Sprint Engine / beseen.studio

You are writing the final deliverable for a digital marketing analytics sprint. SPRINT BRIEF: - Objective: subsun want to get into fun ai creative, look at some of the videos their competitors have in the database which are similar vibe, and make some examples for subsun of where to start - Client: subsun - Context: None ANALYSIS RESULTS FROM ALL COMPUTATION STEPS: === Fetch all competitor ads with video creative (data_fetch) === Output: Fetched 1000 rows from competitor_ads Variables: { "row_count": 1000 } DataFrame 'competitor_ads': 1000 rows Columns: ['ad_archive_id', 'page_name', 'primary_text', 'headline', 'description', 'creative_type', 'ad_delivery_start_time', 'ad_delivery_stop_time', 'is_currently_active', 'eu_total_reach', 'preview_url', 'cta_type'] Sample: [{"ad_archive_id": "1422872115262952", "page_name": "Meller", "primary_text": null, "headline": null, "description": null, "creative_type": "video", "ad_delivery_start_time": "2025-01-24T00:00:00+00:00", "ad_delivery_stop_time": null, "is_currently_active": false, "eu_total_reach": 28256, "preview_url": null, "cta_type": null}, {"ad_archive_id": "891028349624167", "page_name": "Meller", "primary_text": null, "headline": null, "description": null, "creative_type": "video", "ad_delivery_start_time === Fetch competitor performance data (data_fetch) === Output: Fetched 1000 rows from competitor_ad_daily Variables: { "row_count": 1000 } DataFrame 'competitor_ad_daily': 1000 rows Columns: ['ad_archive_id', 'check_date', 'is_active', 'eu_total_reach'] Sample: [{"ad_archive_id": "834603209249733", "check_date": "2026-04-02", "is_active": false, "eu_total_reach": null}, {"ad_archive_id": "1859671824970514", "check_date": "2026-04-02", "is_active": false, "eu_total_reach": null}, {"ad_archive_id": "1196119602563212", "check_date": "2026-04-02", "is_active": false, "eu_total_reach": null}] === Fetch competitor page info (data_fetch) === Output: Fetched 25 rows from competitor_pages Variables: { "row_count": 25 } DataFrame 'competitor_pages': 25 rows Columns: ['page_id', 'page_name', 'category', 'country_code'] Sample: [{"page_id": "442604319128955", "page_name": "Ace & Tate", "category": "eyewear", "country_code": "GB"}, {"page_id": "1415151682063242", "page_name": "Meller", "category": "eyewear", "country_code": "GB"}, {"page_id": "126989427319274", "page_name": "Le Specs", "category": "eyewear", "country_code": "GB"}] === List all competitor video files (storage_list) === Output: Found 3194 files (3194 videos, 0 images) Variables: { "file_count": 3194, "video_count": 3194, "image_count": 0 } DataFrame 'competitor_media': 3194 rows Columns: ['filename', 'folder', 'mimetype', 'size_bytes', 'is_video', 'is_image', 'ad_archive_id', 'public_url', 'created_at'] Sample: [{"filename": "1022133007184519.mp4", "folder": "032c", "mimetype": "video/mp4", "size_bytes": 64650, "is_video": true, "is_image": false, "ad_archive_id": "1022133007184519", "public_url": "https://tavthvmkyupgxgpccikn.supabase.co/storage/v1/object/public/competitor-previews/032c/1022133007184519.mp4", "created_at": "2026-03-24T10:01:45.355Z"}, {"filename": "1052507523633977.mp4", "folder": "032c", "mimetype": "video/mp4", "size_bytes": 72861, "is_video": true, "is_image": false, "ad_archive_id === Merge video ads with actual video files and performance data (python_compute) === Output: Total video ads in database: 1000 Video ads with actual video files: 0 Video ads with performance data: 0 Top competitors by video count: Series([], ) DataFrame 'perf_metrics': 200 rows Columns: ['ad_archive_id', 'max_reach', 'avg_reach', 'days_tracked', 'first_seen', 'last_seen', 'active_days'] Sample: [{"ad_archive_id": "1007034614839515", "max_reach": NaN, "avg_reach": NaN, "days_tracked": 1, "first_seen": "2026-04-02", "last_seen": "2026-04-02", "active_days": 0}, {"ad_archive_id": "1015548884692086", "max_reach": NaN, "avg_reach": NaN, "days_tracked": 1, "first_seen": "2026-04-02", "last_seen": "2026-04-02", "active_days": 0}, {"ad_archive_id": "1016630034373903", "max_reach": NaN, "avg_reach": NaN, "days_tracked": 1, "first_seen": "2026-04-02", "last_seen": "2026-04-02", "active_days": 0} === Identify AI and fun creative indicators in copy (python_compute) === Output: Error in copy analysis: name 'enhanced_video_data' is not defined Error: NameError: name 'enhanced_video_data' is not defined Traceback (most recent call last): File "<sprint_sandbox>", line 5, in <module> NameError: name 'enhanced_video_data' is not defined During handling of the above exception, another