Claude vs ChatGPT-Excel vs Copilot vs ServiceNow: Q3 Matrix

There are now four credible AI options layered on top of Microsoft 365 in production this week. They overlap in confusing ways, none of them is a clean substitute for any other, and Microsoft Build on May 19 is going to add a fifth. If you’re a CIO or VP of IT in a Microsoft shop trying to make a Q3 procurement decision, the matrix has gotten harder, not easier.

The four options shipping right now:

Claude × Microsoft 365 — Anthropic’s Excel, PowerPoint, and Word add-ins, plus Outlook in beta. Strong on cross-app workflows. Expanding rapidly.
ChatGPT for Excel and Google Sheets — OpenAI shipped these out of beta to GA on May 5, 2026. Powered by GPT-5.5, free preview through June 2 for Business / Enterprise / Edu / K-12 plans. Workbook reasoning, formula tracing, error explanation.
Microsoft Copilot + Agent 365 — Microsoft’s first-party stack. Centralized agent management, Copilot Cowork, ecosystem partners (Genspark, Zensai, Egnyte, Zendesk). The default Microsoft answer.
ServiceNow Autonomous Workforce embedded specialists — Announced at Knowledge 2026 on May 5, embedded in Outlook / Word / PowerPoint via Microsoft AI Control Tower governance. CRM / IT / HR / Finance specialists. L1 IT Service Desk available now; security/risk September 2026 GA.

This post is the Q3 procurement decision matrix. Four axes, four-vendor pairings, the one thing not to do this week, and the v1.1 refresh banner for May 20 after Microsoft Build resets the matrix.

What’s actually shipping in production this week

Vendor marketing pages all sound similar. The shipping behavior is what matters.

Claude × Microsoft 365

Anthropic’s add-ins are live for Excel, PowerPoint, and Word. Outlook is in beta. The strongest current use case is cross-app workflows — pull a paragraph from a Word document, summarize a PowerPoint, draft an email response, all in one Claude conversation that retains context across the apps. Recent expansions: Moody’s MCP integration for analyst workflows, a handful of Wall Street firm partnerships, and ongoing Excel formula assistance that’s improving fast through May.

The macro context that’s worth knowing for procurement: Anthropic’s share of enterprise LLM revenue rose to roughly 40% by 2025, with OpenAI down to about 27% and Google up to ~21% per analyst roundups. Large enterprises — especially in financial services and complex engineering — are increasingly standardizing on Claude for coding and analytical workflows even when Microsoft 365 remains the primary workspace. Beyond the well-known Moody’s MCP integration, Anthropic has been shipping Cowork and Claude Code plugins plus specialized connectors for finance data providers, and analyst commentary points to AMD and Synopsys integrating Claude into chip-design flows as evidence the same pattern (Claude as second-engine inside Microsoft 365) is replicating in regulated/high-stakes engineering.

The shop pattern that fits cleanest: Anthropic-preferred organizations that have already chosen Claude as the primary model for legal or risk reasons. The cross-app continuity is genuinely useful and harder to replicate with the OpenAI add-ins, which are spreadsheet-only.

ChatGPT for Excel + Google Sheets

OpenAI’s GA shipped May 5 with a free preview through June 2 for Business, Enterprise, Edu, and K-12 plans. The shipping product is a sidebar inside Excel and Google Sheets that does workbook reasoning — understands which formulas connect to which, explains output changes when you edit a cell, traces and fixes errors. The cross-platform piece (it works in Excel and Google Sheets, with the same conversation context) is the differentiator from Anthropic’s Excel-only add-in.

Pricing tiers: Free and Go plans get limited usage; Plus and Pro use the same agentic usage limits as Codex. For Enterprise workspaces, the feature is disabled by default — admins enable it through role-based access controls. After the June 2 cliff, organization-level credits start metering each interaction; heavier tasks (large tables, multi-sheet models) consume more credits per call. Practical guides aimed at consulting and banking teams report internal benchmarks where the spreadsheet assistant boosted three-statement modeling task accuracy from ~44% to ~87% using GPT-5.x-class models — which is why investment banks, PE shops, and FP&A teams are running aggressive pilots in the May-June window.

