Writing Custom Instructions

Custom instructions let you fine-tune how your Pingd agent communicates and prioritizes information. Think of them as standing orders — preferences your agent always follows.

Where to Add Instructions

Organization Level (Admins)

Settings → Agent → Organization Instructions

These apply to every user. Use for company-wide standards:

• "Always reference deal stage names from our sales process: Discovery, Evaluation, Proposal, Negotiation, Closed"
• "Our fiscal year starts February 1st"
• "Use ARR (Annual Recurring Revenue) not MRR when discussing revenue"

User Level (Individual)

Settings → Agent → My Instructions

Personal preferences that override org defaults for your experience:

• "I prefer bullet points over paragraphs"
• "Always include the account owner's name when listing deals"
• "Focus on my West Coast territory accounts"

Writing Effective Instructions

Be Specific

❌ Vague: "Be helpful with my pipeline"

✅ Specific: "When I ask about pipeline, always include: total value, deal count, weighted value, and comparison to same period last quarter"

Use Action Words

❌ Passive: "Numbers are important"

✅ Active: "Always include specific numbers and percentages. Never say 'a few' or 'several' — give exact counts."

Set Priorities

❌ No priority: "Show me everything"

✅ Prioritized: "Focus on deals over $50K first. Only mention smaller deals if specifically asked."

Example Instructions Library

For AEs (Account Executives)

- Always mention deal stage and close date when discussing opportunities
Flag any deal that hasn't had activity in 14+ days
When prepping for meetings, include the last 3 touchpoints with the account
Use our stage names: Qualified → Discovery → Demo → Proposal → Negotiation → Closed Won/Lost
My quota is $2M this quarter — reference attainment percentage when relevant

For SDRs (Sales Development Reps)

- Focus on lead response time and follow-up cadence
When showing contacts, always include their title and last engagement date
Highlight accounts with buying signals (website visits, content downloads, job postings)
My target is 15 meetings booked per month

For Sales Managers

- When reviewing pipeline, break down by rep
Flag coaching opportunities — reps with stalled deals or low activity
Include week-over-week pipeline movement (new, advanced, slipped, lost)
Always show forecast vs. actual for the current quarter
Use my team's territory assignments when filtering data

For VP/Director

- Lead with the executive summary — top-level numbers first, details on request
Compare to prior year performance when discussing metrics
Highlight rep performance outliers (top and bottom)
Include churn risk in pipeline reviews
Focus on enterprise segment ($100K+ ARR)

Connector-Aware Instructions

When you have multiple data sources connected, these instructions help your agent understand your specific data setup:

- Our Snowflake data uses OWNER_EMAIL for account ownership — map this to Salesforce Account.Owner.Email
When showing customer health, use the health_score from our S3 analytics exports
Account ARR data is most accurate in our Snowflake warehouse accounts table
For product usage metrics, query our Redshift events table first
Our MySQL customer database has the authoritative billing information
Cross-reference pipeline data from Salesforce with forecasting data from BigQuery

System-Specific Rules:

- Snowflake queries: Use fiscal_quarter field instead of calendar quarter
S3 data files: The latest data is always in the /current/ prefix
Salesforce data: Custom fields use our naming convention (Company_Name__c format)
Redshift events: User activity data has a 2-hour delay, mention this caveat

Best Practices

1. Start with 3-5 instructions — Don't overwhelm the agent with 50 rules. Start small, add as you go.

1. Review monthly — Your preferences change. Update instructions when they get stale.

1. Be consistent with terminology — If you define terms in Custom Data Definitions, use those same terms in your instructions.

1. Test after changes — Ask a few questions to verify your instructions produce the desired output.

1. Don't contradict org instructions — If the org says "use ARR" and you say "use MRR," your override wins, which may confuse discussions with colleagues.

Pro Tip: Think about what you repeatedly ask your agent to do differently, then bake that into your instructions so you don't have to keep asking.

Instruction Merge Order

When your agent processes a query, instructions are merged:

1. System defaults (base agent behavior)
2. Organization instructions (company-wide)
3. User instructions (personal overrides)

Later layers override earlier ones for any conflicting directives.

See also: Configuring Your Agent · Custom Data Definitions