Applied AIUpdated 12 min read
AI for small business: 30 tasks ranked by effort and impact

In this article
Three out of four small businesses say they already use AI. In Goldman Sachs' survey of 1,256 owners in its 10,000 Small Businesses program, 76% use it, but only 14% say it's fully built into their core operations (Goldman Sachs, 2026). The official numbers are lower: 18% of U.S. businesses used AI in any business function between November 2025 and January 2026 (U.S. Census Bureau, 2026). In Latin America, the ILIA index puts adoption at 6.8% of small businesses versus 27.6% of large ones (CENIA and ECLAC, 2026).
The gap between "we use it" and "it runs part of the business" is rarely about the tool. It's that nobody decided which task it's for, what data it gets and who checks the output. Here's that part, worked out for 30 tasks.
How to read the tables
Effort is what you need to get it running; impact is how much time or money it moves in a company of 10 to 50 people. These are our estimates: adjust them with your own numbers.
| Rating | Effort | Impact |
|---|---|---|
| Low | An existing AI tool and a saved prompt. Up and running in an afternoon. | Saves a few minutes here and there or improves a piece of writing. |
| Medium | Connecting two tools (with Zapier or Make, for example) or preparing reference documents. A few days. | Saves one person several hours a week. |
| High | Needs the company's data organized in one system and connected. Weeks. | Moves sales, collections or inventory, or saves several people hours. |
The tools are examples. ChatGPT, Claude and Gemini are about equally good at drafting, summarizing and reading documents. Copilot makes sense if your company runs on Excel, Outlook and Teams; Gemini, if it runs on Google Workspace. Pick one and get good at it before trying another.
The 30 tasks by department
Sales
| Task | What it does | Typical tool | Effort | Impact |
|---|---|---|---|---|
| 1. Turn the sales rep's voice memo into an order | Transcribes the audio and sorts it into customer, products, quantities and delivery date, ready to paste into the order. | ChatGPT or Gemini with audio | Low | Medium |
| 2. Draft a quote from the customer's email | Pulls out what they're asking for and drafts the quote using your price list and terms, saved in a project. | A ChatGPT or Claude project | Low | High |
| 3. Prep for the visit or call | Sums up the customer's recent orders, complaints and open items in five lines. | Copilot in Excel or your CRM's AI | Medium | Medium |
| 4. Follow up on quotes with no reply | Every three days, lists open quotes and drafts the follow-up. | Zapier with an AI step | Medium | High |
| 5. Suggest each customer's reorder | Works out what they're due to order based on what they buy and how often. | AI connected to orders and inventory | High | High |
Customer service
| Task | What it does | Typical tool | Effort | Impact |
|---|---|---|---|---|
| 6. Answer common questions on WhatsApp or Messenger | Handles hours, shipping, payment options and the catalog around the clock, and hands off to a person when it doesn't know. | Meta Business Agent or a chat agent | Medium | High |
| 7. Sort incoming messages | Tags each message as an order, complaint, quote request or question and routes it to the right person. | Zapier or Make with an AI step | Medium | Medium |
| 8. Draft a reply to a complaint | Suggests a reply in your company's voice based on the case; a person reviews and sends it. | ChatGPT or Claude | Low | Medium |
| 9. Tell customers where their order is | Looks up the actual order status and delivery date and answers in the chat without anyone digging. | Agent connected to the order system | High | High |
| 10. Summarize the week's complaints | Reads exported messages and returns the five most common reasons, with examples. | ChatGPT or Claude with the file | Low | Medium |
Operations and inventory
| Task | What it does | Typical tool | Effort | Impact |
|---|---|---|---|---|
| 11. Turn a supplier invoice into an inventory receipt | Reads the PDF or photo and returns the line items (SKU, quantity, cost) ready to load without retyping. | Gemini or ChatGPT with document upload | Low | Medium |
| 12. Write up a process | Turns a voice memo from the person who does the job into step-by-step instructions with owners and checks. | Claude or ChatGPT | Low | Medium |
| 13. Find slow-moving stock | Matches the inventory export against sales and lists what hasn't moved in 90 days, with its cost. | Copilot in Excel or ChatGPT data analysis | Low | High |
| 14. Draft the purchase order | Suggests what and how much to buy based on recent sales, stock on hand and each supplier's lead time. | AI connected to sales, inventory and purchasing | High | High |
| 15. Alert when an item drops below the minimum | Checks stock daily and sends purchasing a note with what needs to be ordered. | Zapier or Make on the inventory sheet | Medium | Medium |
Finance and collections
| Task | What it does | Typical tool | Effort | Impact |