Honest about the limits: early versions don’t fully support VBA or Power Query, and there are context-window limits on very large workbooks. Enterprise architects are using the free preview window to gather telemetry — usage by team and workbook type — to set post-cliff guardrails like quotas per department and routing routine cleaning tasks to cheaper internal models.

The shop pattern that fits: Excel-heavy finance and analyst teams, especially shops that also live in Google Sheets for some of the workflow. The free preview through June 2 is the “pilot it now, decide on commitment in mid-June” window.

Microsoft Copilot + Agent 365

The first-party Microsoft stack. Copilot has been GA for over a year; Agent 365 went GA on May 1, 2026 at $15/user/month (per-user subscription, no separate consumption-based meter for the agent layer yet). It’s also bundled into the new Microsoft 365 E7 “Frontier Worker Suite” at $99/user/month — that bundle includes E5 (~$60 from July 2026), Microsoft 365 Copilot ($30), Entra Suite ($12), and Agent 365 ($15); separately those sum to about $117, so the bundle discount is roughly 15%. The May 5 update added Copilot Cowork — multi-agent orchestration for cross-departmental workflows — plus ecosystem partner agents (Genspark, Zensai, Egnyte, Zendesk).

What buyers are actually doing in production (the patterns from licensing analysts and Microsoft MVP commentary):

“Copilot-first” — keep E3/E5; buy Copilot + Agent 365 à la carte for 10-20% of users (front-office, finance, legal). Most popular pattern.
“E7 anchor” — adopt E7 for 5-15% of seats only (trading desks, research analysts, deal teams, clinical leadership) where you genuinely need high-end Copilot + Agent 365 + Entra Suite. Hold the rest of the org at E3/E5 + targeted Copilot.
“Wait and cap” — renew existing M365 SKUs early to dodge the July 1, 2026 price increases (11-20%), and formally cap Agent/Copilot penetration at 2-5% “power users” until clearer ROI data emerges.

Honest about the limits: the Microsoft MVP community has been blunt that Agent 365 is priced purely per-user rather than by activity or agent count, which makes it feel expensive for partial rollouts where most workers don’t yet have advanced agents. Analyst commentary frames Agent 365’s initial value as “bringing shadow agents under control” — unified logging, approval workflows, and policy enforcement — rather than net-new capability. It’s the governance/identity control plane, not yet the autonomous workforce platform.

The shop pattern that fits: any Microsoft 365 E7 customer who’s already paying for Copilot. The audit trail is the strongest in the market because it’s built into the M365 admin layer, not bolted on. If your compliance team needs centralized AI agent governance and you’re already a Microsoft shop, this is the path of least resistance.

ServiceNow Autonomous Workforce embedded specialists

The newest entrant. ServiceNow announced at Knowledge 2026 that their AI specialists — domain-specific agents for CRM, IT, HR, Finance — will embed in Outlook, Word, and PowerPoint through the Microsoft Agent 365 Marketplace, governed by Microsoft AI Control Tower. The specialists do entire business processes start-to-finish: an HR specialist handles offboarding workflows; an IT specialist handles L1 ticket triage and resolution; a Finance specialist handles invoice processing.

L1 IT Service Desk is the first specialist available. Security/risk specialists are GA September 2026. The integration is currently in preview; Microsoft 365 Marketplace availability is “later this year.”

The shop pattern that fits: enterprises with deep ServiceNow workflows already in production where the missing layer is making those workflows act inside Microsoft 365 surfaces rather than only inside ServiceNow’s UI. Audit-driven, compliance-driven, and CRM/IT-heavy organizations.

The 4-axis decision frame

Four axes. Each one points to a different vendor.

Axis 1 — Workflow primacy

What’s the dominant work surface in your shop?

Excel-heavy (finance teams, analyst shops, FP&A, data teams): ChatGPT for Excel + Sheets is the strongest fit. The workbook-reasoning behavior is genuinely better than what’s available in Copilot’s Excel features today.
Cross-app (consultants, executive support, anyone moving between Word, PPT, and Outlook): Claude × M365 or Copilot + Agent 365. The cross-app continuity matters more than per-app depth.
CRM + IT + HR cross-departmental: ServiceNow specialists. The reason: those specialists already understand the underlying business processes; you’re not retraining them on every workflow.
General productivity: Copilot + Agent 365 is the default. It does everything well enough.

