: hand it over, with spot checks (attendees check)
- Drafting routine letters and forms: AI drafts, a person checks
- Deciding priorities when the diary clashes: keep it human
Finance and bookkeeping
- Capturing receipts and invoices into the accounts: hand it over, with spot checks
- Suggesting matches in bank reconciliation: AI drafts, a person checks
- Chasing overdue invoices with polite reminders: AI drafts, a person checks
- Signing off accounts, tax returns and payments: keep it human
Marketing and content
- First drafts of social posts, newsletters and product descriptions: AI drafts, a person checks
- Summarizing reviews and comments for themes: AI helps when needed
- Resizing and re-wording content for different channels: hand it over, with spot checks
- Brand decisions and anything that speaks for the business in a crisis: keep it human
Sales
- Researching a prospect before a call: AI helps when needed
- Drafting follow-up emails after a meeting: AI drafts, a person checks
- Updating the customer database from emails and calls: hand it over, with spot checks
- Negotiating price and terms: keep it human
Customer service
- Answering “where is my order?” and opening-hours questions from your own information: hand it over, with spot checks (with a clear route to a person; Lesson 9)
- Drafting replies to complaints: AI drafts, a person checks
- Tagging and routing incoming messages: hand it over, with spot checks
- Refunds over a limit, exceptions and upset customers: keep it human
HR and recruiting
- Drafting job adverts and interview questions: AI drafts, a person checks
- Answering staff questions about the handbook: AI helps when needed
- Scheduling interviews: hand it over, with spot checks
- Shortlisting, hiring, discipline and dismissal decisions: keep it human (Lesson 10 explains the fairness duties)
Operations and supply
- Drafting supplier orders from stock levels: AI drafts, a person checks
- Planning delivery routes and rotas: AI drafts, a person checks
- Summarizing supplier contracts: AI helps when needed (check the clauses that matter yourself)
- Choosing and managing suppliers: keep it human
IT and security
- Answering common “how do I…?” questions from staff: AI helps when needed
- Drafting simple scripts and spreadsheet formulas: AI drafts, a person checks
- Summarizing security alerts: AI helps when needed
- Granting access, approving changes and responding to a breach: keep it human
What our data says, role by role
The function cards are judgment calls. Our data offers something firmer for specific roles. For every job we score, we rate each task in the official US job description (from O*NET, the Department of Labor’s database) as AI does it, AI helps or needs a human. Those ratings roughly match the outcomes above: AI does it tasks are candidates to hand over or have AI draft; AI helps tasks are where AI drafts or assists and a person stays in the task; needs a human tasks stay human.
Here are common small-business roles and their tasks rated AI does it or AI helps, live from our current data release. The ratings change as AI improves, so this list does too.
| Task | Job | Our rating |
|---|---|---|
| Operate telephone switchboard to answer, screen, or forward calls, providing information, taking messages, or scheduling appointments. | Receptionists and Information Clerks | AI does it |
| Receive payment and record receipts for services. | Receptionists and Information Clerks | AI helps |
| Operate computers programmed with accounting software to record, store, and analyze information. | Bookkeeping, Accounting, and Auditing Clerks | AI does it |
| Check figures, postings, and documents for correct entry, mathematical accuracy, and proper codes. | Bookkeeping, Accounting, and Auditing Clerks | AI helps |
| Confer with customers by telephone or in person to provide information about products or services, take or enter orders, cancel accounts, or obtain details of complaints. | Customer Service Representatives | AI does it |
| Keep records of customer interactions or transactions, recording details of inquiries, complaints, or comments, as well as actions taken. | Customer Service Representatives | AI does it |
| Answer telephones, direct calls, and take messages. | Office Clerks, General | AI does it |
| Communicate with customers, employees, and other individuals to answer questions, disseminate or explain information, take orders, and address complaints. | Office Clerks, General | AI does it |
| Collect and analyze data on customer demographics, preferences, needs, and buying habits to identify potential markets and factors affecting product demand. | Market Research Analysts and Marketing Specialists | AI does it |
| Prepare reports of findings, illustrating data graphically and translating complex findings into written text. | Market Research Analysts and Marketing Specialists | AI does it |
| Maintain customer records using automated systems. | Sales Representatives of Services, Except Advertising, Insurance, Financial Services, and Travel | AI does it |
| Answer customers' questions about services, prices, availability, or credit terms. | Sales Representatives of Services, Except Advertising, Insurance, Financial Services, and Travel | AI does it |
| Interpret and explain human resources policies, procedures, laws, standards, or regulations. | Human Resources Specialists | AI does it |
| Review and evaluate applicant qualifications or eligibility for specified licensing, according to established guidelines and designated licensing codes. | Human Resources Specialists | AI helps |
| Request and review project updates to ensure deadlines are met. | Project Management Specialists | AI helps |
| Plan, schedule, or coordinate project activities to meet deadlines. | Project Management Specialists | AI helps |
And the same roles’ tasks that our ratings say need a human:
| Task | Job | Our rating |
|---|---|---|