|---|---|---|---|---|
| 16. Write payment reminders | Drafts a message based on how overdue the invoice is: friendly at 15 days, firm at 45, with the invoice and balance. | ChatGPT or Claude with a template | Low | High |
| 17. Reconcile the bank account with invoices | Suggests which payment matches which invoice; a person approves the matches and reviews the unclear ones. | Your accounting software's AI or Copilot in Excel | Medium | High |
| 18. Read expense receipts | Pulls the date, vendor, amount and tax from a photo of the receipt and suggests the account. | Accounting app with receipt capture | Low | Medium |
| 19. Explain the monthly P&L | Reads the income statement and explains the five biggest changes from last month in plain English. | ChatGPT, Claude or Gemini with the file | Low | Medium |
| 20. Forecast cash for the next 13 weeks | Combines receivables coming due, supplier payments and payroll to show which week you'll run short. | AI connected to receivables, purchasing and bank data | High | High |
Admin and HR
| Task | What it does | Typical tool | Effort | Impact |
|---|---|---|---|---|
| 21. Write the job post and interview questions | Starts from the role's duties and drafts the posting plus ten questions. | ChatGPT, Claude or Gemini | Low | Low |
| 22. Take meeting notes | Transcribes the meeting and lists decisions and action items with owners and due dates. | Copilot in Teams, Gemini in Meet or Zoom | Low | Medium |
| 23. Answer employees' policy questions | Handles PTO, expense and leave questions from the handbook and policies you load into it. | A Claude project, a custom GPT or NotebookLM | Medium | Medium |
| 24. Review a contract before your lawyer does | Flags the clauses worth a closer look: term, penalties, auto-renewal, exclusivity. | Claude or ChatGPT | Low | Medium |
| 25. Build onboarding for a role | Puts together the first-week plan and a short quiz from the written process. | ChatGPT, Claude or Gemini | Low | Low |
Leadership
| Task | What it does | Typical tool | Effort | Impact |
|---|---|---|---|---|
| 26. Get a Monday briefing | Ten lines on what each rep sold, what was collected and what's still open. | AI connected to sales, receivables and orders | High | High |
| 27. Ask your data questions in plain English | Answers questions like "which customers bought less this quarter?" without building a report by hand. | AI connected to the company database | High | High |
| 28. Prep for the bank or partner meeting | Turns the quarter's numbers into a draft deck. | Copilot in PowerPoint, Gemini in Slides or Claude | Low | Medium |
| 29. Keep an eye on competitors | Looks up published prices, new products and customer reviews, with links so you can check. | Deep research in ChatGPT, Gemini or Claude | Low | Medium |
| 30. Write your AI use policy | Drafts one page covering approved tools, data that never gets pasted in and who reviews the output. | ChatGPT or Claude, reviewed by a lawyer | Low | High |
The matrix: where to start
Plot all 30 tasks on an effort vs. impact grid and here's how they fall:
| Quadrant | Tasks | What to do |
|---|---|---|
| Low effort, high impact | 2, 13, 16 and 30 (4 tasks) | Start here this week. All you need is an AI chat and a well-written prompt. |
| Low effort, medium impact | 1, 8, 10, 11, 12, 18, 19, 22, 24, 28 and 29 (11 tasks) | Pick two or three in the areas that waste the most time today. These are the ones that build the habit. |
| Medium effort, medium or high impact | 3, 4, 6, 7, 15, 17 and 23 (7 tasks) | One at a time, starting in week three, once you know how to write prompts that work. |
| High effort, high impact | 5, 9, 14, 20, 26 and 27 (6 tasks) | A chat window won't get you there: these need the company's data in one place. |
| Low effort, low impact | 21 and 25 (2 tasks) | Use them when they come up; they're not worth planning. |
The six high-effort, high-impact tasks have one thing in common: none of them depends on better writing. They all need to know what's in the warehouse, who owes what or what each customer bought, using today's data.
A 30-day plan
- Days 1 to 3: rules and one toolWrite the one-page policy (task 30) and pick a single tool on a business plan for the whole team. That way you've settled what data can go in before someone pastes your receivables into a free chat account.
- Days 4 to 7: measure before you change anythingTake the four low-effort, high-impact tasks and write down how long they take today: minutes per quote, hours a week on payment reminders. Without that number, you won't know in a month whether it worked.
- Days 8 to 14: one prompt per taskWrite the prompt for each task with a real example of input and output, save it in a shared project and test it on ten cases. Fix the prompt, not the output by hand.
- Days 15 to 25: one medium-effort taskAdd one task that connects two tools, like following up on quotes (task 4) or WhatsApp FAQs (task 6). Decide who reviews the output and when it gets handed to a person.