Axis 2 — Existing license stack

If you’re already paying for Microsoft 365 E7 + Agent 365, the marginal cost of layering a second AI vendor on top is the new spend, not the total spend. Default to Copilot first; layer Claude or ChatGPT-for-Excel as task-specific tools where Copilot’s per-app behavior is weaker.

If you don’t have Agent 365 yet and you’re evaluating from scratch, the question is reversed: which AI fits the work, and does Microsoft’s E7 + Agent 365 bundle still make sense at that price point?

Axis 3 — Anthropic vs OpenAI vs neither (the model preference question)

Many enterprise shops have already standardized on one model provider for legal, risk, or contractual reasons. If your shop has signed a multi-year Anthropic enterprise agreement, the matrix collapses: Claude × M365 is your default and the question is what to layer alongside it. Same for OpenAI-first shops.

For best-of-breed shops accepting governance complexity, the answer is multi-vendor — Copilot + Agent 365 + Claude × M365 + ChatGPT for Excel — with the audit-trail consolidation handled at the Agent 365 layer.

Axis 4 — Governance + audit-trail posture

This is the deciding axis for regulated industries. Three patterns:

ServiceNow AI Control Tower + Microsoft Agent 365 is the only path that consolidates audit trails across all AI agents in the Microsoft stack. If compliance is your driver — financial services, healthcare, regulated tech — this is a near-default.
Microsoft Agent 365 alone consolidates audit across Microsoft-side AI but not across third-party agents.
Best-of-breed without ServiceNow’s governance layer means you’re tracking audit logs in three places. Acceptable for low-regulated shops; risky for SOX/HIPAA/SOC2.

The 4 “this is a 2-vendor pairing” patterns

Most Microsoft shops will end up with 2-3 vendors. The pairings that work:

Pattern A — Anthropic-preferred shops: Copilot + Agent 365 (governance) + Claude × M365 (cross-app primary). Use Copilot for the in-document quick assist, Claude for the longer cross-app workflows, ChatGPT-for-Excel only if your Excel-heavy team has a specific workflow Claude doesn’t cover well.

Pattern B — OpenAI-preferred shops: Copilot + Agent 365 (governance) + ChatGPT for Excel/Sheets (Excel-heavy work). Skip Claude × M365 unless a specific cross-app workflow requires Anthropic’s model behavior.

Pattern C — Best-of-breed shops: Copilot + Agent 365 + Claude × M365 + ChatGPT for Excel. Highest capability, highest governance complexity, highest spend. Justifiable when your work spans cross-app workflows AND deep Excel reasoning AND general productivity.

Pattern D — Compliance-first shops: Copilot + Agent 365 + ServiceNow specialists. Skip both Claude and ChatGPT-for-Excel as the primary AI; use ServiceNow’s specialists for the workflows that matter most (HR, IT, Finance) and Copilot as the in-document fallback.

The 1 “wait for Microsoft Build” gate

Microsoft Build runs May 19-22. Microsoft will almost certainly announce Copilot + Agent 365 v2 features that reset the matrix — likely candidates: GitHub-Advanced-Security AI integration, Microsoft Defender-for-Cloud AI agents, and additional ecosystem partner agents. The Microsoft Agent 365 v2 announcement is the single largest change to the Microsoft-stack AI procurement landscape that will happen this quarter.

If your shop has no immediate Q3 deadline: hold the contract refresh until May 20 to capture the announcement layer. The two-week wait costs you nothing and saves you from a v1.0 commitment that’s obsolete inside a fortnight.

If your shop has a hard Q3 deadline (procurement budget cutoff, fiscal year alignment, regulatory deadline): sign month-to-month for the next 30 days, plan to renegotiate after Microsoft Build, and structure the agreement so you can swap vendors without a contract penalty.

What this means for you

If you’re a Microsoft 365 E7 + Agent 365 shop: Copilot is your default. Pilot ChatGPT for Excel through the free June 2 preview to see if your finance team gets value. Add Claude × M365 only if you have a specific cross-app workflow that Copilot is missing. Don’t change the foundation; layer on top.

