Talking to Your CRM in Natural Language: A Practical Guide

Natural language CRM queries sound like a gimmick until you actually use them. Then they feel obvious. Instead of building filters, choosing fields, and clicking "Apply" to find your leads from SaaStr who haven't heard from you in 30 days, you just ask: "Show me leads from SaaStr I haven't contacted in a month." You get the answer in two seconds.

This guide covers what queries actually work, how DenchClaw translates your English into database queries, common patterns that work reliably, and a few edge cases to watch out for.

What's Actually Happening Under the Hood#

When you ask DenchClaw a question in natural language, here's what happens:

1. The AI interprets your intent and identifies the entities (contacts, deals, companies), filters (date ranges, stages, tags), and operations (list, count, sort, calculate) involved
2. It translates that into a SQL query against your local DuckDB database
3. DuckDB executes the query against your actual CRM data
4. The results come back and the AI formats them into a human-readable response

So "show me deals closing this month" becomes something like:

SELECT name, stage, value, close_date
FROM deals
WHERE close_date BETWEEN current_date AND last_day(current_date)
ORDER BY close_date ASC;

You never see the SQL. You just get the answer. But knowing this translation is happening helps you understand why some phrasings work better than others, and what the limits are.

Common Queries That Work Well#

These types of queries are reliable and work out of the box:

Contact lookups:

• "What's Maria Santos' phone number?"
• "When did I last contact Globex Corp?"
• "Show me all contacts at Acme Inc"
• "Find contacts tagged as 'investor'"
• "Who did I add this week?"

Pipeline queries:

• "Show me all deals in the proposal stage"
• "What deals are closing this month?"
• "What's my total pipeline value?"
• "How many deals do I have open?"
• "Show me deals over $20,000"

Follow-up and activity:

• "Who haven't I contacted in 30 days?"
• "Show me contacts with no follow-up date set"
• "What follow-ups are due this week?"
• "Who did I speak with yesterday?"
• "Show me contacts I've met more than 3 times"

Filtering by source or tag:

• "Show me leads from the SaaStr conference"
• "List contacts tagged 'enterprise'"
• "Show me all contacts in San Francisco"
• "Find leads who came from the website"

Counts and summaries:

• "How many leads do I have from last month?"
• "What's the average deal size in my pipeline?"
• "How many contacts do I have at seed-stage startups?"

Complex Queries: Multi-Filter and Calculations#

Natural language handles complex multi-condition queries surprisingly well. Here are examples that work reliably:

Multi-filter queries:

• "Show me enterprise leads in the negotiation stage that I haven't contacted in 2 weeks"
• "Find contacts in New York who are tagged 'investor' and haven't been updated this month"
• "List deals over $10K that have been in the proposal stage for more than 30 days"

Sorting and ranking:

• "Show me my 5 highest-value open deals"
• "Which companies have I added the most contacts at?"
• "Sort my leads by last contact date, oldest first"

Date-relative queries:

• "Show me deals added in Q1 2026"
• "Which contacts did I meet at events in the last 6 months?"
• "What was the last deal I closed?"

Calculations:

• "What's the total value of deals I closed last quarter?"
• "What's my win rate this year?"
• "How many days on average do my deals sit in the proposal stage?"

These work because DuckDB is a powerful analytical database — it handles aggregations, date math, and multi-condition filtering efficiently. The AI translates your question into the right query and DuckDB does the heavy lifting.

Tips for Phrasing Queries That Work Reliably#

A few patterns that consistently produce good results:

Use the field names you've set up. If you have a field called "Source" (not "Lead Source"), say "source" in your query. "Show me contacts where source is LinkedIn" works better than "show me LinkedIn contacts" if LinkedIn is a value in a Source field.

Be specific about time. "Recently" is ambiguous. "In the last 30 days" or "this month" or "since January" are much more reliable.

Specify the object type when it's ambiguous. "Show me Acme contacts" is clearer than "show me Acme" when you have both a company record and multiple contact records for Acme.

Use relative references. "Deals closing this quarter," "leads added this week," "contacts I haven't touched in a month" — relative date queries are well-supported.

Ask for what you want, not how to get it. "Show me stalled deals" works. You don't need to say "find deals where last_activity_date is more than 14 days ago and stage is not Closed." Just describe what you're looking for.

Follow-Up Questions: Refining Your Results#

One of the most useful features is the ability to ask follow-up questions that refine the previous result:

You: "Show me all open deals"
DenchClaw: [Lists 23 deals]

You: "Now just the ones over $15K"
DenchClaw: [Filters to 8 deals]

You: "Sort those by close date"
DenchClaw: [Reorders the 8 deals]

You: "Who's the contact on the top one?"
DenchClaw: [Shows contact info for the highest deal]

This conversational refinement is one of the biggest advantages over traditional filter UIs. You're not rebuilding the query from scratch each time — you're narrowing in.

What Doesn't Work Well (Yet)#

Being honest about the limitations:

Highly ambiguous queries sometimes need clarification. "Show me my best contacts" is ambiguous — best by what metric? Interaction frequency? Deal value? DenchClaw will ask for clarification rather than guess, but it's slightly more work than a precise query.

Custom field names that are vague. If you have a field called "Notes" and you ask about "notes," DenchClaw might be unsure whether you mean the field or general note content. Descriptive field names help.

Complex nested logic. Something like "show me deals where either the contact hasn't been updated in 30 days OR the deal value has decreased in the last week" is technically possible but may need some rephrasing to get right.

Historical trend analysis that requires comparing time periods: "How has my pipeline grown month over month this year?" works, but might be slower and require a more precise phrasing than simple lookups.

The good news: DenchClaw will tell you when it's unsure and ask for clarification rather than returning wrong results silently.

Voice Queries via Phone#

If you're using DenchClaw via Telegram on your phone, you can use Telegram's voice message feature to query your CRM without typing. Record a voice message, Telegram auto-transcribes it, and DenchClaw processes the transcribed text as a normal query.

This works well for hands-free situations: walking between meetings, driving, or any time typing is inconvenient. The transcription accuracy is good enough for most CRM queries, and you can review the transcribed text before sending if you want to verify it.

Frequently Asked Questions#

Do I need to learn any special syntax to query DenchClaw in natural language? No. That's the whole point. You describe what you want in plain English and DenchClaw figures out the query. The tips in this guide help you get more reliable results, but you don't need to learn a query language or memorize commands.

What database does DenchClaw use, and why does it matter? DenchClaw uses DuckDB, a local embedded analytical database that runs on your machine. This means queries are fast (no network round trips), your data never leaves your computer, and complex analytical queries (aggregations, date math, multi-filter) run efficiently. It's designed for exactly this kind of ad-hoc analysis.

Can I query data I imported from another CRM? Yes. If you've imported contacts, deals, and companies from HubSpot, Salesforce, or another CRM into DenchClaw, the imported data is fully queryable via natural language. The import process maps your old fields to DenchClaw fields, so your existing data structure is preserved.

How does DenchClaw handle typos or informal language in queries? Reasonably well. The AI layer handles common typos and informal phrasing without issues. "Whats the deal with acme" works as well as "What is the status of the Acme Corp deal?" The natural language model is robust to casual input.

Can I save frequently used queries? Yes. You can save a query as a named shortcut: "Call this 'stalled deals' and run it weekly." Then you can just ask for "stalled deals" and get the result without re-describing the query. Saved queries can also be used to power automated reports and alerts.

Ready to try DenchClaw? Install in one command: npx denchclaw. Full setup guide →