Microsoft Copilot in Excel Review (2026): Features, Pricing & Verdict
Microsoft Copilot in Excel is the AI data analysis assistant built natively into Excel — the world’s most-used spreadsheet application, deployed across virtually every enterprise and SMB that runs on Microsoft 365. Unlike standalone AI analytics tools that require data to be exported and re-imported, Copilot works directly inside the workbook the user already has open, grounded in the actual data and structure of that specific file, and can edit it in real time: writing and debugging formulas, generating PivotTables from natural language requests, creating charts, summarising dataset insights in plain English, cleaning messy data, and now — following a major update published 25 June 2026 under the “Frontier Finance” positioning — executing repeatable multi-step financial workflows through pre-built and custom Skills, connecting directly to trusted external data sources (LSEG, Ramp, Rogo, Samaya AI, Velixo, Vena) via Copilot connectors, and providing formula-level traceability through Copilot attribution in Show Changes. Agent Mode (rolling out 2026) adds model choice — users can select between Anthropic and OpenAI reasoning models for complex analysis tasks, marking a rare moment where Microsoft offers competitors’ models inside its own product. The platform is grounded in Microsoft Graph — the organisation’s actual document, email, calendar and chat data — which means when Copilot is asked to analyse Q3 revenue it can reference the actual Q3 files the user has access to, rather than a generic template. Forrester projects 116% ROI over three years for enterprise deployments. Pricing is $21/user/month (Copilot Business, annual) or $30/user/month (Copilot Enterprise) as an add-on to an existing Microsoft 365 subscription — Copilot Pro for individuals costs $20/month. The single most important limitation to understand before purchasing: Copilot in editing mode only works with the currently open workbook, and cannot access other files, emails or enterprise data from inside that editing workflow. For cross-workbook reconciliation, multi-file financial modelling, or research-heavy analysis across multiple documents, this single-workbook boundary is a material constraint.
- Best for
- Microsoft 365 organisations needing governed AI analysis inside Excel with formula traceability, financial connectors and native workbook editing
- Headline 2026 update
- Skills (repeatable workflows) · Financial connectors (LSEG, Ramp, Vena) · Copilot attribution in Show Changes · Agent Mode with model choice
- Grounding
- Microsoft Graph — org emails, documents, chats, calendar; open workbook data
- ROI
- Forrester: 116% ROI over 3 years (enterprise)
- Pricing
- Copilot Business $21/user/month (annual) · Copilot Enterprise $30/user/month · Copilot Pro $20/month (individuals)
What Is Microsoft Copilot in Excel?
Microsoft Copilot in Excel is the AI analysis layer embedded directly within Excel — not a plugin, extension or separate tool that requires data export, but a native component of the application that reads and edits the workbook open in front of the user. It converts natural language requests into spreadsheet operations: build me a PivotTable showing revenue by region, write a formula that calculates a 12-month rolling average, explain what this XLOOKUP is doing, flag rows where inventory is below reorder threshold, summarise what changed in this dataset since last month. The 25 June 2026 “Frontier Finance” update marks a significant capability expansion: Skills (pre-built and custom workflow templates that encode multi-step analytical processes), Financial connectors (live market data, fundamentals and research flowing directly into the workbook from LSEG, Ramp, Rogo, Samaya AI, Velixo and Vena), and Copilot attribution in Show Changes (which formula changes Copilot made, why, and what data it drew on — addressing the traceability requirement that finance and compliance teams impose on any AI-generated output).
Core Features
Formula writing, PivotTables, charts — the everyday workflow
For the knowledge worker who spends significant time in Excel but is not a power user, Copilot in Excel eliminates the most time-consuming bottlenecks in daily spreadsheet work. Formula writing — historically the task that most frequently breaks non-analysts’ spreadsheet workflows — becomes a natural language conversation: describe what calculation you need, Copilot writes the formula, inserts it in the correct cell, and can explain what it does in plain English. The same applies to debugging: paste the formula that is not working, describe the expected result, and Copilot identifies the error and corrects it. PivotTable generation follows the same pattern: describe what breakdown or cross-tabulation you need, Copilot builds the PivotTable structure, selects the appropriate fields, and applies the right aggregation method. Chart creation from data ranges, conditional formatting rules, data cleaning (removing duplicates, standardising formats, flagging inconsistencies) and dataset summarisation all operate through the same conversational interface. One independent reviewer reports saving approximately twenty minutes per day on formula-related tasks — modest per session, but material when compounded across a team. A critical operational note: data must be formatted as an Excel Table (not a plain range) to unlock the full Copilot editing experience; unstructured data ranges produce significantly reduced Copilot capability.
