Conversational BI Tools and Conversational Business Intelligence Software Compared by Price per Question

Conversational BI tools let a sales manager or a finance lead type "what was net revenue by region last quarter" and get the answer back as a number, a chart and, in the better tools, the SQL behind it. In October 2026 every major platform sells one, and they cost wildly different amounts for the same work: for a 10-person team asking 1,000 questions a month, list prices range from $0 (free allowances and trials that end on 31 December 2026 and 31 January 2027) to about $403 a month for Power BI Copilot. The meter is the real difference. Some tools charge per seat, some per message, some per token and some per hour of capacity, so you cannot compare quotes until you put them on the same workload.

That is what this page does. Every figure below was read from the vendor's own pricing page, rate card or documentation between 28 September and 2 October 2026, and the full working for each sits on our individual pricing pages, linked as we go.

The same team on eight conversational BI tools

The workload: 10 business users, about 100 questions each a month, 1,000 in total. List prices in US dollars, per month, for the conversational layer. Query compute on your warehouse is extra everywhere, because every one of these tools has to run the SQL it writes somewhere.

Three things stand out once the numbers sit side by side. First, the three lowest figures are temporary or capped: Google and Databricks are free because of trials with published end dates, and Amazon's free tier is a single pool of 1,000 prompts shared by everyone in the AWS account. Second, the per-seat and per-capacity products (ThoughtSpot, Power BI) cost several hundred dollars a month for a team this size before anyone asks a question. Third, only Snowflake prices the question itself, which makes it the only one you can forecast exactly from headcount, at 67 Platform Credits per 1,000 messages.

Per seat, per message, per token or per capacity hour

The meter decides how the bill behaves when usage grows, which matters more than the starting price.

Per message is Snowflake's model. Cortex Analyst charges 67 Platform Credits per 1,000 successful messages through its API, so a question costs $0.134 on Standard and $0.268 on Business Critical. Double the questions, double the bill. Our Snowflake Cortex Analyst pricing breakdown has the full rate card, including the separate AI Credits Snowflake introduced in September 2026 that negotiated discounts do not cover.

Per token is Google's. The Conversational Analytics agent, part of what Google calls Data Cloud agents, bills $3 per million input data tokens and $20 per million output data tokens. Output costs 6.7 times input, so verbose answers drive the bill more than the number of questions. Google originally ended the free trial on 30 September 2026 and has since extended it to 31 December 2026, so any pilot you run this quarter measures a free product.

Per DBU is Databricks'. Genie is free through 31 January 2027, then bills $0.070 per DBU with 150 free DBUs per user per month. Databricks publishes no DBU figure per question, so you cannot convert the allowance into a question count in advance, and the SQL Serverless compute that runs Genie's queries costs ten times the Genie rate. The Databricks Genie pricing article walks through that second layer.

Per capacity hour is Microsoft's. Copilot in Power BI has no license of its own and is not included in a Pro or Premium Per User seat. It needs a workspace on Fabric F2 or higher, which costs $262.80 a month pay-as-you-go or $156.33 on a one-year reservation, and Copilot then draws down that capacity alongside every refresh and report. The details, including why trial capacities do not qualify, are on Power BI Copilot license. Microsoft also retires the older, free Power BI Q&A in February 2027.

Per seat is ThoughtSpot's and, in a different form, ours. ThoughtSpot Pro is $50 per user a month with 25 Spotter AI questions per user, which works out to $2.00 per included question for a team that uses them all. AgentSQL Team is $59 a month billed yearly for up to 10 seats and 1,500 questions, about 4 cents a question at full use.

What separates the tools beyond price

Price per question is the easy part to compare. These four questions decide whether business users keep using the tool after the first week.

Does it need a semantic model before the first answer?

Cortex Analyst needs a YAML semantic model describing your tables, measures and joins. Looker's agent answers on top of LookML. Power BI Copilot is only as good as the dataset model under it. Those layers improve accuracy, but each one is a project that someone on the data team has to build and maintain before a manager asks anything. Text-to-SQL tools that read the live schema skip that step, which is why they can be running the same afternoon, and why they work best when column names are reasonably clear.

Can a reviewer see the SQL?

This is the accuracy question in practice. A tool that returns "Revenue in the West was $1.2M" with no query behind it asks the reader to trust it. A tool that shows the SQL lets an analyst see in ten seconds whether it filtered on the right date column or double-counted refunds. Cortex Analyst returns SQL through its API, Genie and Amazon Q show the generated query, and AgentSQL shows it beside every answer. When an answer still looks wrong after the SQL checks out, the cause is usually upstream, and a data lineage tool that traces which job last wrote the table is how teams find it.

Is the access read-only?

Business users asking questions should never be one prompt away from an UPDATE. Every tool here runs reads in normal use, but they differ in how much the setup enforces it. The safe pattern is a dedicated database role with SELECT only on the schemas people should see, which is how AgentSQL connects to Snowflake, BigQuery, Postgres and MySQL.

Does it work where the data already lives?

Warehouse features are tied to their warehouse: Cortex Analyst to Snowflake, Genie to Databricks, Amazon Q to Redshift, Google's agent to BigQuery and Looker. That is fine if everything lives in one place and a constraint if your product data is in Postgres and finance is in BigQuery. Suite copilots are tied to their BI tool. Standalone tools connect to several databases at once.

Which conversational BI tool fits which team

Already deep in one warehouse, with a data team to build the semantic model: start with that warehouse's own feature. Cortex Analyst on Snowflake, Genie on Databricks, Amazon Q on Redshift. The integration is native and the free allowances are generous, though two of them have end dates you should put in the calendar.

Already paying for Fabric capacity: Power BI Copilot is the natural choice, because the capacity is a sunk cost and Copilot only adds consumption. Buying F2 just to get Copilot for ten people is the most expensive row in the table.

Business users who need answers this week, from Postgres, MySQL, Snowflake or BigQuery, without a modeling project: a flat-priced text-to-SQL tool. AgentSQL connects read-only, writes the SQL against your live schema, returns a chart and a one-line answer, and shows the query every time, for $59 a month for up to 10 seats. The pillar on text to SQL explains how the generation works, and conversational analytics pricing lays out every vendor's meter in more depth.

A search-first BI replacement for the whole company: ThoughtSpot, priced per user, and best when you want the search interface to replace dashboards rather than sit beside them. The ThoughtSpot vs Tableau comparison covers that decision.

How to run a fair two-week trial

Pick twenty questions your team already knows the answers to, from easy ("orders last week") to awkward ("customers whose second order came within 30 days of the first"). Ask each tool all twenty, check every answer against the known figure, and count how many came back right without an analyst rephrasing the question. Then multiply the questions your team would really ask in a month by each tool's meter, using the rates above, and add the warehouse compute. The tool that scores highest on the twenty questions and still fits the budget at real volume is the one to buy. The console at the top of this page is a fair place to start: it answers against a demo store database, and every answer shows its SQL.