Conversational analytics
What conversational analytics costs
Most of the running cost of conversational analytics sits in things a dbt team already pays for: the warehouse and an AI subscription. The real cost is people: who defines the metrics and who acts on the gaps, and Mark Your Data closes that loop to your data team.
What drives the cost
| Cost driver | What drives it | What to expect | How Mark Your Data Assistant keeps it down |
|---|---|---|---|
| Hosting the application | An always-on service that receives the chat messages. | Under €100 a month at list price (estimate). | It runs in your own cloud project. |
| LLM usage | Tokens per question and the model you choose. | With Gemini, under one cent per question on average, measured across our demo runs on sample data. It depends on the provider, models and task complexity. | A fixed pipeline without an agent loop. Where a metric exists, the model picks it instead of writing SQL. Result rows go to the model only if you allow it. Routing can run on a small decision model. Cost per request is in the log, with a limit per request. |
| Database | The data scanned per query. | Your existing warehouse. Good data modelling, such as marts, partitioning and metrics on aggregated models, keeps it down. | Read-only access. Governed answers run the SQL of the metrics your team reviewed. You can cap the data a query scans. |
| Licence | Mark Your Data. | Setup plus a yearly service, per organisation, priced after the POC. | You decide after the POC, once you've seen it work on your own questions. |
| Maintenance time | Defining metrics and acting on gaps. | Depends on how fast your team closes gaps. The first pilot will measure it. | The closed loop: the log shows which gaps users hit and why, and a fix is one dbt pull request, so you fix what users actually hit, once. |
Build the business case with your own numbers
Start with the time your team spends on ad-hoc questions today, using your own numbers. Then add:
- Waiting time for the business: how long a question waits in the queue today. Answers took 3 to 8 seconds in testing on sample data.
- Disputed numbers: measure the share of governed answers at the start, then every month.
- Your dbt project: every closed gap is a definition your dashboards can use too.
- The alternatives: compare building text-to-SQL yourself, a platform assistant and Mark Your Data Assistant on total cost, including who does the loop and how portable your definitions are.
What it means for you
- Manager: You can build the case with your own numbers, and compare the options on total cost.
- Analytics engineer: You can see tokens and cost per request in the log, and cap the data a query scans.
Questions and answers
How much does conversational analytics cost?
Conversational analytics costs hosting, model usage, warehouse usage, a licence and your team's time. For Mark Your Data Assistant, the licence is setup plus a yearly service, per organisation, priced after the POC. Hosting and warehouse usage run on your own cloud accounts. The real cost is people: who defines the metrics and who acts on the gaps.
What does one question cost in AI and warehouse usage?
With Gemini, model usage for Mark Your Data Assistant is under one cent per question on average, measured across our demo runs on sample data, because a fixed pipeline has no agent loop and the model picks an existing metric instead of writing SQL. It depends on the provider, models and task complexity. Warehouse cost depends on data scanned.
How much engineering time does it take to start and to keep up?
Mark Your Data Assistant can start with one metric or saved query, and the time after that depends on how fast your team closes gaps. The log shows which gaps users hit and why, and each fix is one dbt pull request, so you fix what users actually hit, once. The first pilot will measure the hours.
How do we build the business case for conversational analytics?
Start with the time your team spends on ad-hoc questions today, using your own numbers. Then compare building it yourself, a platform assistant and Mark Your Data Assistant on total cost, including who does the loop and how portable your definitions are.
Isn't a platform assistant like Genie or Copilot cheaper?
Often cheaper to start, yes. Genie One and Genie Agents are free for users through 31 January 2027, and Copilot in Power BI needs paid capacity. So ask what a question costs once billing starts, and how you keep control of gaps and spend. Mark Your Data Assistant logs cost per request and closes the loop to your data team.