What the agent does

The agent is the analyst you hand questions to. Ask in plain English and it works on your data: it finds the right tables, writes and runs the queries, runs the statistics, and builds the chart, report, deck or dashboard.

next.dimensionlabs.ioProduct footage

Ask a question, get a chart. Recorded in the product.

It is built for the deep questions, the ones that start with “why”:

• Why is revenue down?
• Why are these accounts churning?
• Why is this product launch underperforming?
• Why is this segment behaving differently from that one?

New to the agent? Start with Your first chat, then read Quick answers and analysis plans.

An agent, not a chatbot

A chatbot answers from what it already knows. The agent takes actions on your data: it finds the tables, runs the queries and tests, and builds the output.

That means a system that explores for you. It still needs direction, but you work like an executive: it finds the opportunities, lays out the options and drafts the analysis, and you decide.

In practice the agent brings together the skills of a market researcher, a statistician and an expert in your industry. It uses them when the analysis calls for them, reports on its progress as it goes, and checks in with you along the way.

How it works through a question

1. Plans. It writes a short task list before it starts.
2. Inspects your data. It checks which tables, Dimensions and columns exist, whether they’re populated, and what dates they cover. This is where it decides whether your question can be answered yet.
3. Runs queries. Counts, joins, filters, comparisons, transcript searches. You can watch each step.
4. Writes back. A headline answer, how it got there, the numbers, quotes from the underlying conversations, and a suggested next step.

Your first chat walks through a full example, including what an answer and a refusal look like.

What it reasons over

The agent joins your conversation data (what customers said) with your structured business data (what they did and what the business measures). Within that, it works from three kinds of fields:

• Metadata you provide: the columns that came with your conversations, such as rating, store, plan or channel, and the business data files you upload.
• Standard Dimensions: the general fields every workspace has available, such as sentiment and contact reason, called System Prompts in the app. They are filled in by a backfill, or every day if Dimension Labs has turned on daily enrichment for your workspace. See When enrichment runs.
• Custom Dimensions: the fields you design for your business. See Design custom Dimensions.

The limit is not what the agent can think about. The limit is what data it can see. As you add sources and Dimensions, the range of questions it can answer well grows. See What the agent can see.

What it’s good at

• Questions about why. “Why did one-star reviews rise in March?” “Which contact reasons come before a cancellation?”
• Joining what customers said to what they did. Conversations next to orders, renewals, revenue.
• Testing whether a difference is real. It runs statistical tests to separate real, meaningful differences from the variation you’d expect by chance, and tells you when a gap could be chance. It does this on its own when the analysis calls for it, and when you ask for a specific test or comparison. See How we prove a driver.
• Charts and breakdowns. Trends over time, segment comparisons, distributions and correlations. Ask for the rows behind any chart or number.
• Building outputs. Reports, decks, dashboards and charts, in your company’s brand. See Reports, decks and sharing.
• Helping design Dimensions. When your data doesn’t yet capture a concept, it says so and can help you add it: it samples your data and drafts a definition with you. See Designing Dimensions with the agent and Design custom Dimensions.

How it differs from a BI tool

• You don’t need SQL or your schema. The agent works out which fields to pull and how to join them.
• It uses your definitions. It reasons over your custom Dimensions and follows your project’s standing instructions, so its answers reflect how your team defines the problem, not a generic template.
• It points out gaps. When your data can’t support an answer, it says so and suggests how to fill the gap. Often that’s a new Dimension.
• It explains why a finding matters, not just what the number is.

What makes it yours

Out of the box the agent is general. Two things make it specific to your business:

• Your Dimensions define what it reasons over. See Design custom Dimensions.
• Your standing instructions shape how it interprets your data and how it responds: your terms, your caveats, how your reports should read. See Projects and standing instructions.

Together, they are why the same question on the same data gives your team a better answer than it would give anyone else.

What it won’t do

• Guess. If the data can’t support an answer, it says what’s missing and what would fill the gap.
• See data you haven’t given it. The limit is the data in your workspace.
• Do as well with vague questions. “Tell me everything about my customers” gets a less useful answer than “Which customer segments are showing rising cancellation intent this quarter?” See Ask good questions.
• Manage your account. The agent reasons over your data, not your account. For login, billing, account settings or anything else, email [email protected].

Next

• Your first chat
• Quick answers and analysis plans
• Ask good questions
• What the agent can see
• Projects and standing instructions