← All insights

WorkflowsMicrosoftGoogleAnthropicOpenAI24 October 2026· 5 min read

7 AI automations a small business can run today

Seven practical AI automations for small businesses, from sorting enquiries to weekly reports, with the tools that run them and how to start safely.

The best AI automations for a small business are the routine tasks that happen often and are easy to check: sorting enquiries, drafting replies, turning meetings into tasks and building weekly reports. In 2026, Microsoft, Google, Anthropic and OpenAI all added agents that carry out work across your apps. You can now set many of these up without a development team.

How quickly businesses are adopting AI agents

Microsoft’s 2026 Work Trend Index found that the number of active agents in Microsoft 365 grew 15 times in a year. In the same report, 58% of AI users surveyed said they produce work they could not a year ago. The survey covered 20,000 workers using AI across 10 countries.

For a small business, the opportunity is practical. Take routine admin off people’s plates, so they spend more time with customers.

Seven AI automations worth starting with

1. Sort and route new enquiries

Each contact form submission or email is read, tagged by type and urgency, and sent to the right person. Obvious spam is set aside. A small, low-cost model is usually enough for this.

2. Draft replies to common questions

For questions about pricing, availability or opening hours, the agent drafts a reply from your own information. A person reviews and sends it. Replies go out faster and stay consistent.

3. Turn meetings into tasks

Meeting notes become a summary, a task list with owners and a draft follow-up email. Built-in note takers, such as Google Meet’s “Take notes for me”, already handle the first step.

4. Pull data out of documents

Invoices, receipts, purchase orders and forms are read, and the key fields go into a spreadsheet or accounting tool. The agent flags anything it is unsure about for a person to check.

5. Build a weekly report

Every Monday, the agent gathers figures from your analytics, sales and enquiry tools, then writes a short summary of what changed. Claude’s scheduled tasks can run this on a set schedule, and weekly reports are one of their common uses.

6. Monitor reviews and draft responses

New reviews are collected, sorted into positive and negative, and given a draft response. Negative reviews go straight to the owner to handle personally. We build this as review replies.

7. Research leads before a call

Before a sales call, the agent gathers public information about the company and its website, then writes a one-page brief.

The tools that run these automations

Tool What changed in 2026
Microsoft Copilot Cowork Generally available worldwide from 16 June. Needs a Microsoft 365 Copilot licence, with usage billed in Copilot Credits.
Google Workspace Studio Custom starters and steps, webhooks and third-party integrations (beta), rolling out from 17 September.
Claude Cowork and chat merged into one Claude on 16 September, rolling out to Pro, Max and Ultra first. Scheduled tasks are on all paid plans.
ChatGPT ChatGPT Work launched on 9 July as an agent that can act across your apps and files.

The right tool depends on where your work already happens. A business on Google Workspace can start with Workspace Studio. One on Microsoft 365 can look at Copilot Cowork. Work that spans several apps, or needs to run without anyone starting it, may suit a dedicated automation tool or a small custom tool built on an AI model’s API. If you would like one built for you, see our AI automation projects, such as enquiry triage.

Shared standards make agents easier to connect

The Model Context Protocol (MCP) is an open standard for connecting AI tools to apps and data. Anthropic donated MCP to the Agentic AI Foundation in December 2025. The foundation sits under the Linux Foundation. It was co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg.

ChatGPT, Gemini, Microsoft Copilot and other AI products have adopted MCP. In practice, more of the software you already use can connect to AI agents without custom work.

How to start an AI automation safely

  1. Pick one task that happens often and is easy to check, such as sorting enquiries.
  2. Keep a person in the loop. Let the agent draft and suggest. Have a person approve anything sent to customers or involving money.
  3. Set a spending limit on every AI account the automation uses.
  4. Keep a log of what the agent did, so you can review and improve it.
  5. Measure the time saved after a month, then choose the next task.

The main tools now build in some of these controls. Claude asks before taking an action by default. In Workspace Studio, admins can set human approval requirements for sensitive steps. Copilot Cowork is off by default, and admins can set spending limits for the whole organisation, groups or individual users.

References

8 official sources
  1. 2026 Work Trend Index: Agents, human agency, and the opportunity for every organization Microsoft WorkLab · 5 May 2026
  2. Copilot Cowork is now generally available Microsoft Copilot Blog · 16 June 2026
  3. Automate workflows with custom starters and steps, third-party integrations, and webhooks in Workspace Studio Google Workspace Updates · September 2026
  4. Claude Cowork and chat are now one Claude Anthropic (Claude blog) · 16 September 2026
  5. Schedule recurring tasks in Claude Cowork Claude Help Center
  6. ChatGPT is now a partner for your most ambitious work OpenAI (news feed) · 9 July 2026
  7. "Take notes for me" in Google Meet Google Meet Help
  8. Donating the Model Context Protocol and establishing the Agentic AI Foundation Anthropic · 9 December 2025

Further reading