✍️How to write a good prompt

A short guide on how to write good prompts on Boolee.

Prompt Guide for the Boolee Agent System

Overview

This guide helps you write effective prompts to get the best possible results from the Boolee Agent System. The system can perform various types of data analysis, from simple queries to complex machine learning predictions.

Basic Structure of a Good Prompt

1. Specify Data Sources & Context

Channel: [Google Ads, Facebook, TikTok, etc.]
Level: [Account, Campaign, Ad Group, Ad, Keywords]
Time Period: [last 30 days, this week, January 2024, etc.]

2. Specify Analysis Depth

• Quick: Fast overview and summary

• Medium: Detailed analysis with insights

• In-Depth: Comprehensive analysis with ML methods and predictions

Prompt Categories & Examples

1. Exploratory Analysis (Data Explorer)

Structure: Analyze [Data Source] at [Level] and show me [what you want to see]

Examples:

• Analyze Google Ads campaigns from the last 30 days and show me performance trends

• Examine Facebook Ad Sets in January 2024 and identify the best and worst performers

• Analyze TikTok Ads at account level for Q1 2024 and give me an overview of overall performance

2. Specific Metrics Analysis (Data Analyst)

Structure: Show me [specific metrics] for [Data Source] at [Level] [Time Period]

Examples:

• Show me CTR, CPC and Conversion Rate for Google Ads campaigns from the last 7 days

• Analyze Cost per Acquisition and ROAS for Facebook Ads at ad group level in December 2023

• Compare Impressions and Click-Through-Rates between different TikTok campaigns this week

3. Predictions & Forecasting

Structure: Create a prediction/forecast for [Metric] based on [Data Source]

Examples:

• Make a prediction for Google Ads spend for the next 30 days based on current trends

• Create a forecast for expected conversions from our Facebook campaigns for next week

• Forecast the budget consumption of our TikTok Ads for the next quarter

4. Anomaly Detection

Structure: Search for anomalies/irregularities in [Data Source] [Time Period]

Examples:

• Search for anomalies in Google Ads performance from the last 14 days

• Identify anomalies in Facebook Ad costs and CTR this week

• Detect unusual patterns in TikTok conversion rates from the last 30 days

5. Dashboard & KPI Addition

Structure: Add [specific metrics] as KPI to the dashboard for [Data Source]

Examples:

• Add Cost per Click and Quality Score as KPIs for Google Ads campaigns

• Create a KPI dashboard with ROAS, CPM and Engagement Rate for Facebook Ads

• Add Video View Rate and Cost per View for TikTok campaigns to the dashboard

6. Help & Recommendations

Structure: Give me recommendations/help for [specific problem]

Examples:

• Give me recommendations for optimizing my Google Ads campaigns with low CTRs

• Which metrics should I track for Facebook E-Commerce campaigns?

• Help me understand and improve the performance of my TikTok Ads

Advanced Prompt Options

Time Period Specifications

• Last X days/weeks/months
• Specific date ranges (January 1st to March 31st, 2024)
• Comparison periods (January vs. February, year-over-year)
• Granularity (daily, weekly, monthly)

Visualization Preferences

• Chart type: Line chart, Bar chart, Pie chart, Heatmaps
• Tables with sorting and filtering
• Trend analysis with time series
• Interactive dashboards with drill-down functions

Note: If a desired visualization is not possible or not optimal for the data, Boolee automatically selects the best available alternative.

Machine Learning Methods

• Anomaly detection: Identify unusual patterns in data
• Regression: Understand relationships between metrics

Prompt Quality Checklist

✓ Good prompts contain:

• Clear data source (Google Ads, Facebook, etc.)

• Specific level (Campaign, Ad Group, etc.)

• Defined time period

• Concrete question or goal

• Desired metrics (if relevant)

❌ Avoid:

• Vague formulations ("show me everything")

• Missing time specifications

• Unclear data sources

• Too complex multi-requests in one prompt

Prompt Templates

Template 1: Performance Analysis

Analyze [CHANNEL] [LEVEL] for [TIME PERIOD] and show me:
- [METRIC 1], [METRIC 2], [METRIC 3]
- Trends and anomalies
- Top 5 performers and Bottom 5 performers

Template 2: Comparison Analysis

Compare [CHANNEL] performance between [TIME PERIOD 1] and [TIME PERIOD 2]:
- [SPECIFIC METRICS]
- Identify improvements and deteriorations
- Give recommendations for optimizations

Template 3: Forecast Request

Create a [TIME PERIOD] forecast for [CHANNEL] based on:
- Historical data from the last [TIME PERIOD]
- [SPECIFIC METRICS]
- Consider seasonal trends

Pro Tips for Better Results

1. Be specific: "Google Ads Search campaigns" instead of just "Google Ads"

2. Define success: What is a "good" CTR for your industry?

3. Provide context: Mention special events (Black Friday, product launches)

4. Iterate: Ask follow-up questions based on initial results

5. Use history: Reference previous analyses for context

Typical Workflow Patterns

Pattern 1: Exploratory Analysis → Specific Questions

1. "Analyze Google Ads performance from the last 30 days"
2. "Why did the CTR drop so significantly in week 3?"
3. "Show me the keywords with the worst performance"

Pattern 2: Problem Identification → Solution

1. "Search for anomalies in Facebook Ad costs this week"
2. "What are the causes for the increased CPCs?"
3. "Give me recommendations for cost optimization"

Pattern 3: Dashboard Building

1. "Show me an overview of my current Google Ads KPIs in the dashboard"

---

Tip: The system learns from your requests. The more you use it and provide feedback, the better the results become!