RAG
Docs. Indexes your documentation and answers how-does-this-work questions. Auto-re-indexes on GitHub merges. Example: "How is churn calculated?"
A Tableau AI Assistant for BI teams
Your team has questions. Stop making them wait
d.Ask is a Tableau AI Assistant that puts a plain-English layer on top of your Tableau dashboards, databases, and documentation. No new dashboards. No SQL. Just answers in minutes, not days.
Built by Data Dune: BI consultants for Pfizer, McKinsey, Siemens.
"Deep domain knowledge around Tableau, data visualization and integration" Dick Olsson, Director, Principal Engineering Lead @ Pfizer
- 13x
- faster than a Jira ticket
- 41x
- cheaper per question
- 40h+
- analyst hours saved per week
The problem
Sound familiar?
Your analysts are good. They shouldn't be spending their days answering what's-the-number questions.
Every question is a Jira ticket.
Business users wait two days for a number that takes five minutes to pull. Your analysts spend 40% of their time on lookups, not analysis.
Only three people know how this works.
Business logic lives in Tableau calc fields, scattered READMEs, and Slack threads. When someone leaves, the knowledge goes with them.
The dashboard doesn't slice that way.
Rigid dashboards answer yesterday's questions. Today's question needs a new view. That is next sprint at best.
Architecture
A Tableau AI Assistant with three sources.
Ask a question in plain English. d.Ask queries three sources in parallel, cross-checks them, and returns one sourced answer. No SQL needed.
You ask a question
Plain English. No syntax. No training required.
RAG Agent
Searches your documentation: READMEs, metric definitions, known issues.
Database Agent
Generates SQL and queries your database. Supabase, Snowflake, Postgres.
Tableau MCP Agent
Connects to Tableau Cloud live. Reads calc fields, links dashboards, checks formulas.
Combiner Agent
Cross-checks all three. Returns one sourced answer. Flags contradictions.
Sources
How the agents answer.
d.Ask sees documentation, numbers, Tableau metadata, and one combined answer side by side.
Database
Numbers. Generates and runs SQL against your actual data. Source of truth for every number, count, and date. Example: "What's Q3 churn this quarter?"
Tableau MCP
Dashboards. Reads live metadata from Tableau Cloud: calc field formulas, parameters, direct links to workbooks and views. Example: "What formula is behind this metric?"
Combiner
Final answer. Sees all three outputs side by side. DB wins on numbers. RAG wins on context. MCP wins on Tableau. One answer with citations.
The talk
The thirty-five minute version, from the conference stage.
How the four agents are wired, what breaks when the layer underneath disagrees with itself, and the parts that did not work first time.
Nothing is requested from Google until you press play. If you would rather not, the talk is also on YouTube.
Tableau Conference 2026. Thirty five minutes.
What this costs your team right now.
The same four questions. Your analyst vs. d.Ask. These numbers are measured, not estimated.
- $95.85
- analyst cost for 4 questions
- $2.32
- d.Ask cost for same 4 questions
- 88
- FTE-equivalent throughput
- $29k
- per month replaces
- 80%
- of data questions never get asked. d.Ask makes asking free.
135 min at standard rate
~10 minutes
from one system at 50k q/mo
$600k in analyst capacity
ROI
Your analyst today vs. d.Ask.
| What | Your analyst today | d.Ask |
|---|---|---|
| Time per question | 135 min | ~10 min |
| Cost per 4 questions | $95.85 | $2.32 |
| Available | Business hours | 24/7 |
| Sources cited | Rarely | Every answer |
| Handles more questions | Hire more analysts | Same system |
Use cases
Questions your team asks every week.
Directors, engineers, and team leads are all waiting on data. Here is what d.Ask returns in under two minutes.
Director of Analytics
"Why is Q3 revenue flat despite more deals?" d.Ask finds: deal count +15%, but avg deal value -30%. Team shifted to smaller, faster deals while the larger EMEA pipeline stalled. Links to the revenue dashboard included. Sources: Database, Tableau.
VP of BI
"How is churn rate calculated and what's the actual number?" Returns the Tableau formula, the current figure from the database, and doc context explaining why the 3% to 8% jump traces to a silent pipeline failure from August. Sources: Database, Docs, Tableau.
Data Engineer
"The delivery_times dashboard hasn't updated. What's going on?" d.Ask traces: pipeline ran but loaded 0 rows after Aug 15. Source API renamed a field. Links directly to the pipeline README with the fix. Sources: Database, Docs.
Team Lead
"Which open deal should I focus on next? Why that one?" d.Ask picks a deal based on value, region, and consultant load. It remembers the context from your previous question. No repeating yourself. Sources: Database.
Get started
Start free. Go production with us.
The free workflow gets you running in an afternoon. Production-ready takes expertise. That is where Data Dune comes in.
Free download. Free forever.
Working n8n workflow with 4 AI agents. In-built RAG ready to load your docs. Works with Tableau Cloud + your database. 5-step setup guide included. Production-tuned prompts, auth, permissions, monitoring, persistent memory, and Slack / Teams integration are not included.
Data Dune implementation, from one week's engagement
Everything in the free workflow. Prompts tuned to your data and business logic. Auth, permissions, row-level security. Persistent conversation memory. Monitoring, alerting, SLAs. Slack or Teams integration. Multi-tenant where you need it. 20 to 25 hours to production-ready.
FAQ
Questions buyers ask.
The practical bits: security, accuracy, fit, ownership, and how quickly you can get moving.
How does d.Ask handle data security and row-level access?
It inherits permissions from your existing database and Tableau, including row-level access. d.Ask never stores query results, and production deployments run in your tenancy or on-prem.
How accurate is the SQL d.Ask generates?
The Database agent is constrained to known schemas and validates queries before running. Results are cross-checked against the Tableau and RAG agents; disagreements are flagged, never silently resolved.
Why d.Ask instead of Tableau Pulse, Power BI Copilot, Looker or ThoughtSpot?
Those tools answer questions about dashboards you already built inside one BI stack. d.Ask sits on top of whatever BI tool you already run and answers across your database, docs, and Tableau metadata combined. That includes questions no dashboard exists for. It's a Tableau AI Assistant, not a replacement BI platform.
Where does d.Ask live? Does our data leave our environment?
100% in your environment. d.Ask runs in your tenancy or on-prem, queries your database and Tableau directly, and stores nothing. We provide the know-how and the setup; the system, the data, and the keys stay with you.
Which LLM does it use, and is our data sent to it?
Bring your own. d.Ask works with OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, or a private model. Only the question text and minimal schema context go to the LLM, never raw query results, customer records, or full table contents. For regulated environments we deploy against a private endpoint.
How much does it cost to run?
The free workflow is free forever. LLM running cost is roughly $0.50–$0.60 per 4 questions on a standard model. That is about 41x cheaper than analyst time. Our implementation is a fixed-scope engagement from one week; no per-seat fees, no SaaS lock-in.
What happens after Data Dune leaves? Can our team maintain it?
The workflow runs in n8n with documented prompts, agents, and integrations. Your engineers own it; we provide handover docs and 30 days of post-implementation support.
How long until it's live?
The free workflow runs in an afternoon. Production-ready setup, including auth, monitoring, and your prompts, is typically a 2-3 week engagement.
Next step
A prototype takes hours. Production takes expertise.
We build d.Ask into your stack. Your Tableau. Your database. Your docs. Your team gets answers in minutes, not days.
No credit card. No sales call required. Works in an afternoon.