Frontier Finance — Skills, connectors and traceability (June 2026)
The June 2026 Frontier Finance update is the most significant expansion of Copilot in Excel’s capability since launch, specifically targeting the finance and FP&A workflow requirements that general-purpose AI analysis tools cannot yet satisfy. Skills are the headline feature: pre-built workflow templates that encode multi-step analytical processes (build a three-statement financial model, create a board package, produce a variance analysis) as reusable instructions that Copilot executes consistently on demand. Custom Skills can be defined by saving a SKILL.md file to OneDrive — Copilot picks it up and applies the process the team has defined. Partner-built Skills from finance and ERP vendors are arriving in Q3 2026. Financial connectors bring live market data, fundamentals and research from LSEG directly into the workbook — eliminating the manual data pull step that consumes significant analyst time in organisations that need current market data for modelling. Connectors from Ramp (corporate spend data), Rogo (earnings call transcripts and financial research), Samaya AI (investment research), Velixo and Vena (ERP/FP&A data) complete the initial connector library. Copilot attribution in Show Changes addresses the single most common objection from finance and compliance reviewers: it logs which changes Copilot made, what formula or data it used, and provides a traceable audit trail from AI output back to the underlying data source. Personalization and workbook rules allow Copilot to apply team-specific formatting conventions, naming standards and formula preferences consistently. Microsoft Finance itself runs Copilot in Excel across FP&A, Accounting, Tax, Compliance and Treasury in real workflows — providing genuine internal pressure-testing of the product before release.
Microsoft Graph grounding and security architecture
Copilot in Excel’s grounding in Microsoft Graph is the feature that most distinguishes it from standalone AI analytics tools in governed enterprise environments. Microsoft Graph is the intelligence layer that indexes all Microsoft 365 content the user has permission to access — emails, Teams chats, SharePoint documents, OneDrive files, calendar events, and meeting records — providing Copilot with organisational context that transforms generic AI analysis into organisation-specific analysis. A query like “compare this quarter’s Excel model against the assumptions in last month’s board presentation” becomes executable rather than requiring the user to manually locate and open both files. The security architecture inherits existing Microsoft 365 permissions, sensitivity labels and data loss prevention policies — meaning Copilot only surfaces and processes data the requesting user already has permission to access, and sensitive files labelled accordingly are handled within the same governance controls as non-AI access. Agent Mode, rolling out progressively in 2026, adds model choice: users can select between Anthropic’s reasoning models and OpenAI’s for complex multi-step analysis tasks — a meaningful acknowledgement that different model architectures excel at different analytical workload types, and a rare instance of Microsoft integrating competitor models directly into a flagship product.
Scored Categories
Pricing
| Plan | Price | Best for |
|---|---|---|
| Copilot Chat (free) | $0 | Included with eligible M365 subscriptions; web-grounded AI chat only — no deep Excel/Word/PowerPoint integration |
| Copilot Pro (individual) | $20/user/month | Individuals; includes Excel, Word, PowerPoint, Outlook integration; priority model access |
| Microsoft 365 Copilot Business | $21/user/month (annual) $25.20 (monthly) |
Organisations up to 300 users; full M365 Copilot experience including Excel, Teams, Outlook; requires eligible M365 base plan |
| Microsoft 365 Copilot Enterprise | $30/user/month | Enterprise; includes Copilot for Finance, Sales, Service at no extra cost in 2026; requires M365 E3/E5 or equivalent base licence |
Strengths
- Native Excel integration — no export/import; edits the open workbook directly
- Microsoft Graph grounding — organisation’s actual data context, not generic AI
- Frontier Finance: Skills, financial connectors (LSEG, Ramp, Vena), Copilot attribution
- Agent Mode with model choice — Anthropic and OpenAI reasoning models
- Formula writing, PivotTables, charts, data cleaning from natural language
- Inherits M365 security: permissions, sensitivity labels, DLP policies
- Forrester: 116% ROI over 3 years (enterprise); SMB ROI 132–353%
- 24/7 integration — works inside the tool the team already uses daily
Weaknesses
- Single-workbook boundary — editing mode cannot access other files simultaneously
- Requires existing M365 base subscription — significant total cost if starting from scratch
- Complex reasoning (model audit, financial assumption review) — Claude for Excel ranked higher by independent benchmarks
- App Skills removed Feb 2026 — Python-style analysis now in separate “Analyst” tool
- Precision caveat — Microsoft explicitly warns against use in finance/legal without human review
- Data must be formatted as Excel Table — unstructured ranges reduce capability significantly
- No free trial — Copilot Chat (free) does not include Excel integration
Verdict: 8.5 / 10 — The Definitive AI for Governed Microsoft 365 Excel Workflows
Microsoft Copilot in Excel’s 8.5 reflects a product that is genuinely the best AI data analysis tool for one specific, enormous use case: organisations already on Microsoft 365 that need governed, security-compliant, organisation-data-grounded AI analysis inside the spreadsheet application their teams already use daily. The Frontier Finance June 2026 update — Skills, financial connectors, Copilot attribution, Agent Mode with model choice — is one of the most significant AI-in-spreadsheet capability expansions of 2026, targeting the finance and FP&A workflows where previous versions fell short of professional-grade requirements. For that M365-native use case, it is unmatched. The honest caveats are the single-workbook editing boundary (material for multi-file financial modelling and reconciliation workflows), the Microsoft warning on precision in sensitive finance decisions, and independent benchmark evidence that Claude for Excel outperforms Copilot for complex reasoning-heavy workbook analysis. For organisations already on M365, Copilot in Excel is the right starting point for data analysis AI; for pure analytical depth on complex models, augment with reasoning-first tools where needed.
