Sourcetable, a San Francisco–based spreadsheet startup, claimed on July 8, 2026, that its AI-powered spreadsheet achieved a perfect score of 24 out of 24 on the company’s own Sourcetable Benchmark, handily beating Microsoft 365 Copilot in Excel (19/24) and Google Sheets (17/24). The announcement—made via a press release timed to coincide with growing enterprise scrutiny of AI productivity tools—immediately raises questions for the millions of Windows users who rely on Copilot for spreadsheet work. But the results come with a critical caveat: the benchmark was designed and administered by the vendor itself, and no independent validation was provided.

What the Benchmark Actually Measured

Sourcetable disclosed that its benchmark evaluates an AI assistant’s ability to handle 24 distinct spreadsheet tasks grouped into four broad categories: formula generation, data analysis and insight, data cleaning and transformation, and error handling and explanation. Each task was scored on a pass/fail basis, with a point awarded only if the AI produced a correct, ready-to-use result without human intervention. The tasks were drawn from a mix of simple operations (like calculating a running total) and complex, multi-step workflows (like generating a pivot table from messy data and providing a narrative summary).

Microsoft 365 Copilot (specifically the Excel integration available in Current Channel build 16227.20280 or later, per Microsoft’s June 2026 update) scored 19 out of 24. Google Sheets, which relies on its Duet AI features, achieved 17. Sourcetable—which built its own AI layer on top of a familiar spreadsheet interface—aced every single item.

The company did not release the full list of tasks or the raw evaluation data, citing competitive sensitivity. However, Sourcetable’s CTO noted in the press release that Copilot struggled with tasks requiring deep contextual understanding across multiple sheets, while Google Sheets faltered on advanced formula debugging. That matches anecdotal reports from Windows forum discussions and user feedback: Copilot often excels at simple data queries but stumbles when asked to interpret nuanced business logic without extensive prompting.

What This Means for You

For Everyday Windows Users

If you rely on Copilot in Excel for quick insights, conditional formatting, or formula suggestions, the benchmark suggests you are using a capable but not flawless tool. A score of 19/24 means it gets most things right, but you should still verify its output—especially when working with financial data or nested formulas. The five failed tasks likely involve scenarios where the AI hallucinates a reference or misinterprets a column header. The practical takeaway: treat Copilot as a helpful junior analyst, not an autonomous expert. Always double-check its work before sharing reports.

For Power Users and IT Decision-Makers

The gap between Sourcetable’s perfect score and Copilot’s 19/24 may prompt a closer look at spreadsheet AI alternatives—but only if spreadsheet automation is a core part of your workflow. Sourcetable is still a niche player; it lacks the integration depth, enterprise security certifications, and ecosystem reach of Microsoft 365. For organizations standardized on Windows, OneDrive, and Teams, switching spreadsheet platforms is a non-trivial exercise. IT leaders should instead use this benchmark to start internal conversations about whether Copilot’s current performance meets their accuracy thresholds, and to watch for upcoming improvements (Microsoft has already committed to a major Excel AI update later in 2026, codenamed “Bucharest”).

For Developers Building on Copilot

Independent software vendors and internal development teams that extend Copilot’s capabilities via plugins or the Microsoft Graph should note the areas where Copilot lagged. If Sourcetable’s benchmark is even directionally accurate, the biggest shortcomings lie in cross-sheet reasoning and error-handling logic—areas where custom add-ins might temporarily fill the gap until Microsoft ships its next update.

How We Got Here

Three years ago, AI in spreadsheets was mostly limited to simple auto-complete and pattern recognition. The launch of Microsoft 365 Copilot in March 2023 changed the game, bringing large language models into the ribbon. Google followed with Duet AI for Workspace. Since then, both have iterated rapidly, but a persistent frustration has been the lack of standardized, independent benchmarks to compare these assistants.

Enter Sourcetable. Founded by ex-Meta and Flexport engineers, the startup launched its first AI features in 2025 and has been positioning itself as the “AI-native” alternative to legacy spreadsheets. Like Notion’s evolution from notes to databases, Sourcetable aims to rethink the spreadsheet paradigm around a chat interface and automated data pipelining. Its benchmark is clearly a marketing move, but it also fills a real information vacuum—even if the source is biased.

Microsoft, for its part, has steadily improved Copilot in Excel. The June 2026 update added support for natural-language data profiling (“What patterns do you see in column B?”) and better multi-sheet formula generation. Google Sheets’ Duet AI, meanwhile, remains constrained by the platform’s script-based automation model, which is less flexible than Excel’s Power Query and VBA heritage.

What to Do Now

  1. Test Copilot against your real-world tasks. Don’t trust a score alone. Pick five of your most frequent spreadsheet operations—maybe a weekly report, a budget template, or a data cleanup routine—and run them through Copilot. Document where it succeeds and where it fails. This gives you a personalized accuracy gauge.

  2. Engage with Microsoft’s feedback channels. If you find consistent failure modes (e.g., Copilot messes up date formatting in conditional rules), use the “thumbs down” button in the Copilot pane. Microsoft’s engineering team actively mines that data to prioritize fixes.

  3. Keep an eye on independent evaluations. No widely recognized third party has yet tested the Sourcetable Benchmark claims. Analysts at firms like Gartner and Forrester are likely to weigh in within the next quarter, and community-driven tests on platforms like WindowsForum.com may emerge. Wait for those before making any purchasing decisions.

  4. If you’re evaluating Sourcetable, take advantage of its free tier (spreadsheets up to 10,000 rows) to verify the perfect-score assertion in your own environment. Pay particular attention to data import/export fidelity with Excel—a frequent pain point when leaving the Microsoft ecosystem.

  5. For IT admins, review your organization’s AI usage policies. If users are already bringing unsanctioned AI spreadsheet tools into the enterprise, this benchmark could amplify shadow IT. Consider whether your compliance and data-residency requirements could even accommodate a platform like Sourcetable, which runs on AWS and may not yet offer Azure Active Directory integration.

Outlook

Sourcetable’s benchmark is as much a challenge to Microsoft as it is a marketing ploy. Expect Microsoft to respond—either with its own comparative testing, with faster feature updates, or both. The bigger story is the rapid maturation of AI in productivity tools. By the end of 2026, the accuracy gap between a startup’s purpose-built AI and a tech giant’s broad assistant may narrow considerably, making spreadsheet users the real winners. For now, Windows users should stay curious, stay critical, and let the data—not just the press releases—guide them.