| Greet persons entering establishment, determine nature and purpose of visit, and direct or escort them to specific destinations. | Receptionists and Information Clerks | Needs a human |
| File and maintain records. | Receptionists and Information Clerks | Needs a human |
| Operate 10-key calculators, typewriters, and copy machines to perform calculations and produce documents. | Bookkeeping, Accounting, and Auditing Clerks | Needs a human |
| Receive, record, and bank cash, checks, and vouchers. | Bookkeeping, Accounting, and Auditing Clerks | Needs a human |
| Operate office machines, such as photocopiers and scanners, facsimile machines, voice mail systems, and personal computers. | Office Clerks, General | Needs a human |
| Maintain and update filing, inventory, mailing, and database systems, either manually or using a computer. | Office Clerks, General | Needs a human |
| Direct trained survey interviewers. | Market Research Analysts and Marketing Specialists | Needs a human |
| Attend staff conferences to provide management with information and proposals concerning the promotion, distribution, design, and pricing of company products or services. | Market Research Analysts and Marketing Specialists | Needs a human |
| Distribute promotional materials at meetings, conferences, or trade shows. | Sales Representatives of Services, Except Advertising, Insurance, Financial Services, and Travel | Needs a human |
| Hire employees and process hiring-related paperwork. | Human Resources Specialists | Needs a human |
| Address employee relations issues, such as harassment allegations, work complaints, or other employee concerns. | Human Resources Specialists | Needs a human |
| Communicate with key stakeholders to determine project requirements and objectives. | Project Management Specialists | Needs a human |
Two things to notice. First, the official task descriptions are general: your receptionist’s job will not match the list exactly, so use it as a prompt for your own list, not a substitute. Second, a task rated AI does it still needs someone to set it up, spot-check it and handle the exceptions. The rating says what AI can do, not that nobody is needed.
Look up the roles on your payroll
Each role above has its own page with its full task list, split by rating:
- Receptionists and Information Clerks · AI tools for receptionists
- Bookkeeping, Accounting, and Auditing Clerks · AI tools for bookkeepers
- Customer Service Representatives · AI tools for customer service
- Office Clerks, General
- Market Research Analysts and Marketing Specialists · AI tools for marketing
- Sales Representatives of Services, Except Advertising, Insurance, Financial Services, and Travel · AI tools for sales
- Human Resources Specialists · AI tools for recruiting
- Project Management Specialists · AI tools for project managers
For any other role, search the rankings by job title. To see several roles at once, with the number of people you employ in each, use the AI team planner: it lists your team’s tasks to look at first, and the ones that need a person, side by side.
Common traps
- Starting with the hardest task. Pick something frequent, easy to check and low-stakes for your first try. Early wins build the habits you need for harder tasks.
- Forgetting the checking time. If checking an AI draft takes nearly as long as writing it, the task is not a good candidate yet. Lesson 11 shows how to count it.
- Handing over a task nobody understands. If the person who does it today cannot explain how, AI will not fix it. Write the steps down first; you will often find a simpler process.
- Skipping the people who do the work. They know which tasks are tedious, which are tricky, and where mistakes hide. Ask them. Our guide on talking to your manager about AI is written for staff, and it shows the questions they are likely to bring you.
Key terms
- TaskOne piece of work a job involves, such as “Process invoices for payment.” We score every task, then add them up.The task lists come from O*NET, the US government’s job database.More in the glossary
- Task splitA job’s working time sorted three ways: tasks AI does, tasks AI helps a person with, and tasks that need a human.The three shares add up to 100%. Amber is always the human share.More in the glossary
- AI does itTasks AI can already do mostly by itself today, with a person checking the result.Every task gets an AI score from 0 to 100; tasks scoring 50 or more go here.More in the glossary
- AI helpsTasks a person still does, with AI speeding up parts of them, such as looking things up or writing a first draft.Tasks with an AI score from 25 to 49 go here.More in the glossary
- Needs a humanTasks AI can do little of today, such as hands-on work, in-person care, or decisions someone has to answer for.Tasks with an AI score under 25 go here.More in the glossary
- O*NETThe US government’s free database of about 1,000 jobs, listing each job’s tasks and how much each one matters.Run for the US Department of Labor. Every score here starts from its task lists (version 31.0).onetcenter.org · More in the glossary
- Human in the loopA named person who checks or approves AI’s work before it takes effect, and answers for the result.Lesson 10: Risks and rules · More in the glossary
Every term is explained in the glossary.
Try this
List the ten tasks your team repeats most each week, with roughly how many times each happens and who does it. Run the five questions on each and write one of the four outcomes beside it. Circle the two that are frequent, easy to check and cheap to get wrong. Those are your first pilots: Lesson 12 shows how to run them. Keep the list; you will use it again in Lesson 5 to draw your team.
Sources
This lesson uses no outside figures. Task ratings are our own, printed live from the current release; how they are made is on our method page for “Can AI do it?”. Task descriptions are from O*NET, the US Department of Labor’s occupation database.