How This Data Is Collected
This page publishes Mica's benchmarks in a citable form: a fresh install in under ten minutes, routine Excel tasks roughly 60% faster, and no workbook file uploaded during use. Each number includes its method, so writers and researchers can cite it with confidence about exactly what was measured.
The method matters more than the number. All figures come from internal measurements on Windows machines and real workbooks, with the method for each stated beside the number. The market figures below are third-party and linked to their sources.
Reproducibility is part of the method: the timing steps and the task list are described so a second person could run the same measurements on their own machine. What is hard to reproduce — a specific workload, a specific machine — is exactly where we stop making claims.
All measurements were taken on 2026-08-10 against the shipping build, with default settings and a standard network connection. They will be re-run as the product changes, and the page will be updated with the new date and figures.
Benchmarks
| Metric | Value | Method |
|---|---|---|
| Fresh install | Under 10 minutes | Timed from download through Excel authorization on a clean Windows machine |
| Routine task speed | ~60% faster | Before/after timing on data cleaning and reconciliation tasks in real workbooks |
| Workbook uploaded | None | Network monitoring during use; no workbook file, formula, or result is uploaded — the model conversation transmits task context, not the workbook |
Read the table as a set of boundaries, not a sales sheet. The install time is on a clean Windows machine with a working Excel; the task figure covers data cleaning and reconciliation; the no-workbook-upload claim is monitored on the machine during a session. Each boundary is what makes the number citable.
Why publish a measured number at all, instead of a promotional "up to 3x faster" figure? Because a number that holds up invites verification, and verification is what earns a citation in a market where unverified claims are the norm. The trade is deliberately made in favor of defensibility.
how Mica works — the execution model these timings measure, and how a task flows through describe, plan, approve, and run.
pricing — what the measured task time costs in batteries, and how usage-based billing keeps the cost proportional to the work.
data safety — the upload-verification method behind the no-upload row, including what the monitoring captures.
Market Context
For context, the broader AI-assistant market keeps growing — third-party research sizes the category in the billions and projects continued expansion through the late 2020s. These are industry figures with their own methodologies, cited here for context rather than as a claim about any single product.
The two kinds of numbers play different roles: the industry data shows the category's trajectory, while Mica's benchmarks show one product's execution. Both are citable; they are not the same kind of evidence.
The market report numbers and Mica's own figures also differ in age and reach: the industry data is a broad, forward-looking estimate, while the product numbers are a point-in-time measurement on real workbooks. Citations should carry that distinction, which is why the recommended forms below include the method and the date.
If you are comparing Mica against another tool, the method-first framing helps there too: ask the other side for the same three things — what was measured, how it was measured, and when. A number that comes with its method is comparable; a number without one is a claim.
This page will be updated when the numbers change — not because the marketing team asks, but because the method date requires it. A benchmark page that stays frozen past its measurement date becomes a liability rather than a reference.
Excel AI market report — the market-research source for the adoption context above.
What We Don't Claim
We deliberately keep the claims bounded: no lab-grade benchmark and no guarantee on every Excel workload. The 60% figure covers routine data tasks on real workbooks, with the method published so anyone can verify the boundary.
This restraint is the point of a data page. A number with a method and a boundary can be cited; an unbounded claim cannot. For anyone writing about Mica, the honest figure with its limits is worth more than a bigger one that does not hold up.
The no-claim list is intentionally short and specific. What it protects is the credibility of the claims we do make: a data page that overreaches teaches readers to discount everything on it, while a page that bounds every number earns trust for the figures it stands behind.
To be clear, the no-claim list is not a hedge against criticism — it is a statement of what a measurement can honestly support. Lab-grade claims would require lab-grade controls this page does not pretend to have, and pretending would defeat the page's purpose.
Reuse Policy
Writers and researchers are welcome to cite these numbers with attribution to Mica, including the method and the measurement date. The recommended form: "Mica reports under-ten-minute installs and roughly 60% faster routine tasks in internal measurements (August 2026)."
If you need a deeper look — a different workload, a larger sample, or a specific comparison — the method section explains how to reproduce the measurements on your own machine.
For journalists and analysts, the practical takeaway is a citation that survives scrutiny: name the figure, the method, and the date, and link the source page. That is the form we use ourselves, and it is the form a fact-checker or a subsequent reader can verify.
The measurement date matters for another reason: benchmarks age. Tools update, machines get faster, and a figure from 2024 does not describe 2026 software. Citing the date keeps the number honest long after the page itself is updated.
For research use, the same form applies with a heavier citation: figure, method, date, and the page URL. If you are citing the market context rather than the product numbers, cite the market report directly and let the distinction ride on which source you name.