Many finance teams already have two things sitting on the same desktop: a TM1 model they pull reports from through Planning Analytics for Microsoft Excel, and an AI assistant they are not quite sure what to do with. Connecting the two is more useful than it first appears. PAFE handles the part AI is bad at, getting the right numbers at the right level and from the right version of the model. Claude handles the part the spreadsheet is bad at, explaining what the numbers mean and turning that explanation into something presentable.
The common “AI-finance” workflow starts with an export. Someone runs a report, saves it as a file, cleans it up, and hands it to an AI tool. That workflow inherits every problem the export had. If the extract was pulled at the wrong consolidation, filtered to the wrong version, or assembled by hand with lookups against a second file, the AI will analyze it with complete confidence and produce a well-written wrong answer.
Starting from a PAFE report removes most of that risk. An Exploration View, Quick Report, Dynamic Report, or Custom Report is not a copy of the data, it is a live window into the cube. The hierarchy is the model's hierarchy. The consolidations are the model's consolidations. The scenario, version, and currency are whatever the report's context dimensions say they are, and they are visible on the sheet. When Claude reads that sheet, it is reading something the model produced, not something an analyst assembled.
Claude for Excel is an add-in installed from the Microsoft Marketplace, available on Pro, Max, Team, and Enterprise plans. It runs as a sidebar inside Excel and sits alongside PAFE without conflict, so nothing about your existing report library must change.
That is the whole setup. There is no connector to configure and no model to expose, which is a large part of why this is worth trying before committing to anything heavier. You will simply get a Claude button in Excel that opens a chat window you can use to ask questions about your reports:
Claude for Excel runs as a sidebar add-in. It reads the open workbook cells, formulas, and tab structure and answers with citations back to specific cells, so every claim it makes can be clicked and checked.
The workflow is unremarkable, which is the point. Refresh your PAFE report, open the sidebar, ask a question in plain English.
"Which accounts are furthest from budget this quarter?":
"Build a waterfall showing the bridge from prior year to current":
"Summarize this P&L in five bullets for the board pack":
The chart requests are worth calling out separately, this is where the time savings are most obvious. Building a revenue bridge waterfall by hand in Excel is a tedious exercise of hidden helper columns and hand-picked colors. Asking for one takes a sentence. The same is true of combo charts, cohort heatmaps, and KPI scorecards assembled from several tabs of a multi-sheet report.
Claude can also apply Excel-native operations directly: sorting and filtering, editing pivot tables and charts, conditional formatting, data validation, and print setup. So "make this presentable" is a reasonable instruction, not just "explain this to me".
The quality of the answer Claude gives depends heavily on how the report was assembled, and the habits that make a PAFE report readable to a colleague are the same ones that make it readable to Claude.
Keep the context dimensions visible on the sheet rather than buried in the report definition. A grid of numbers with no indication of scenario, version, or currency is ambiguous to anyone, and an AI assistant will resolve that ambiguity by guessing. Use member aliases that carry meaning. Claude reasons about "Marketing — EMEA" far better than it reasons about a cost center code, and if your model stores both, put the descriptive alias on the report. Give each report its own tab and title the tab accordingly. Where a report is a means to an end rather than the deliverable, say so in the prompt, pointing at the sheet by name and stating what it contains costs one sentence and removes most of the interpretation risk.
Refresh before you ask, not after. Claude reads the values sitting in the cells the moment you send the message, so a stale report produces a stale analysis.
It is also worth knowing what Claude will not touch. It does not support macros or VBA, and it does not have data tables. It cannot trigger a PAFE refresh, change your TM1 connection, or write back to the cube. The add-in is confined to the workbook that is open in front of it.
Excel burns through usage limits much faster than chat. Each prompt in Excel is expensive, reading the workbook, reasoning across tabs, and making a sequence of tool calls to apply changes. Anthropic's guide of about 45 messages per five-hour window on Pro assumes short text chats; spreadsheet work burns through that far faster, and Pro users can run out mid-session. Heavy users realistically need Max or Team headroom, which is a budget conversation worth having before rollout rather than after.
Responses are not instant. Simple questions can be quick, but questions that require deeper analysis of the report or requests such as updating formatting or creating a new sheet can take minutes.
Everything still needs checking. Anthropic is explicit that the add-in is not recommended for final client deliverables without human review, or for audit-critical calculations without verification. Budget review time rather than assuming it away.
We would frame Claude for Excel on top of PAFE as the accessible floor rather than the ceiling. It is a meaningful improvement over exporting to CSV and hoping for the best, it requires no implementation project, and it puts capable analysis in the hands of anyone already living in Excel. It is not a substitute for AI that can reach into the model, but for the everyday work of reading, explaining, and presenting your numbers, it is hard to beat, and it is ready the moment you install it.
If you would like to talk through how this fits your reporting cycle, contact us at solutions@acgi.com.