Ad Hoc Deflector
Handle routine data requests without dropping everything
A Slack bot that intercepts routine data requests, searches existing dashboards and past queries for the answer, and only generates SQL when the request is clear and safe. Analysts step in when it genuinely needs human judgment.
INGREDIENTS
PROMPT
Create a skill called "Ad Hoc Deflector". Set up a Slack sub-agent that monitors a designated channel for data requests. When someone asks a question (e.g., "What was our revenue last month by region?"), the agent should: (1) Search existing dashboards and saved queries for an existing answer — if found, reply with a link and the relevant data. (2) If no existing answer, parse the question into a SQL query, show the query for transparency, and execute it only in read-only mode when the request is clear and safe. (3) If the question is ambiguous, high-risk, or requires judgment (e.g., "Why did churn go up?"), escalate to a human analyst with context instead of guessing. Log every request, whether it was auto-answered or escalated, and the time it would have taken a human. Generate a weekly summary showing volume, auto-answer rate, and recurring questions that should become dashboards.
How It Works
Every analyst knows the pain: 15 Slack DMs asking for "quick pulls" that each
take 30 minutes. Most of these questions have already been answered somewhere —
in a dashboard nobody remembers, a query someone ran last month, or an email
thread from Q3. This skill is the first line of defense.
What You Get
- A Slack sub-agent that monitors a designated channel for data requests
- Automatic search of existing dashboards, saved queries, and past conversations for relevant answers
- Read-only SQL generation and execution for new but straightforward requests
- Formatted results delivered directly to the requester
- Escalation to the human analyst only when the request requires judgment
- A log of all requests, responses, and time saved
Setup Steps
- Ask your Claw to create an "Ad Hoc Deflector" skill with the prompt below
- Configure a Slack channel for data requests (e.g., #data-requests)
- Connect it to your data warehouse for query execution
- Point it at your dashboard catalog and saved query directory
- Set escalation rules (what triggers human involvement)
Tips
- The time savings log is powerful evidence for building self-serve infrastructure
- Recurring questions that come up weekly should be flagged for dashboard creation
- Start with read-only access — it should never modify production data
- The Slack integration means stakeholders get faster answers, not worse ones