Claude for Small Business now ships 43 ready-to-run workflows across 37 partner connectors, after a September 15, 2026 expansion that added 27 integrations including Shopify, Xero, Gusto, Square and Stripe. Anthropic says it has been installed more than 900,000 times. Most owners should switch on three or four of the 43, not all of them.
This is a practitioner's read of what the package does, which workflows earn their place, and where it stops and custom work starts.
What is Claude for Small Business?
Claude for Small Business is a free plugin for the Claude desktop app that bundles connectors and pre-built workflows for finance, operations, sales, marketing, HR and customer service. It runs on any paid Claude plan — Pro, Max, Team or Enterprise — and every workflow starts in a mode where you approve before anything sends, posts or pays.
It launched on May 13, 2026 with seven tool partners: QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace and Microsoft 365. That release carried 15 workflows and 15 skills, almost all of them back-office: closing the month, chasing invoices, planning payroll.
The September release moved into the front of the house — lead response, proposals, social content — and added connectors for Salesforce, Shopify, Square, Stripe, Xero, Gusto, Zoom, Atlassian, TikTok Ads and Zapier, among others. One naming note: Anthropic's pricing page now says "Claude Cowork is now just Claude", but the setup docs still refer to Cowork.
Which of the 43 workflows are actually worth switching on?
Start with the workflows that read data and summarise it, not the ones that act on it. A brief that pulls from your books and CRM is low-risk and shows you quickly whether your data is clean enough for anything more ambitious. Workflows that send, pay or sign belong later.
The five Anthropic features most prominently, with the connectors each can draw on:
| Workflow | What it does | Connectors |
|---|---|---|
| Monday Brief | One-page weekly summary of cash, sales, pipeline and overdue invoices | 37 |
| Speed to Lead | Answers and qualifies inbound enquiries, books time, logs to CRM | 15 |
| Proposal Builder | Drafts proposals from your pricing and past work | 17 |
| Social Content Engine | Plans and drafts social posts on a weekly cadence | 16 |
| Close the Month | Reconciles and prepares month-end for your accountant | 14 |
Monday Brief is the right first install for almost everyone. It is read-only, it touches every system you connect, and its output is easy to check against what you already know. If the brief is wrong, the problem is your data, and you have learned that for free.
Speed to Lead suits businesses where response time decides the sale — trades, clinics, agencies, anyone whose enquiries arrive out of hours. It is also the workflow that most needs a human checkpoint, because it speaks to customers in your name.
Close the Month suits owners who already have tidy books. It prepares; it does not sign off. The Academy guide says plainly that "the calls that matter — what to charge, what to sign, what to send your accountant — are yours and your professionals' to make."
Proposal Builder and Social Content Engine are the most variable. Their output is only as good as the past work and voice examples you give them, and both produce drafts, not finished artefacts.
What does it cost to run?
The plugin itself carries no separate fee; you pay for the Claude plan it runs on. At published prices, Pro is $20 a month ($17 billed annually), and Team starts at $25 per seat monthly ($20 annually) for teams of 2 to 150. Anthropic recommends Team for any business with more than one person.
There are two costs the plan price does not show. Multi-step workflows consume more of your usage allowance than chat does, so heavy use can push a solo owner from Pro towards Max. And some features depend on your plan in the connected tool — the Academy guide notes that staging sends or generating designs may need a higher tier there.
For a fuller treatment of where AI running costs actually come from, we worked the numbers through in what an AI agent actually costs to run.
Is your business data used to train Claude?
On Team and Enterprise plans, no — Anthropic states that by default it does not train on business data on those plans. On Pro and Max, which are consumer plans, training is a setting each user chooses. A solo owner on Pro should check that setting before connecting their accounting system.
The consumer terms update of August 2025 introduced that choice for Free, Pro and Max. If a user opts in, data is retained for five years; if not, the existing 30-day retention applies. That difference is a good reason, on its own, to run any multi-person business on Team.
