how to set up claude for a small business.
on may 13, 2026, anthropic launched claude for small business. it comes free with the pro, max, and team plans: 15 ready-to-run workflows for things like invoice chasing, month-end close, cash position, and contract review, plus connectors to quickbooks, hubspot, paypal, google workspace, and microsoft 365.
which sounds like the problem is solved. it is not. anthropic handed every owner the same 15 generic templates and then left. the templates do not know your books, your pipeline, or the way you actually quote a job. the gap between "claude exists in my account" and "claude runs a piece of my business every week" is setup work, and that gap is where most AI money dies.
i spent ten years in enterprise sales watching companies buy software that never got implemented. the pattern is identical here, just smaller. so this is the setup guide i wish every owner read before touching anything.
first, the number nobody puts in the launch post
most AI initiatives fail to meet expectations, and the research is blunter than the headlines: MIT's 2025 study found 95 percent of generative AI pilots show no measurable return, and RAND puts the broader failure rate above 80 percent. in 2025, 42 percent of companies abandoned most of their AI initiatives, up from 17 percent the year before. only 7 percent of organizations have AI fully deployed and integrated.
and here is the part that matters: the most-cited cause across that research is not the model. it is the workflow. the redesign of how the work moves, not the AI tool itself, decides whether anyone sees a result. people buy the subscription, paste a few prompts, and go back to doing the work by hand. they bought a tool and skipped the system.
so the goal of a claude setup is never "we have AI now." the goal is one specific piece of work that used to be done by hand and is now drafted by claude every single week, in a way you can see and check. i wrote more about the failure pattern in why your ai is not working.
what to automate first
the research on what businesses actually pay AI consultants for ranks the demand clearly. highest first:
- workflow and process automation. connect the tools, automate the handoffs, kill the manual drag: crm updates, scheduling, internal reporting, the stuff that eats your tuesday.
- lead generation and response. research prospects, qualify, draft outreach, follow up. this is the one businesses pay the most for, because it touches revenue directly.
- customer service and support triage. 90 percent of CX leaders report positive ROI from AI support tools.
- document and invoice processing. often the highest measured ROI: cost per invoice, cycle time, error rate.
- marketing and content. about three in four marketers already use AI here, per Salesforce's latest count.
if you only build one thing, build lead response. the money most businesses lose is not from bad leads. it is from slow or missing follow-up. a setup where claude drafts a personalized first reply to every inquiry, plus the follow-up sequence, recovers revenue that was already leaking out. one recovered job often pays for the entire setup.
the selection rule, straight from the businesses that succeeded: start with high-volume, repetitive, low-risk tasks where mistakes are cheap and feedback is instant. your first build is never payroll. it is the quote email you write the same way nine times a week.
the approve-everything model
the loudest objection from owners is trust, and it is the correct objection. one business owner put it this way: "in no world would i give claude or any AI agent direct write access to financial operations." agreed. you should never set it up that way.
the model that works is draft and approve. claude does the reading, the sorting, and the drafting. a human clicks send. nothing posts, sends, or pays without your approval. that one rule changes the risk math completely: the worst case of a bad draft is thirty seconds of your time, while the worst case of an action taken without you is a real mess in your real accounts.
it also fixes the second trust problem, the black box. teams reject AI they cannot see into; one of the most common failure stories is a team spot-checking every output and doing the work twice. a proper setup is legible: the instructions live in plain-text files you can open and read, so you always know why it did what it did. i explain how those files work in claude skills, explained.
what a working setup looks like
concretely, a small-business claude setup has four parts:
- one mapped process. the single highest-volume, lowest-risk task, written down step by step the way your best person does it.
- instruction files claude follows. the process saved as plain-text files in your account, so the quality is repeatable instead of depending on how well you prompt that day.
- connected accounts. the quickbooks, hubspot, or workspace connectors wired to your actual data, so claude works on real inputs instead of pasted snippets.
- approval gates. claude drafts, you approve, every time.
the payoff, honestly stated: IDC and Microsoft's study measured an average of $3.70 back for every $1 spent on generative AI, realized over roughly 13 months, not weeks. the near-term win is hours, and how many depends on how much of your week is repetitive work. both numbers come from businesses that picked one workflow and finished it, not from businesses that bought five tools.
one more thing while it is timely: set up claude on the interactive side, where you work with it directly. that side stays on your flat subscription after anthropic's june 15 billing change; always-on background agents are the side that gets metered. details in the june 15 note.
do it yourself, or have it done
everything above is doable yourself if you have the hours and the patience to map your own process honestly. plenty of owners do not, which is why the most common comment under my videos is some version of "can i hire you to help me set this up?" if you go the hired route, read the buyer's guide first, because the market is full of hidden pricing and rented systems.
want the diagnostic first?
the systems diagnostic is $500, the price is on the page, and you get a written map of the one process in your business worth automating first, with the build plan to do it. you decide on your own schedule.
get the $500 diagnostic