exception occurred: Traceback (most recent call last): File "/app/app/sandbox.py", line 198, in execute_code exec(compile(code, "<sprint_sandbox>", "exec"), namespace) File "<sprint_sandbox>", line 67, in <module> NameError: name 'enhanced_video_data' is === Rank top performing creative videos for manual review (python_compute) === Output: Error in video ranking: name 'enhanced_video_data' is not defined Variables: { "review_urls": [] } === Creative score distribution by competitor (chart) === Charts generated: ['/app/artifacts/sprint_12/creative_scores_by_competitor.png'] === Analyze copy patterns and messaging themes (python_compute) === Output: Error in copy analysis: name 'priority_videos' is not defined Variables: { "copy_patterns": {} } Charts generated: ['/app/artifacts/sprint_12/creative_scores_by_competitor.png'] === Create competitor benchmark matrix (python_compute) === Output: Error creating benchmark matrix: name 'enhanced_video_data' is not defined Charts generated: ['/app/artifacts/sprint_12/creative_scores_by_competitor.png'] === Performance vs Creative Approach Matrix (chart) === Charts generated: ['/app/artifacts/sprint_12/performance_creative_matrix.png', '/app/artifacts/sprint_12/creative_scores_by_competitor.png'] === Generate specific creative briefs for SubSun (python_compute) === Output: Error generating creative briefs: name 'priority_videos' is not defined Variables: { "creative_briefs": [], "subsun_strategy": {} } Charts generated: ['/app/artifacts/sprint_12/creative_scores_by_competitor.png', '/app/artifacts/sprint_12/performance_creative_matrix.png'] === Generate comprehensive competitive intelligence report (llm_synthesis) === Output: ### Executive Summary - Key Findings and Opportunities **Key Findings:** - **Current State:** SubSun currently has no video ads with actual video files or performance data in its database, leading to a significant gap in leveraging AI-driven creative strategies. - **Competitor Analysis:** Top competitors do not have any video ads either, indicating that the market is ripe for innovative and engaging content. However, leveraging AI can set SubSun apart by providing dynamic, personalized, and high-performing videos. **Opportunities:** - **Innovative Content Strategy:** SubSun has a unique opportunity to lead in creating AI-driven creative video content which can significantly enhance brand visibility. - **Data-Driven Insights:** By integrating AI, SubSun can gain valuable insights into consumer behavior and preferences, enabling more targeted and effective ad campaigns. ### Competitive Landscape - Who's Doing AI Creative and How They're Performing **Current State of Competitors:** - **Top Competitors Analysis (Example):** - **Competitor A:** Has not launched any video ads with actual content yet. - **Competitor B:** Similarly, has no video ads in their database. **Performing Metrics (Hypothetical Examples):** - **Engagement Rates:** Both competitors have low engagement rates due to the absence of dynamic and personalized video content. - **Conversion Rates:** Limited data, but expected to be below industry averages due to lack of compelling visual content. ### Success Patterns - What Makes Creative Videos Perform Well **Key Elements for High-Performing Ads:** 1. **Personalization:** AI-driven videos that adapt to individual viewer preferences show significantly higher engagement. 2. **Dynamic Content:** Use of dynamic elements such as real-time weather updates, location-based services, or interactive features can increase interaction rates. 3. **Storytelling:** Effective storytelling with emotional appeal and clear brand messaging leads to higher conversion rates. ### Strategic Recommendations - Specific Actions for SubSun 1. **Invest in AI Technology:** - Partner with leading AI firms to develop a robust video creation tool that can generate personalized videos based on user data. 2. **Develop Initial Creative Concepts:** - Create three detailed and diverse concepts as outlined below. 3. **Run A/B Testing:** - Use the initial creative concepts for A/B testing across different demographics to refine the content. ### Creative Briefs - 3 Detailed Concepts SubSun Should Test 1. **Dynamic Weather Ad Concept:** - **Concept:** A video that adapts based on real-time weather conditions, showcasing sunglasses in sunny, rainy, and overcast scenarios. - **Target Audience:** Urban professionals who are frequently outdoors. - **Engagement Strategy:** Use voiceovers explaining how SubSun sunglasses enhance outdoor activities regardless of the weather. 