Most of what a finance team produces gets used once. You pull the data, build the view, send it, and the file is never seen again.

This is costly for two reasons:

1. The work is tedious to regenerate (ask any IB Analyst). You format the table, add the variance column, paste it into a deck, fix whatever shifted in the paste, and write some strung-together commentary at 9 pm because the numbers only went live at 8.

2. The second is that some of those requests keep coming back. Maybe it is the same question every third deal, the same monthly cut that three people want re-sliced, or the same calculation that four account executives run by hand and somehow get four different answers. Those are the kinds of requests that are better handled by a tool, but building one has traditionally meant asking someone in engineering to carve out two weeks (they usually ghost me after this).

Claude Artifacts covers both. I’ll run you through the mechanics first, then walk through three finance deliverables built end to end.

WTH is an Artifact

An Artifact is a functional app Claude builds in a panel beside the conversation. You describe what you want, Claude writes the code and renders the interface, and the result can be published for anyone to access. The code is available if you want to inspect it, but I doubt you will :)

Dashboards are the most common use case, but you can produce calculators, internal tools, websites, interactive explainers (like employee training guides), and games. I’ve even generated an app that lets you rank and rate cookies.

🚨 There are four capabilities you should know before you build one. An Artifact can turn uploaded data into an interactive app. It can also call Claude from inside that app based on whatever a user enters; it can save information, and it can connect to outside applications so it can take actions such as posting something to Slack (ex: your Levain Baker Double Choco Chunk review)

What you need before you start

• A Claude account. Artifacts work on the free plan and on every paid plan. Every edit counts against your daily and weekly usage, so a free account can run out mid-project if you iterate too aggressively.

• One sentence describing what the thing does. You do not need a complete feature list. “A dashboard for a monthly revenue review” is enough to start, and it’s arguably easier to steer from there than from a long initial spec sheet. You may even find that Claude includes a feature or view you missed.

Setting Up and Using Claude Artifacts

Setup

1. Click Artifacts in the left sidebar. This is where everything you build lives, split between what you created and what others shared with you.

2. Click New Artifact in the top right. You'll be brought to a new chat session where Claude will present several categories: apps and websites, productivity tools, documents and templates, or something else entirely

3. Click through the series of questions to scope out your Artifact.

4. Or skip all of the above and ask for it inside a conversation you already have open. For instance, if you’re in a session currently comparing budget v actual files, you can ask Claude to create a dashboard to summarize the variances. Both routes end at the same Artifact.

5. Publish the artifact when Claude finishes, then click the expand icon to review it at full width.
   
   

Key settings to configure

1. Data source. When you ask Claude to build an Artifact, Claude will often ask what data source to use before it starts. "Placeholder data" is one of the answers you can give at that point, and lets Claude invent a realistic set so you can shape the layout before any company data is involved. I suggest using it while you're still deciding what the app should look like. Alternatively, you can use actual data (via Google Sheet, imported file, connected database, etc.)

2. Saved data. When you ask Claude to add memory to an artifact, it isn't flipping a single switch to "private" or "shared" — it's deciding, field by field, where each piece of data lives. By default, everything goes into one shared pool that every single person who opens the link can read and write, which is exactly why a team tracker works: someone checks off a task, and everyone else sees it update. If you want something to stay private to just you, you have to tell Claude that specifically, because private data gets stored in a separate spot tied only to your account, and Claude won't set that up unless you ask. So if you're building something with sensitive numbers in it, like a sales pipeline, and you never mention privacy, that data defaults to the shared pool and becomes visible to anyone with the link, not just you.

3. Share visibility. Sharing an artifact means setting two things: who can open the link and which version they see. The link defaults to your organization, so you have to turn on public access separately if you want anyone online to open it, and it can either follow "latest" automatically or stay frozen on one version even as you keep editing. If it's frozen and you don't re-share, your reader just keeps seeing the old build with no sign anything changed.