Frequently Asked Questions
What are Skills in Copilot in Excel and how do they work?
Skills (generally available from June 2026) are pre-built or custom workflow templates that encode multi-step analytical processes as reusable instructions Copilot executes on demand. A pre-built skill for “three-statement financial model” encodes the sequence of steps — income statement structure, balance sheet linking, cash flow derivation, formula relationships — that Copilot follows to build the model from the user’s input data, rather than requiring the user to prompt each step individually. Custom Skills are defined by saving a SKILL.md file to OneDrive: the file describes the workflow steps, naming conventions, formula preferences and output structure the team uses, and Copilot picks it up and applies it consistently every time that skill is invoked. Partner-built Skills from ERP and finance vendors (LSEG, Ramp, Rogo, Samaya AI, Velixo, Vena) are arriving in Q3 2026, integrating those vendors’ data and workflows directly into the Excel skill ecosystem. The practical value for finance teams is standardisation at scale: instead of each analyst building a three-statement model from memory (with inevitable variation in structure, formula conventions and output format), the Skill enforces a single consistent methodology across the team, with Copilot handling the execution and the analyst focusing on the assumptions and judgement calls.
What is the single-workbook limitation and when does it matter?
Copilot in Excel’s editing mode operates on the currently open workbook only — it cannot simultaneously read, reference or edit data from other Excel files, Outlook emails, SharePoint documents or Teams messages during an editing session. For single-workbook analysis tasks (analysing a dataset, building a model from data already in the file, summarising results, formatting outputs), this boundary is irrelevant — Copilot has everything it needs. The limitation becomes material in three specific scenarios: multi-file financial modelling (where assumptions in one workbook need to inform formulas in another), workbook reconciliation (comparing two files to identify discrepancies), and research-informed analysis (where the analyst needs to reference an Outlook email or SharePoint document alongside the spreadsheet). For these workflows, the user must manually open each file, copy relevant data into the working workbook, and then use Copilot — adding manual steps that partially offset the productivity gain. Microsoft’s Copilot Chat (accessible separately from the Excel editing interface) can search across the full Microsoft Graph index during conversation, which provides cross-document context at the chat level. However, the Excel editing workflow — where Copilot actually inserts formulas and modifies the workbook — remains bounded to the single open file as of mid-2026.
How does Copilot in Excel compare to Claude for Excel?
Independent benchmarking as of mid-2026 positions Microsoft Copilot in Excel as the superior choice for governed Microsoft 365 environments, and Claude for Excel (via the Anthropic-Excel workflow) as the superior choice for complex reasoning-heavy work — workbook audits, financial assumption reviews, formula logic explanation at depth, data cleaning of messy inherited files, and multi-step analysis requiring sequential reasoning. The distinction is architectural: Copilot’s advantage is native workbook editing, Microsoft Graph grounding, M365 security inheritance, and the June 2026 Frontier Finance features (Skills, connectors, traceability). Claude’s advantage is reasoning depth — explaining complex formula logic, identifying subtle model errors, auditing financial assumptions for coherence, and handling the kind of complex reasoning tasks that require the model to maintain a detailed understanding of workbook structure across many sequential analytical steps. In practice, many finance teams will find value in both: Copilot for daily formula writing, PivotTable generation, data summarisation and Skills-based workflow automation within the M365 governance framework; Claude for periodic deep model audits, assumption reviews and complex analytical challenges where reasoning depth is the primary requirement.