Skip to content

A Landscape of AI Data Analysis Tools

The three broad categories and what each is for

As Code Interpreters vs. Plain Chat covers, one question matters most for any tool in this bundle. Does it run a real computation against your data, or predict a plausible-sounding answer? That question extends past chat tools. Two other categories deserve the same test: AI built into business intelligence dashboards, and AI built into spreadsheets themselves.

Chat-based code interpreters are the category the rest of this track has covered directly. Claude's code execution tool, OpenAI's Advanced Data Analysis, and Gemini's code execution mode all fit here. Each writes and runs real code in a sandboxed environment. The earlier pages in this track show the vendors disagree on what that guarantees about file size. The BlueFin benchmark shows real code execution does not guarantee a correct answer either (Kundurthy et al., 2026).

BI-embedded AI features live inside a business intelligence tool already used for reporting. Microsoft's Power BI Copilot answers natural-language questions by generating and running real DAX queries against the live semantic model. It does not read a summary of the data (Microsoft, 2026). That guarantee comes with a real precondition. The underlying model has to be genuinely prepared for AI, with descriptive field names and sound structure. Without that preparation, Microsoft's own documentation states the tool can produce generic or misleading output (Microsoft, 2026). A separate, narrower toggle also applies. Q&A is enabled by default for most data sources, though some, like DirectQuery or Direct Lake models, need it turned on manually (Microsoft, 2026).

Spreadsheet AI features are built directly into the spreadsheet application itself. Copilot in Excel writes real, functioning formulas into the workbook, the same kind of formula a person could write by hand. It does not describe an answer in prose (Microsoft, 2026). Its edits save automatically to the live file. Every collaborator with access can see the change (Microsoft, 2026).

A quick decision guide by task type

A chat-based code interpreter fits a one-off analysis of a file already in hand. It needs no setup beyond uploading the file. A BI-embedded feature fits a recurring question against a dataset an organization already tracks in a dashboard. That only holds once the underlying model is prepared for it. A spreadsheet AI feature fits building or auditing a calculation inside a workbook people will keep editing by hand. Its output stays as ordinary, inspectable formulas, not a one-time answer.

None of these three categories replace the others. Each assumes a different starting point: a standalone file, a maintained data model, or a live workbook. Picking the wrong one for the job often means falling back to a prose answer. The tool could have run a real computation instead, had it been set up that way.

What to check before adopting a new one

Confirm three things before adopting any tool in these categories for a recurring task. First, confirm it is actually computing against the data, not summarizing it. Use the demonstrable test from Code Interpreters vs. Plain Chat: a known total should match exactly, not approximately. Second, check what preparation the tool actually requires. Power BI Copilot's own documentation states plainly that a poorly structured model produces worse answers, not just slower ones (Microsoft, 2026). Third, read what the vendor itself says about verifying an answer before trusting it. Power BI Copilot's own guidance says to review its generated DAX query before applying it (Microsoft, 2026). It also says to expand its "How Copilot arrived at this" explanation, to check which fields it actually used (Microsoft, 2026). Copilot in Excel carries the same instruction, to review, edit, and verify anything it creates before relying on it (Microsoft, 2026). A vendor telling a user to check its own output is not a formality. It is the same caution Why Dumping a File Rarely Works opened this track with.

Sources

  • Kundurthy, Na, Moraine, Mohta, Winter, Fang, Ling, Strubell, Kirshner — BlueFin: Benchmarking LLM Agents on Financial Spreadsheets, 2026. a sandboxed code execution tool exposes the workbook as an in-memory openpyxl object with restricted built-ins... frontier LLMs demonstrate poor performance on the challenging benchmark, with the strongest LLMs achieving less than 50% average scores across tasks... frontier models all struggle to clear the 50% threshold, with GPT-5.5 and Opus 4.7 at the forefront. View source · verified 2026-07-06 · primary
  • Microsoft — Ask Copilot questions about your data - Power BI, 2026. [Copilot for Power BI] answers natural-language questions about your data by querying your semantic model and returning the results as a visual. Copilot can also generate DAX queries to answer questions that require ad hoc calculations, such as creating new measures that are not contained in the model. View source · verified 2026-07-06 · primary
  • Microsoft — Copilot for Power BI overview, 2026. You need to prepare data to work with Copilot. Model owners need to invest in prepping their data for AI to ensure Copilot understands the unique business context, prioritizes the right information, and delivers responses that are consistent, reliable, and aligned with your goals. Without this prep, Copilot can struggle to interpret data correctly - leading to generic, inaccurate, or even misleading outputs. View source · verified 2026-07-06 · primary
  • Microsoft — Ask Copilot questions about your data - Power BI, 2026. You need to enable Q&A for your model because Copilot uses the same underlying engine to build queries from your inputs. For most data sources, Q&A is enabled by default. For some sources, such as DirectQuery or Direct Lake models, you might need to enable Q&A manually. View source · verified 2026-07-06 · primary
  • Microsoft — Generate formula rows and columns with Copilot in Excel, 2026. Copilot in Excel can create new columns or rows that calculate values from your existing data, and explain how each formula works. It recommends a formula such as XLOOKUP and adds a new column for the results. View source · verified 2026-07-06 · primary
  • Microsoft — Frequently asked questions about Copilot in Excel, 2026. Copilot updates your workbook using Excel's built-in features. When Copilot edits your workbook, changes are saved automatically. Anyone with access to the file can see those changes, including people in a coauthoring session. View source · verified 2026-07-06 · primary
  • Microsoft — Privacy, security, and responsible use for Copilot in Power BI, 2026. Review Copilot-generated DAX queries before you apply them to your model. View source · verified 2026-07-06 · primary
  • Microsoft — Ask Copilot questions about your data - Power BI, 2026. Expand the How Copilot arrived at this section to learn more about how Copilot understood your question and to verify that the correct fields, measures, or filters were selected from your model. View source · verified 2026-07-06 · primary
  • Microsoft — Frequently asked questions about Copilot in Excel, 2026. Copilot uses AI to generate suggestions and can sometimes make mistakes, misinterpret information, or produce inaccurate results. Review, edit, and verify anything Copilot creates before you rely on it. View source · verified 2026-07-06 · primary