- Days 26 to 30: count and decideCompare before and after, drop whatever didn't save time and list the tasks that stalled because the data was scattered across spreadsheets, chats and email. That list is your next step.
The risks and how to handle them
Customer data
Data privacy and security is the barrier owners mention most: 50% named it in the Goldman Sachs survey (2026). The real-world risk is someone pasting a customer list, an aging report or a contract into a free personal account. Business plans from the major vendors don't use your data to train their models by default; OpenAI, for example, says so for ChatGPT Business, ChatGPT Enterprise and its API (OpenAI, 2026). Check that term with whichever vendor you pick, and keep Social Security numbers, health information and bank details out even if the plan allows it.
Made-up answers
A language model can state something false with total confidence. In 2024, a Canadian tribunal ordered Air Canada to pay a passenger after its website chatbot promised a discount the airline didn't offer: the company is responsible for what its chatbot says (Moffatt v. Air Canada, 2024). The practical rule is the one daily users already follow: 86% treat AI output as a starting point, not a final answer (Microsoft, 2026). Anything that reaches a customer with a price, a deadline or a term gets checked by a person, or comes from your system's data and not from the model's memory.
Use policy
The AI use policy doesn't need to be longer than one page. Before you send it to the team, check that it answers these points:
- Which tools are approved, and on which account (the company's, not a personal one).
- What data never gets pasted in: ID numbers, health information, bank accounts, passwords.
- What can be used with customer data, and on which plan.
- Who reviews AI output before it reaches a customer, a supplier or a bank.
- Which decisions stay with people: approving prices, discounts, credit or terminations.
- Who to tell if someone pasted something they shouldn't have.
- How often it gets reviewed (every six months is reasonable), and a lawyer's review of the personal data section.
When AI isn't the place to start
At some companies, AI isn't the first job. Wait, or start somewhere else, if:
- Nobody will review what it produces. Without review, every mistake reaches the customer and the time saved goes into fixing it.
- The process isn't clear yet. If everyone quotes or collects their own way, AI repeats the mess faster: write the procedure first.
- The task happens only a few times a month. Writing and testing a good prompt takes hours that two uses won't pay back.
- You don't have a business account and the task involves customer data. Until you do, use AI only with public information.
When a standalone AI chat isn't enough
Most AI use today is a chat window open next to the real work. In the U.S. Census Bureau study, the top generative AI tasks are writing, analyzing documents and searching for information, and 65% of firms that use it stick to three tasks or fewer (U.S. Census Bureau, 2026). That's a good start, and it hits a ceiling fast.
You feel that ceiling on the high-effort tasks. Across 20,000 workers who use AI, Microsoft found that organizational factors (culture, manager support, how work is set up) carry twice the weight of individual ones in the impact people report: 67% vs. 32% (Microsoft, 2026). It's an association, not proven cause and effect, but it points the same way as the matrix.
A chat window is no longer enough if any of these sound familiar:
- Every time you use AI, you export and paste the same data: inventory, receivables, the price list.
- The right answer depends on what happened today: stock levels, payments received, orders shipped.
- You want the AI to do something, like create the order or send the appointment reminder, not just draft it.
- Sales, inventory and collections data live in separate tools that don't talk to each other.
At that point, the work left to do is about data: orders, inventory, receivables and customers in one system, with permissions per person. With that in place, AI reads the real stock level instead of last week's spreadsheet, and tasks 5, 9, 14, 20, 26 and 27 become an AI integration on top of data that already exists.
If you're not there yet, don't skip the first few weeks. The low-effort tasks teach you which questions you'd ask your data, and that list is exactly what you need to design the system.
Sources
- Goldman Sachs 10,000 Small Businesses Voices, "Survey: Small Businesses Embrace AI — But Need Training and Support to Fully Harness It" (Business Wire), 2026 (opens in a new tab)
- Bonney, Breaux, Dinlersoz, Foster, Haltiwanger and Pande, U.S. Census Bureau, "The Microstructure of AI Diffusion" (CES 26-25), 2026 (opens in a new tab)
- CENIA and ECLAC, "ILIA 2026: 4th edition of the Latin American Artificial Intelligence Index" (in Spanish), 2026 (opens in a new tab)
- Microsoft, "2026 Work Trend Index," 2026 (opens in a new tab)
- Meta, "Meta Business Agent," 2026 (opens in a new tab)
- American Bar Association, "BC Tribunal Confirms Companies Remain Liable for Information Provided by AI Chatbot" (Moffatt v. Air Canada), 2024 (opens in a new tab)
- OpenAI, "Enterprise privacy at OpenAI," 2026 (opens in a new tab)