4. Pinning an artifact. You can pin an Artifact from the three dots next to it in the Artifacts list, which moves it into a pinned section at the top so you can find it fast later. What it gets you back to is just the finished output itself, not the chat where you built it. Any edit you want to make still has to happen back in that original conversation — pinning doesn't give you a shortcut to keep working on it, only a shortcut to view it again (silly, I know)

5. Code View. Code View is something you can open right on the artifact itself, showing you the actual underlying code instead of the rendered page. The reason to open it is validation: if you're assuming some calculation is happening behind the scenes, checking the code is how you confirm it's actually running using the correct logic.
   
   

How to take an Artifact further

Add Claude inside the app

Suppose you want to turn a static deal calculator into something that also provides judgment on the inputs. Start a new Artifact, choose “something else,” and describe the tool:

Create an internal tool for our account executives. They enter deal size, contract term, and requested discount. Use Claude to return a short written rationale for the discount plus two questions the deal desk should ask before approving. Keep the response under 150 words.

A different set of inputs produces a different answer. The AE does not need to write a prompt because you already decided what Claude should evaluate and how it should respond.

This is also what makes the capability useful beyond your own workflow. You are not sharing a Claude conversation with someone else. You are giving them a specific way to use the model, with the framing, constraints, and structure already built into the tool.

If you plan to share the Artifact with a team, you should also open the shared link from another account before rolling it out broadly so you understand what account access viewers need. For this, just use a personal account or some other dummy account to test.

Let the tool save information

Once that works, you can ask it to retain what users create:

Add the ability to save a rationale, and a section at the bottom listing every saved one with the deal name and the date.

Claude asks who the saved data should belong to: just you, or everyone who opens the page. This determines whether you are building a private scratchpad or a shared record the team can use.

This is the capability that allows the same basic Artifact to function more like a tracker, pipeline log, or diligence list.

Connect it to another application

Artifacts can also reach outside Claude through connectors running on MCP, or Model Context Protocol, which is a standard way for AI tools to interact with other apps (like Slack or Notion) and take actions in them.

For the same deal desk tool, you could:

Add a button that posts a saved rationale to our deal-desk Slack channel.

For recurring data: use a Live Artifact

There is one other capability that sits somewhat separately from the progression above.

In the Claude desktop app, a Live Artifact can pull current information from connected applications and local files on your machine, with a reload button that forces a refresh.

That makes more sense for something you expect to revisit every week or month. For a one-time analysis, the uploaded snapshot is usually enough.

Compounding time savings

One build could save you an hour. Making the app reusable is what changes the economics over time.

A month later, you can feed in the next export and get the same shape of output from the same app. That makes it worth spending a second pass hardening the Artifact you expect to run repeatedly. There is much less reason to do that for something you built to answer a single question.

Common errors, and how to fix them

The dashboard looks slick, so nobody checked it. Clean formatting can make a chart feel more trustworthy than it necessarily is. Make sure you tie the headline figures back to the source export before sharing it, and consider building the numbers-validation view into the app so that check is part of every run.

The Excel file confuses it. Multi-sheet workbooks and merged header cells can create errors that are not immediately obvious. Export one flat sheet with a single header row instead.

You change a forecast input and only one cell moves. The build did not carry the change into later periods. Tell Claude exactly how the value should propagate and rebuild that behavior.

Your colleague is looking at an old version. Sharing points to the version you selected when you copied the link. After another round of changes, share the current version again.

The team can’t open what you built. Publishing to the web and sharing internally are separate actions, and internal team sharing depends on having a Team or Enterprise plan. Confirm what your account supports before promising everybody a link.

Saved entries are visible to people who shouldn’t see them. That behavior comes from how you answered Claude when it asked who should own the saved data. Tell Claude which access model you want and have it changed.

The dashboard is stale. A standard Artifact does not refresh automatically. Re-upload the data whenever you need a current version, or rebuild the tool as a Live Artifact in the desktop app.