Microsoft 365 Copilot for Excel: pricing, limits, and what it can’t do
Copilot assumes your data is already in Microsoft's cloud, already clean, and already formatted as a table. Real exports are none of those things.
Copilot in Excel is a good product. It explains formulas, builds charts, updates assumptions from plain English, and it's woven through Word, Outlook, Teams and PowerPoint in a way nothing else is. If your data already lives tidily in SharePoint, it's hard to beat.
But most business data doesn't start there. It starts as an export — from Shopify, QuickBooks, Stripe, a CRM, a client's emailed workbook. This page is about what Copilot requires before it can help, and what that costs you every month.
The short version
| Quiriz | Copilot in Excel | |
|---|---|---|
| Analyse across multiple files | Joins datasets natively | One table at a time — merge with Power Query first |
| Where the file must live | Anywhere — upload, Drive, OneDrive, email, Slack | OneDrive/SharePoint, AutoSave on |
| Messy exports | Auto-mapped and typed on import | Struggles with merged cells, multi-row headers, stray totals |
| Reads the whole table | SQL over every row | Reads "relevant sections" — partial insights |
| Same number every cycle | Governed metrics compiled to SQL | Re-derives from scratch each run |
| Google Sheets and Slack | Yes, both | Excel only |
| Cost per person | Per workspace — askers don't need seats | $32/user/month bundled with Business Premium |
| Word, PowerPoint, Outlook, Teams | Not our job | Excellent — Copilot wins this |
| Respecting existing SharePoint permissions | Our own access model | Inherits Graph permissions — Copilot wins this |
1. Copilot reads across your files. It doesn't compute across them.
This distinction does most of the work on this page. Copilot can reference and summarise other files in your SharePoint or OneDrive, and Copilot in OneDrive compares up to five at once. Useful — but summarising five documents isn't joining five datasets.
In Excel itself, Copilot works with one structured data source at a time, and it has to be a proper Excel table with headers. It reads the active table or your current selection. It can't detect or merge data from other open workbooks.
Microsoft's own answer to multi-file analysis is to merge everything with Power Query first, then format the result as a table. Which is fair advice — it's just the entire job you wanted help with. You do the joining; then the AI gets to answer.
2. "Relevant sections" is doing a lot of work
Copilot handles tables up to two million cells. But it doesn't load the whole spreadsheet into one context — it accesses relevant sections at a time, which Microsoft describes as producing accurate but partial insights on large or multi-sheet workbooks.
For summarising, that's fine. For a total, it isn't. A figure computed from part of a table is wrong, and it will foot perfectly against the part it read. Nothing in the output tells you which part that was.
There's a second version of the same problem. Copilot's documented weak spots — unstructured ranges, merged cells, hidden columns, data scattered across areas — are a fairly precise description of a real export. If a subtotal row is sitting inside your data, it gets counted as data, and every total comes out inflated but internally consistent.
Quiriz reads the whole dataset because it never reads it into a model at all: your question compiles to SQL and runs against every row. On import it also flags columns containing Total, Subtotal or All values so they're excluded from sums, rather than quietly counted twice.
3. Answers in the cell, from data nobody has open
Copilot needs the workbook open, saved in OneDrive, with AutoSave on. Quiriz's Excel add-in works the other way round — it's a window onto datasets you loaded weeks ago:
=QUIRIZ.ASK() returns the answer as a value in the cell — computed against stored datasets, joined across sources, with nothing else open. The sidebar also pushes the active sheet back into Quiriz.Try the test that separates them: open a blank sheet and ask about last quarter's revenue. A sidebar bound to the open workbook has nothing to read.
4. Getting exports in, and keeping them current
Give Copilot credit first: files already sitting in SharePoint need no upload and no refresh at all. If your data is born inside Microsoft 365, that's genuinely frictionless.
For everything else, the checklist runs: download the export → upload it to OneDrive, because Copilot can't see a file on your desktop → clean it and format it as a table with headers → merge with Power Query if it spans files → and do all of it again next month, because there's no dataset and no refresh for external data.
Quiriz: point at a folder and bulk upload, connect Drive or OneDrive for automatic updates, email a file in, or push straight from Excel, Sheets or Slack. Mapping and type detection run automatically — including the merged cells, multi-row headers, stray totals, leading zeros and text-formatted dates that break other tools. Reports re-run themselves when the data refreshes.
5. The part people get backwards
Because Copilot comes from Microsoft, buyers assume it's the economical option. It's the opposite: at $32/user/month it's the most expensive AI option most SMBs will compare — more per person than ChatGPT Team or Claude Team.
Quiriz charges per workspace. An answer costs us a fraction of a cent to serve, so we don't meter questions and don't charge for the colleague who asks one thing a month. You pay for your data, not your people.
Where Copilot genuinely wins
Copilot inherits your Microsoft Graph permissions, so it respects SharePoint and OneDrive access control with no new configuration — if you've already modelled permissions in M365, that's real. It processes inside your Microsoft 365 tenant boundary and doesn't train on your data. And it works across Word, PowerPoint, Outlook and Teams, which we don't attempt at all.
For a well-structured workbook already in SharePoint, analysed by one person, Copilot is a strong tool and you don't need us for it.
You probably shouldn't choose. Use both.
Most people can't remove Copilot anyway — it arrives bundled, chosen at tenant level. Fortunately the split is clean, and it follows one rule: a language model shouldn't be the thing that computes the number.
Quiriz — the number
- Ingest messy multi-source exports
- Join across datasets
- Governed metric definitions
- The same figure every cycle
- Per-step rows for audit
Copilot — everything after it
- Formatting and layout in the workbook
- Charts, conditional formatting, slicers
- PowerPoint decks from the numbers
- Word summaries, Outlook drafts, Teams updates
- Narrative and interpretation
Here's the neat part. Copilot's precondition is data that's already in OneDrive, already clean, already a table with headers. That's precisely what Quiriz produces. Quiriz takes the eleven messy exports, types and joins them, applies your governed definition, and writes a clean structured table into the workbook. Copilot then does what it's genuinely excellent at, on data it could never have assembled itself.
Quiriz is the Power Query step Microsoft tells you to do first — done automatically, and repeatably.
And it works in reverse. When Copilot or a colleague enriches that data in Excel, push it back into Quiriz from the add-in. It becomes a governed dataset again — versioned, scoped by business unit, team and project, refreshable, and askable from Sheets and Slack too. Otherwise an edited workbook is one person's file at one moment, with no lineage and no way for anyone else to query it.
Which one is right for you
Copilot alone is fine if…
- Your data lives in SharePoint and nobody exports anything
- Analysis is per-workbook and the workbook is tidy
- You want AI across Word, Outlook and Teams too
- One person does the analysis
Add Quiriz if…
- Data arrives as exports from several systems
- The same report goes out every cycle and must match
- Several people need answers, most only occasionally
- Some of your work happens in Google Sheets or Slack
- Different people should see different data
Also compare: ChatGPT for data analysis · Claude for Excel · AI for Google Sheets · AI in Excel: the full guide
Try it on your own data. No signup — upload a messy export and ask it something.
Try with your data → Get the Excel add-in See the Slack demo
Copilot features and pricing verified August 2026. Microsoft ships quickly — if something here is out of date, tell us and we'll correct it.
Related: AI Pricing Comparison 2026 — verified per-seat prices for Copilot, ChatGPT, Claude and Gemini and what a 10-person team really pays. AI Hallucination Statistics 2026 — why the same model scores 0.7% or 94% depending on the test.