Permissions follow the source system. Anthropic's launch post puts it directly: "If an employee can't see something in QuickBooks or Drive today, they can't see it through Claude." That is only reassuring if your existing permissions are right — which is worth auditing before anyone connects anything.
Where do the packaged workflows stop?
They stop at your data quality, at the last irreversible step, and at anything specific to how your business works. The packaged workflows assume standard setups; multi-entity accounting, custom CRM fields and approval chains with more than one signatory are where they start to strain.
Data quality is the first wall. A workflow that reads a CRM full of duplicates writes a confident brief built on duplicates. Anthropic's own field notes from its small business tour describe an owner who "caught Claude using a floor-area multiplier instead of real wall measurements" on a bid. That kind of error is caught only by someone who knows the job.
The last step is deliberately left to you. Anthropic's September post notes that "some workflows stop short of the last step" — Claude stages payroll in Gusto for you to submit rather than submitting it. That is the right design, and it means the time saved is in preparation, not in removal of the task.
Accuracy is high, not perfect. One five-person customer quoted in the September announcement describes it as getting the result right "90% of the time". For a draft, that is excellent. For anything that pays, posts or signs unattended, a one-in-ten miss rate is the reason approval mode exists.
Your process is not the template. Anthropic includes /build-connector and
/build-agent for gaps, and they work for simple cases. Once the gap is a
bespoke approval chain, a system with no connector, or logic that spans several
tools, that is a build, not a configuration.
When do you need custom automation instead?
When the workflow is specific to your business, runs unattended at volume, or has to meet an audit requirement. The packaged workflows are built for an owner approving each run at a desk; a process handling hundreds of items a day without supervision needs evaluation, logging and failure handling the plugin does not provide.
The building blocks underneath are the same. Agent Skills came out of beta on the Claude API on August 19, 2026, and the Claude Agent SDK gives developers the same harness, connected to business systems through MCP servers. A custom build is the plugin's approach, made specific and made observable.
That is the work our AI automation and agent engineering practice does. More often than not, though, the first project is underneath the automation: reconciling the CRM, the books and the spreadsheets so an agent has one version of the truth. That is business systems and data foundations work, and it is what decides whether any of the 43 workflows produce something you can trust.
We set out which kinds of automation reliably survive contact with real operations in which AI automation applications actually reach production.
Should a small business use it at all?
Yes, if you start narrow and treat its output as a draft until it has earned trust. The US Census Bureau's survey shows overall business AI use between 17% and 20% through early 2026, and that use rose among firms with 20 or more employees but "didn't change significantly" below that size.
Those Census Bureau figures cover December 2025 to May 2026. They suggest the gap is not interest but implementation: Anthropic's own exit surveys from its tour found nearly two-thirds of attendees asking for hands-on help after the workshop.
A sensible sequence for the first month:
- Week one: install, connect accounting and CRM, run Monday Brief, and check it against what you know.
- Week two: fix whatever the brief exposed — duplicate contacts, miscategorised transactions, stale pipeline stages.
- Weeks three and four: add one workflow that acts, in approval mode, and count how often you change its output before approving.
That last number is the one that tells you whether to let a workflow run on its own. If you are changing one draft in three, it is not ready; if you are changing one in thirty, it may be.
For teams that want staff using these tools well rather than just having them installed, our AI enablement and training work covers the habits that matter most: checking output, writing clear instructions, and knowing when to stop and escalate.
The short version
Claude for Small Business is a well-designed starting point with sensible defaults: approval before action, permissions inherited from your tools, and a fallback to spreadsheets when a connector is missing. Install Monday Brief first, fix what it reveals, and add workflows that act only once the drafts stop needing correction.
The ceiling is your data and your process, not the model. If you want a costed view of which of your workflows should be packaged, which should be built, and which should not be automated at all, that is what the automation blueprint produces.