What is Conversational Analytics?
Conversational Analytics lets you ask plain‑language questions about your Coconut data and get answers as charts, tables, or summaries—without building reports or knowing SQL (Structured Query Language). It uses the same governed analytics layer behind our dashboards, so metric definitions and permissions stay consistent and trusted.
Typical questions include:
“Over the last 30 days, which branches had the highest no‑show rate?”
“How many appointments did we complete last quarter by channel?”
“Show me trends in booked appointments this year by location.”
Instead of hunting for the right report, you ask your question in everyday language and Conversational Analytics returns a visualization or table that directly answers it.
How to access Conversational Analytics
If your organization has access to Conversational Analytics and your user role allows it, you can open it from Analytics in two main ways.
From the Analytics homepage
Go to Analytics.
Select the Ask button in the top‑right corner of the Analytics homepage.
This opens the Conversational Analytics view, where you’ll see:
A chat‑style interface on the right, where you type your questions.
Your recent conversations on the left, so you can pick up where you left off.
From here, you’ll typically start with a Coconut‑configured Agent (see below) so you don’t have to choose a dataset or Explore yourself.
From an Analytics view
Open any Advanced Analytics report.
At the top of the page, open the Go To dropdown.
Choose Conversational Analytics to jump directly into the conversational experience, using the same underlying data that powers the report you’re viewing.
This is especially useful if you’re already in a specific Analytics view and want to ask follow‑up questions without leaving that context.
Note: Conversational Analytics is currently available as a Beta feature. Access may be limited to certain organizations, roles, or cohorts while we refine the experience. If you don’t have access but would like to, contact your internal admin or your Coconut representative.
Choosing how to start: Agent or Explore
Behind the scenes, Conversational Analytics can answer questions in two main ways:
Using a data agent (recommended default)
Starting from a specific Explore (dataset)
In both cases, your questions are run against the same governed Looker model that powers your dashboards and Explores, and your existing Analytics permissions continue to apply.
Using an Agent (recommended starting point)
When you select an Agent in Conversational Analytics, you’re chatting with a pre‑configured, Coconut‑managed assistant that already knows:
Which Explores to use – each agent can include up to five underlying Explores.
How to interpret your business language – through instructions, examples, and preferred metrics tailored to common banking and operations questions.
Agents are created and tuned by Coconut, then shared so you can simply pick an Agent and start asking questions. You’ll typically reach them by:
Opening Conversational Analytics from the Ask button or Go To → Conversational Analytics, then
Choosing a New conversation or selecting a shared Agent.
Use an Agent when:
You want a simple, guided starting point without choosing a dataset or Explore first.
You’re asking broader business questions that may span multiple datasets (for example, engagements plus branches or staff).
You’d rather type less for your question - the Agent has common business terms, synonyms, rules, and instructions built in by default.
Starting from an Explore
If you work directly in Looker, you can also start a conversation from an Explore:
In Looker, navigate to a specific Explore and choose Start a conversation to open a chat that is anchored to that one Explore.
Any questions you ask in that conversation will be answered using only that Explore’s fields and filters.
Use an Explore‑based conversation when:
You already know which dataset you want (for example, an “Engagements overview” Explore).
You want tight control over what’s in scope and to validate answers directly against the fields and filters you see in Explore.
You’re extending or double‑checking an existing Explore query and plan to open results back in Explore using Open in Explore for further tweaking.
You’re willing to (or want to) provide your own metric definitions, rules, instructions, synonyms, or calculation rules.
Which option should I choose?
As a rule of thumb:
Most users should start with an Agent. It’s the fastest way to get trustworthy answers from the right combination of datasets, without picking Explores yourself.
Use Explore‑based conversations if you’re an advanced user who already builds or validates Explores in Looker and wants fine‑grained control over the dataset in scope.
No matter which you choose, the same governance and permissions apply, and you can always open a given answer in Explore for deeper analysis.
Asking Questions
Once you’re in Conversational Analytics, you’ll see your recent chats on the left and a chat‑style interface on the right where you type your questions.
Start a new conversation
Click New conversation (or select an available Agent).
Type your question in everyday language, for example:
“Show total completed appointments by branch for the last 90 days.”
“Compare booked vs. completed appointments this quarter.”
Press Enter to send.
Conversational Analytics interprets your question using the Agent or Explore that’s in scope, then returns an answer as a chart, table, or text summary.
Refine your question with follow‑ups
You don’t need to start over for every variation. The conversation remembers what you’ve already asked, so you can refine in place. For example:
“Now break that down by channel.”
“Filter to only Branch A and Branch B.”
“Show this as a bar chart instead of a table.”
This is often the fastest way to iterate: ask a broad question first, then narrow by segment, time period, or visualization.
Use filters and timeframes in your wording
Conversational Analytics understands common filters like date ranges, locations, staff, and channels when you mention them in your question. For best results, try to include:
Time frame: “last 30 days”, “this quarter”, “last year”
Segment: “by branch”, “by staff member”, “by channel”
Conditions: “where no‑show rate is above 5%”, “only booked appointments”
If an answer looks off, adding more precise timeframes or filters is often enough to correct it.
Working with your results
When you submit a question, Conversational Analytics will:
Run a governed query against the same analytics model used by your dashboards and Explores.
Return a visualization, table, or summary that directly answers your question.
From there, you can:
Switch views by asking for a different visualization (for example, “show this as a line chart over time”).
Ask follow‑ups to drill deeper (for example, “Which three branches are highest?”).
Rename the conversation so you can easily find it later in your recent chats (for example, “Q2 Branch No‑Shows”).
Open in Explore to see the underlying query and modify it directly in Looker, if your permissions allow it.
For any high‑stakes reporting or executive presentations, we recommend double‑checking critical numbers against your standard dashboards or trusted Explores, just as you would with any new analysis.
Privacy, security, and safeguards
Conversational Analytics is built on top of the same governed analytics layer that powers your existing Coconut dashboards and reports:
Your existing permissions apply. Users can only see data they’re already allowed to see in Analytics; Conversational Analytics does not grant new data access on its own.
Conversations stay within your environment. Coconut cannot see individual Conversational Analytics chat histories or the specific answers returned; we rely on explicit feedback you choose to share when you report an issue or success.
Enterprise‑grade governance. All queries are executed through a governed analytics model with row‑ and column‑level controls, helping ensure that answers reflect your organization’s approved metric definitions and access rules.
As with any early‑stage AI experience, results can occasionally be incomplete or incorrect. If something looks off:
Rephrase your question with more detail (timeframe, branches, channels, etc.).
Compare against an existing dashboard or Explore.
Use the in‑product Give feedback option (where available) to flag especially helpful or confusing responses so we can continue improving the experience.