You open Claude to draft the variance commentary. It asks what your company does. You explain it again, the same way you did last week.

Or the opposite happens. You set Claude up a year ago, and it still writes for a role you have since left, or brings context from one piece of work into another where it does not belong.

Both problems come from the same place: settings nobody has looked at in a while.

Balanced AI Insights starts a new format today. Every week, I will share one AI workflow or playbook for a real finance task, tested on real work, followed by the AI updates worth your attention.

I will keep it realistic and show where each workflow can break.

Most workflows will use Claude, and they assume your Claude account is set up a certain way. So the first edition covers the basics: the settings that make everything I post next work on your account.

I am doing this review on my own account this week too. Claude has added and renamed several settings since I last went through mine.

Keep this edition. Every future edition will link back to it under "Don't know where to start?"

Today I'll show you how to set up your Claude account for the workflows ahead, using two prompts: one for the initial setup and one for review.

Watch me go through my own settings (83 seconds).

Your Settings may look different. Claude is moving accounts to a new Settings layout, and accounts switch over at different times. I list the new path first, with the older one in brackets. If a path doesn't match your screen, look for the setting by name. On a Team or Enterprise plan, your company's admin controls some of these settings; I flag where.

Turn on web search first (Settings → Capabilities). Claude needs it to check the current settings; without it, it may suggest settings that no longer exist. Then open a new chat outside any project and paste the prompt that fits you. Claude interviews you, writes your Instructions for Claude, and recommends which features to turn on for your plan and your work. Steps 2 to 5 show where each setting lives, so you can check the recommendations against your screen.

New to Claude: Prompt 1. Paste into a new chat outside any project.

Reviewing your setup: Prompt 2. Paste into a new chat outside any project.

Your setup will drift as your role, clients and Claude itself change. Review it every three months with Prompt 2.

Read what Claude wrote before you save it; you deliver it, you own it. Then paste it into Settings → Your Claude → Instructions (older layout: Settings → Account → Instructions for Claude). It applies to every conversation, so keep client, entity and one-off details out; those belong in a project (Step 3). If you used Global instructions in Cowork, they now live in this same box.

Memory sits under Settings → Your Claude (older layouts: Settings → Memory, or Settings → Capabilities). "Generate memory from chats" lets Claude save what it learns as topics while you work, and "Search and reference chats" lets it look up your past conversations. Claude decides what to keep, so click Manage next to Your memory and edit or delete anything wrong or stale. "Include sensitive topics in memory" is off by default; leave it off unless you need to change it.

Team or Enterprise: your company's owner turns memory on or off for the whole organization, and owners cannot view your individual memories.

Reviewing your setup? Delete memory topics from finished work and former roles.

Each project has its own memory. Inside a project, Claude uses that project's memory plus your account memory from chats outside projects, and it never draws on memory from your other projects. The key switch is "Use account memory" on each project's page; with it off, the project stays fully separate.

If you are fractional or work across several clients: create one project per client and turn off "Use account memory" on each, so nothing from one client reaches another.

If you are a full-time CFO or head of finance: create one project per recurring workstream, such as the close, board reporting or the budget, and leave account memory on so Claude knows your company everywhere. Turn it off for confidential work, such as a deal or a compensation review.

For a one-off sensitive question, start an incognito chat outside any project (the ghost icon, top right); it is not saved to memory or chat history.

Team or Enterprise: incognito chats still appear in your company's data exports and follow its retention policy.

Reviewing your setup? Check "Use account memory" on every client or confidential project.

On Free, Pro and Max plans, Settings → Privacy has a toggle called "Help improve our AI models". It controls whether your chats and coding sessions can be used to train future models. Turn it off, always; for any client or company work, this is the minimum accepted setup. The change applies to chats from that point on.

Team or Enterprise: the toggle is not on your screen; under Anthropic’s commercial terms, your chats are not used for training unless your company has chosen to share data with Anthropic.

Reviewing your setup? Open the page and confirm the toggle is off.

Under Settings → Capabilities, turn on "Code execution and file creation" (called "Create and edit files" in the new layout). Claude needs it to build Excel files, and skills only run with it on; most workflows here will hand you a .skill or .xlsx file, so this switch matters most. While you are there, check that web search is on, and under Settings → Connectors link only the apps you will actually use, such as email, calendar or drive.

Then install the Claude desktop app and try Cowork. Cowork works on files in a folder on your computer that you connect, and Claude reads and writes only inside folders you have connected. Cowork is now part of the regular conversation in Claude; it is rolling out gradually to Pro and Max plans.

Computer use: keep it off. The desktop app has a setting, Settings → Capabilities → Enable computer use (or Settings → System, under This computer), that lets Claude click and type in the apps on your computer. It is still in beta, and Anthropic's own guidance is to keep it away from financial accounts and other people's personal data. For finance work, connectors are the safer route: you choose exactly which apps Claude can reach, and nothing else. I recommend keeping computer use disabled, at least while it is in beta.

Team or Enterprise: the owner enables code execution and skills under Organization settings → Plugins & skills, and can turn web search off for the whole company.

Reviewing your setup? Look at your connectors and remove the ones you no longer use.

Where this stops. Settings move between releases, so treat the paths above as a guide and your screen as the source. Memory is Claude's judgment about what matters; you deliver the work, you own it, so you keep doing the review in Step 2.

Keep this edition and come back to it every three months; Prompt 2 takes you through the review.

This newsletter is free for everyone, and the best way to help me help you is more readers. Forward it to anyone who might find it useful, especially someone just starting with Claude; connect with me on LinkedIn (linkedin.com/in/anna-tiomina); and like, comment on or repost my posts.

And if you have a workflow in mind that you would like to see here, reply to this email; it is the easiest way to reach me.

Anna Tiomina
CFO & AI Trainer