What Mosaic adds.

Start a project from a corporate data lake, warehouse, or other governed source instead of only a file upload. Connect permission-aware access to customer-owned storage and SQL-backed tables so the study stays in one workspace rather than a chain of exports.

A product API key loads the project into Python or R over Apache Arrow — into pandas, Polars, or tibbles — without a CSV round trip. Run packages and models you choose, then write columns back to the same project for charts, banners, drivers, and stories.

Pick one column. Every other question is ordered by how strongly it differs. View the result as charts or a table, sorted by significance or by data order. A coded theme column can be ranked the same way.

Each segment is compared with everyone else, using a two-proportion z-test. Holm is applied within each answer. Cohen's h is the effect size. A shaded cell is higher or lower than the rest of the sample.

Each cell is compared with the count a chi-square model expects. The shading uses the standardized residual. Holm is applied within each answer column.

Pick a segment from the menu, or click its column header. Every other segment is tested against that baseline. Total can be the baseline. Same z-test and Cohen's h. Holm is applied within that answer and that baseline.

Build the banner in the product: bases, column percentages, indices, significance letters, and Holm. Rank the questions, then ask a question about the table you built.

A workbook written from the in-product banner: categorical and numeric tables, battery summaries, net rows, optional mean, standard deviation, and median rows, significance shading, and brand styling. A spec can be drafted and published through the agent.

Counts or shares as horizontal bars. A horizontal bar is the same view when you are comparing combinations.

Counts or shares as vertical columns, including numeric values by group.

Each row is the mix of answers. A Likert scale can sit on its neutral point, with that neutral block in the middle or between the two sides.

The same composition view, drawn as vertical columns.

Side-by-side bars for each combination, rather than one stacked bar.

A line across ordered categories, for counts or for a numeric measure.

Several series stacked on a line, so you see the total and the mix together.

The stacked line filled in, for the share of each series across an ordered axis.

A trend over time, for one series or for combinations, including a numeric trend by group.

The dense rank of each series at every time point, instead of the underlying value.

Each category as a dot sized by count or share. Bubble size can fit the data or run from 0 to 100%. The same plot can show individual numeric values by group.

The spread of a numeric measure by group, drawn horizontally or vertically.

How many responses fall in each value range.

The typical value in each group, as bars.

The relationship between two numeric columns, or combinations plotted on two axes.

A scatter with quadrant guides and a summary for each group.

The correlation structure of numeric columns in two dimensions.

A shaded grid of the values in a numeric pivot.

Slices show the share of each category.

The most common words in a text column, at a glance.

Promoters, passives, and detractors, and the same balance cut by segment.

The exact count for each answer, or for each numeric range, in a sortable table.

The count for every combination, on the same cut as the chart.

Key statistics by group, in a table.

Small multiples, one panel for each value of a third column.

Brand metrics over survey waves, as lines.

Clustered columns for each brand at every wave.

Each brand as a dumbbell between two survey waves.

The signed change in a brand metric between two waves.

A long category list defaults to the top 15, with the rest grouped as Other. Line, dot, and bubble charts can show a logo on each category.

A battery of just-about-right scales becomes one matrix. Each cell is the share who answered just about right. Colour and an arrow show which end the other answers lean toward, using that question’s own wording, such as too little or too much. Sort by survey order, or by the just-about-right share from high to low or low to high.

Sort a chart by value, A to Z, or natural order for scales, dates, age bands, and waves. Likert and rating batteries can sort by overall balance, positive response, net positive, or average score. Two series can sort by the size of the difference. Other orders cover strongest preference, how evenly answers are spread, and how unusual a pattern is against the overall result. Ranked differences can sort by significance or by data order.

Correlation PCA on a battery. Loadings, variance explained, a component map, and optional Varimax rotation.

Treat a battery as category entry points. Mental penetration, mental market share, network size, share of mind, and advantage, plus a brand-by-situation matrix with audience or wave cuts.

Latent class sits alongside K-means, K-medoids, and HDBSCAN. Compare candidates on silhouette, PERMANOVA, and, for latent class, AIC, BIC, and entropy, then save the one you keep as a column.

Ordinal Elastic Net and Multinomial Elastic Net, alongside logistic regression, linear regression, Random Forest, and Elastic Net. The model can be chosen automatically. Class-specific coefficients show up when one ranking would hide a category.

CASE WHEN and window aggregates, on top of NPS, CSAT, CES, UMUX, bins, weights, and arithmetic formulas. Each calculation records the columns it depends on.

A segment map places respondents in two dimensions: principal components of the inputs, a Gower dissimilarity map, or class-membership probabilities for latent class. Each segment also shows its size and the characteristics that set it apart.

Paste a discussion guide. Mosaic creates one text column per question. Those columns, and reviewed theme codes, can then sit in the ranked-differences banner, key drivers, and clustering like any other variable.

Ask Maple works on the project in front of you. It can clean the data: rename columns, recode categories, and group or ungroup questions. It can apply brand colors, category colors, Likert colors, and numeric colors. It can also build charts and tables, create calculated columns, run models, and explain the result on screen. The numbers come from the statistical engine.

The agent can create a story dashboard, a PDF report, and a PowerPoint deck from the analysis, and leave each one for you to review.

Cursor, Claude, and ChatGPT can use the same project, with the same permissions as the browser.

A filter on a story dashboard updates the live charts and the key-driver results. It can apply to the whole dashboard, one page, or selected blocks. A story can draw on more than one project, and a published release stays separate from the draft.