The AI Business Setup Checklist

Phase 1: Map the business

• +List the five to ten workflows that consume the most staff time each week.
• +For each workflow, write down who runs it, what triggers it, and what done looks like.
• +Mark which workflows are judgment heavy and which are repetitive. Repetitive goes first.
• +Identify where the work already lives: email, CRM, project tool, shared drives, spreadsheets.
• +Pick the one workflow where saved time would matter most. That is your first target.

Phase 2: Pick the platform

• +Shortlist the major business platforms: ChatGPT, Claude, Gemini, Microsoft Copilot, and Grok.
• +Check which one connects to the tools your team already uses every day.
• +Compare business plans on admin controls, data handling, and user management, not just model quality.
• +Read each vendor's data policy for its business tier and confirm how your company data is used.
• +Decide seat count for the first rollout. Start with the people who will actually use it weekly.
• +Pick one platform. Running two from day one splits attention and doubles setup work.

Phase 3: Build the foundation

• +Set up the business workspace under a company owned domain, not someone's personal account.
• +Assign at least two workspace admins so access never depends on one person.
• +Turn on single sign on or two factor authentication where the plan supports it.
• +Define roles: who can add members, who can add integrations, who can only use.
• +Set default permissions to least access. Grant more only when a workflow needs it.
• +Document the setup: who is admin, what is enabled, where credentials are stored.

Phase 4: Organize company knowledge

• +Gather the documents that explain how your company works: services, pricing rules, processes, policies, tone.
• +Turn tribal knowledge that lives only in people's heads into short reference documents.
• +Cut outdated material. Wrong information in the system is worse than no information.
• +Organize documents by workflow, not by department, so each task gets what it needs.
• +Decide what must never go into the platform: client secrets, regulated data, anything under an NDA you have not reviewed.
• +Give every document an owner and a review date so the knowledge stays current.

Phase 5: Build reusable assets

• +Turn your best repeated prompts into named, shared assets: projects, custom instructions, skills, or agents, depending on what your platform supports.
• +Write standing instructions that encode your voice, formats, and rules once, so nobody retypes them.
• +Build templates for your most common outputs: proposals, reports, client emails, briefs.
• +Name assets so the team can find them: workflow first, version second.
• +Keep one shared library. Private duplicates drift apart within weeks.

Phase 6: Connect approved systems

• +List the systems your target workflow touches: CRM, email, calendar, project tool, file storage.
• +Check which of those have supported connectors on your platform. Availability may vary by plan.
• +Connect only what the first workflow needs. You can add more later.
• +Review what each connection can read, and where applicable write, before enabling it.
• +Use scoped access or a dedicated service account where the platform and the tool support it.
• +Record every connection in one place: what it is, who approved it, what it can touch.

Phase 7: Build the first workflow

• +Take the target workflow from Phase 1 and write it as steps the AI performs and steps a person performs.
• +Define the exact inputs the AI receives and the exact output format it must produce.
• +Set human approval points anywhere output leaves the company or money is involved.
• +Build it from the assets in Phase 5 and the connections in Phase 6, not from scratch.
• +Write a one page runbook: how to trigger it, what good output looks like, what to do when it fails.

Phase 8: Test with real work

• +Run the workflow on ten to twenty real, recent examples, not on made up samples.
• +Compare the AI output against what your team actually shipped for those same cases.
• +Log every failure and sort it: missing knowledge, unclear instructions, or the wrong task for AI.
• +Fix knowledge and instructions first. Most failures trace back to Phase 4 and Phase 5.
• +Set pass criteria before you start, and do not go live until the workflow meets them.
• +Keep the human review step in place until the error pattern gets boring.

Phase 9: Train the team

• +Train on your workflows and your assets, not on generic prompting tips.
• +Show each role the two or three tasks where the platform saves them real time this week.
• +Name an internal owner who answers questions and collects what is working.
• +Set clear rules for what may and may not go into the platform, and repeat them often.
• +Make usage visible: share wins, failures, and improved assets in one channel.

Phase 10: Govern and expand

• +Review usage monthly: what is used, what is ignored, what failed.
• +Retire assets nobody uses and update the ones people rely on.
• +Audit permissions and connections quarterly. Remove access nobody needs.
• +Track vendor updates. Platform capabilities change, and your setup should change with them.
• +Pick the next workflow from your Phase 1 map and repeat Phases 4 through 9.
• +Budget standing time for maintenance. An unmaintained setup decays quietly.

Where this checklist leads