If you’re a financial-services or healthcare shop: ServiceNow specialists + Microsoft Agent 365 governance. The audit-trail consolidation is the right answer. Skip best-of-breed for at least 6 months until the governance market matures further.

If you’re an Anthropic-preferred enterprise: Claude × M365 + Copilot + Agent 365. Use Claude for the strategic cross-app work, Copilot for the routine quick-assist, Agent 365 for the audit layer. ChatGPT for Excel only if a specific finance team requests it.

If you’re an Excel-heavy analyst or finance team: Pilot ChatGPT for Excel through June 2 — the free preview is the cheapest evaluation window of the year. If your team is already on Claude × M365 for non-spreadsheet work, run them in parallel and compare. The workbook reasoning behaviors are different and one will likely win for your specific workflow.

If you’re a SMB without M365 E7: the math may not justify Agent 365 yet. Run the free ChatGPT-for-Excel preview, see if it covers your spreadsheet needs, and revisit the cross-app question in Q4 once the Microsoft Build announcements have been digested.

What it can’t do

It can’t replace your data-governance review. Layering 4 AI vendors on top of M365 multiplies your data-handling surface. Don’t make the procurement decision before your data team has reviewed each vendor’s data-handling agreement, retention policy, and training-data carve-outs.

It can’t bypass M365 license requirements. Most of these add-ins require specific Microsoft 365 license tiers. Check before you contract.

It can’t make the Excel-vs-Sheets decision for you. ChatGPT works in both; Claude × M365 is Microsoft-only. If your shop runs a mix of Excel and Google Sheets, the OpenAI option’s cross-platform behavior is genuinely useful. If you’re Microsoft-only, this advantage doesn’t apply.

It can’t fix a poorly-scoped pilot. Don’t run a 3-vendor parallel pilot with no success criteria. Define what “winning” means before you start — ideally measurable outcomes like “time saved per analyst per week” or “average error rate in formula construction.”

It can’t substitute for the Microsoft Build refresh in 9 days. Whatever you decide today is going to need re-evaluation on May 20. Plan for that explicitly.

The bottom line

The Microsoft-stack AI market is in the most-fragmented state it’s been since Office launched. Four credible vendors, no clean substitute, Microsoft Build looming, and the pricing math still settling.

The default Q3 answer for most Microsoft shops: Copilot + Agent 365 as the foundation, plus one task-specific layer. The task-specific layer depends on what you actually do — Claude × M365 for cross-app, ChatGPT for Excel for spreadsheet-heavy, ServiceNow for compliance-driven, multi-vendor for best-of-breed.

Three things to do this week:

Map your shop into one of the 4 patterns above
Pilot ChatGPT for Excel before June 2 (the free window costs nothing)
Hold the multi-year contract refresh until after Microsoft Build (May 19-22)

If you want the structural Microsoft-stack AI procurement playbook — including the negotiation script for multi-vendor pairings, the audit-clause template for Agent 365 governance, and the vendor-onboarding sequence — that’s the focus of the microsoft-copilot course and copilot-cowork course.

v1.1 — May 20 refresh commitment: This post will be updated within 48 hours of the close of Microsoft Build (May 19-22). Specifically, the “what’s shipping” section and the 4 pairing patterns will be re-evaluated against announcements like Microsoft Agent 365 v2, GitHub Advanced Security AI integration, and Microsoft Defender-for-Cloud AI agents. Bookmark this URL or check back May 21-22.

Sources

v1.1 note (2026-05-10 evening): Updated within hours of initial publication after Perplexity Pro research surfaced concrete pricing data and analyst pairings: confirmed Agent 365 at $15/user/mo (per-user, list); E7 bundle math at $99 (~15% discount vs $117 unbundled); the July 1, 2026 11-20% price increases driving “renew early” buyer behavior; the 3 production procurement patterns (“Copilot-first”, “E7 anchor”, “Wait and cap”); the ChatGPT-Excel 44%→87% three-statement-modeling benchmark; Anthropic’s ~40% enterprise LLM revenue share; AMD + Synopsys Claude validation; and the canonical FSI/healthcare pairings from analyst commentary. Microsoft Build May 19 v1.2 refresh still planned per the original commitment banner.