Text dimensionality
Structure for the 80% of data that has none.
Structured data has shape: rows, columns, schemas. Calls, chats, tickets and surveys have none, no consistent fields to filter, trend or build metrics on. Dimensions are the stable attributes inside that text, written onto every record as typed fields.
Why BI tools can't analyze unstructured data
There is nothing to group by. A transcript is one long string. Until each one carries the same set of fields, you can only read samples of it.
How we help
We automate the extraction pipeline: no hand-tagging, no prompt engineering per project. Dimensions aren't fixed. Adjust the logic as needs evolve and the change propagates consistently across your full dataset.
How Dimensions are used
Dimensions are the schema layer for unstructured data. Once text has them, they behave like any structured field: queryable, filterable, joinable. That is what makes reports, dashboards and monitors possible on raw conversations.
Example Dimensions
A sample of the Dimensions we build, across 11 categories and ready on every plan. We create hundreds more, and custom Dimensions are written in plain English.
How strongly the person feels, independent of whether it's positive or negative, surfaces your highest-passion and highest-risk voices.
The stakes behind a complaint, from mild annoyance to serious harm, as framed by the customer.
The true feeling expressed in the words themselves, extracted from the prose, not the star rating or survey score.
A modeled 1–5 satisfaction score inferred purely from the qualitative text of an interaction.
The likely promoter / passive / detractor classification read from language, even when no score was given.
The primary reason the person reached out or showed up.
How hard the customer had to work to get value or resolution.
Whether this is a new or returning customer, read from the narrative rather than a CRM flag.
Whether the exchange is a complaint, question, praise, cancellation, or sales inquiry.
Where the person sits in their journey with you, prospect, new, onboarding, active, at-risk, or churned.
The dominant reason the customer bought or visited.
Where the issue stands by the end of the interaction, resolved, pending, or unresolved.
The kind of transaction the text is about, sales vs. service, purchase vs. return, dine-in vs. delivery.
The specific business or personal use case the person references.
How well specific needs were accommodated and the courtesy shown.
The energy, mood, and sensory experience of the venue.
The customer's assessment of the core product or offering, food, course, device reliability.
Experience with the app, online ordering, or self-service surfaces not visible in structured data.
The physical condition and cleanliness of the space.
Whether the order or request was fulfilled correctly.
The single most operationally important thing that went wrong.
The single dominant subject of the conversation.
What happened when the customer raised an issue or asked for help.
What the customer says about timeliness and wait time.
The character of the human or service interaction, from exceptional to hostile.
How the customer reacts to gamification and engagement features, streaks, rewards, notifications.
The specific feature or capability the person engaged with.
A specific improvement, fix, or missing capability the customer asks for.
Early signal on a newly launched or updated product, critical for catching launch-quality issues fast.
Which product, model, brand, or module the text is about.
Whether dissatisfaction stems from the product itself or from how it was serviced.
The specific hardware or software failure, and where in the lifecycle it occurred.
The specific billing or charge dispute described.
Exact monetary amounts named in the text, disputes, fees, promises, refunds.
The direction and size of a price, fee, or premium change the customer calls out.
How the customer frames what they got for what they paid.
Reaction to subscriptions, paywalls, ads, and monetization changes.
The concrete business outcome the customer credits to your product.
A composite health read inferred from all the signals in a message.
Positive purchase intent and where the deal sits in the sales process.
Language suggesting the customer may leave, not renew, or switch away.
Language pointing to more seats, new use cases, upsell, or deeper adoption.
An explicit loyalty-program or brand-loyalty stance, the retention stake.
The concern or hesitation blocking the deal.
Whether the person would recommend you, or actively warns others off.
Whether the customer returned, refunded, or is considering it.
The operational owner who could act on the feedback, the routing tag that turns voice into a task.
Language or situations that should be routed for compliance review.
Reports of lost data, progress, or access.
Risk of escalation to a supervisor, social media, legal action, or a regulator.
The distance between what the customer was promised or expected and what they actually got.
Mentions of lawyers, regulators, complaints bodies or rights that could create legal or regulatory exposure.
Reports of injury, hazard, contamination or anything else that puts a person's safety at risk.
Whether the customer feels they were told the truth, or believes they were misled, ignored or deceived.
How time-sensitive the request is, including any date, renewal or deadline the customer names.
How the customer positions you against the alternatives they are weighing or already use.
A specific capability the customer says a competitor does better, worse or differently.
Which competitor the customer names, and whether they are evaluating, switching to or leaving it.
The tools, vendors or process the customer uses today, and what they would replace.
Outside forces the customer cites, such as the economy, budgets, seasonality or industry change.
The type of customer speaking, such as consumer, small business or enterprise, read from what they say.
Who the customer was with, such as family, kids, a partner, colleagues or a large group.
The store, site, city or region the customer is talking about.
Staff members the customer names, and whether the mention is praise or a complaint.
The customer's role or job, such as owner, admin, buyer or end user, as they describe it.
Where the customer bought or engaged, such as in store, online, in app, by phone or through a partner.
How the customer feels about being served by AI at all, from welcome to unacceptable.
How an AI assistant went wrong, such as misunderstanding, a wrong answer, a loop or a dead end.
How the customer compares the AI's help with help from a person.
Requests that are routine enough to be handled by self-service or automation.
Whether the bot resolved the request on its own or the customer had to go elsewhere.
Each time the customer asks for a person, and whether they got one.
What the customer tried on their own before reaching out, such as help articles, the app or the bot.
The exact words that capture what the customer loved, ready to quote.
The single sentence that best captures the conversation, kept word for word.
The exact words that capture what went wrong, ready to quote.
The passage that carries the strongest feeling, kept word for word as evidence for the score.
A short, plain summary of what happened and where it stands.
What we're learning about language
Five published papers on how reliably AI recovers what people meant and rated.
See our researchDimensions at work
Every artifact in the gallery was built on Dimensions, from Panera store diagnostics to NBA fan intelligence.
Check out artifactsStop reporting. Start understanding.
Start with one file. Ask why a number moved. Plans from $299 a month, no demo and no contract on monthly plans.