If your business already uses AI to tidy up an email, the more interesting question is whether it can help with the work around that email. Can it find the relevant customer information, prepare a useful draft and leave it somewhere you can review?
That is why Claude’s small-business integrations are worth a look. The useful test is a task from your own working day, rather than how many connectors appear in the announcement.
What Anthropic announced
On 15 September 2026, Anthropic announced an expansion of Claude for Small Business. Its release post describes 43 workflows and 27 new integrations, including tools such as Xero, Shopify, Square and Zapier. It also announces training events and partner webinars.
The update follows the May launch, which packaged connectors and workflows around common small-business tools. This article is Kerbside’s assessment of the announcement, not a hands-on review of every integration. Features and availability should be checked against the current product documentation before purchase.
The opportunity is less copying between tools
Our view is that the most promising use is reducing the preparation around a task. Consider a customer update: read the job record, check outstanding requests, find the last conversation, then write the reply.
If an assistant can retrieve the correct information and prepare a draft for review, it may remove some of that searching and copying. The benefit depends on the records being useful and the connector exposing the information you need.
For a local business with an inbox and a simple job list, the first improvement might still be keeping that list up to date. A connection to incomplete records gives the assistant incomplete context.
Test one job with a known answer
Choose a recurring task where you can judge the result. A project summary is easier to assess if you know the project well. A new report containing unfamiliar financial figures is harder to check and makes a poor first trial.
Try a controlled example, using invented or appropriately anonymised information where possible. Compare the output with what you would produce yourself. Check which records it used, whether it missed anything and how much editing was needed.
For a proposed customer-update workflow, the checks could be:
- It selected the right customer and project.
- It used the latest approved information.
- It marked missing details rather than inventing them.
- It saved a draft without sending it.
- A person could find and correct the sources.
These are suggested acceptance checks, not claims that a particular connector provides all five. The trial should establish that.
Check the permission before pressing Connect
Reading records and changing records are different responsibilities. Look at the permission request and the actions supported by the connector. Decide whether staff need approval before sending a message, updating a customer record or carrying out another action.
Check the cost of the complete setup too: the assistant subscription, any connected tools and the work needed to maintain the process. Avoid comparing only the headline subscription price with the time spent on one successful demonstration.
Who should look at it now?
It is worth investigating if you already use one of the supported tools and have a specific, repetitive task that requires gathering information or preparing text. Give it a bounded trial and count the checking time.
If you are still deciding where AI belongs, start with one everyday admin task. If the standard connector gets you part of the way but misses a permission or workflow requirement, read when custom AI makes sense.
You do not need to adopt every new capability. A single workflow that survives an ordinary busy week is a better basis for the next decision.
Working on something similar?
Discuss a Claude integration ↗