2. **Location-Based Experience Ad Concept:** - **Concept:** A vide Variables: { "synthesis": "### Executive Summary - Key Findings and Opportunities\n\n**Key Findings:**\n- **Current State:** SubSun currently has no video ads with actual video files or performance data in its database, leading to a significant gap in leveraging AI-driven creative strategies.\n- **Competitor Analysis:** Top competitors do not have any video ads either, indicating that the market is ripe for innovative and engaging content. However, leveraging AI can set SubSun apart by providing dynamic, personalized, and high-performing videos.\n\n**Opportunities:**\n- **Innovative Content Strategy:** SubSun has a unique opportunity to lead in creating AI-driven creative video content which can significantly enhance brand visibility.\n- **Data-Driven Insights:** By integrating AI, SubSun can gain valuable insights into consumer behavior and preferences, enabling more targeted and effective ad campaigns.\n\n### Competitive Landscape - Who's Doing AI Creative and How They're Performing\n\n**Current State of Competitors:**\n- **Top Competitors Analysis (Example):** \n - **Competitor A:** Has not launched any video ads with actual content yet.\n - **Competitor B:** Similarly, has no video ads in their database.\n\n**Performing Metrics (Hypothetical Examples):**\n- **Engagement Rates:** Both competitors have low engagement rates due to the absence of dynamic and personalized video content.\n- **Conversion Rates:** Limited data, but expected to be below industry averages due to lack of compelling visual content.\n\n### Success Patterns - What Makes Creative Videos Perform Well\n\n**Key Elements for High-Performing Ads:**\n1. **Personalization:** AI-driven videos that adapt to individual viewer preferences show significantly higher engagement.\n2. **Dynamic Content:** Use of dynamic elements such as real-time weather updates, location-based services, or interactive features can increase interaction rates.\n3. **Storytelling:** Effective storytelling with emotional appeal and cle CHARTS GENERATED: 7 visualisations were created and will be embedded in the report. Write a comprehensive analytical report. Structure: 1. **EXECUTIVE SUMMARY** — 3-4 sentences with the most important findings. Be specific. 2. **KEY FINDINGS** — Numbered list. Each must cite specific numbers/data from the analysis. 3. **DETAILED ANALYSIS** — Walk through the methodology and what each analysis revealed. Reference specific competitor names, ad types, trends, statistical results. 4. **RECOMMENDED ACTIONS** — Specific, prioritised actions based on findings. Each should explain the expected impact. 5. **METHODOLOGY & CONFIDENCE** — What we did, what worked, what data gaps exist. Rules: - Every claim must be backed by data from the analysis results above. - Never invent data that isn't in the results. - Be direct and actionable — this is for a working marketing team, not a textbook. - If something couldn't be computed due to data gaps, say so honestly. - Use markdown formatting for readability.

Model: qwen2.5:7b Prompt: You are a senior creative strategist writing a comprehensive competitive intelligence report for SubSun sunglasses brand about AI creative video opportunities. Analysis Results: Total video ads in database: 1000 Video ads with actual video files: 0 Video ads with performance data: 0 Top competitors by video count: Series([], ) Error in copy analysis: name 'enhanced_video_data' is not defined Error in video ranking: name 'enhanced_video_data' is not defined Error in copy analysis: name 'priority_videos' is not defined Error creating benchmark matrix: name 'enhanced_video_data' is not defined Error generating creative briefs: name 'priority_videos' is not defined Write a professional report with these sections: 1. EXECUTIVE SUMMARY - Key findings and opportunities 2. COMPETITIVE LANDSCAPE - Who's doing AI creative and how they're performing 3. SUCCESS PATTERNS - What makes creative videos perform well 4. STRATEGIC RECOMMENDATIONS - Specific actions for SubSun 5. CREATIVE BRIEFS - 3 detailed concepts SubSun should test 6. IMPLEMENTATION ROADMAP - Timeline and priorities Focus on actionable insights. Reference specific competitors and performance data where mentioned in the analysis. Keep the tone professional but engaging.

You are the research director for a digital marketing analytics agency (beseen.studio). You must plan a rigorous, multi-step analytical sprint. This is serious analytical work — the output should be something a human analyst would spend a full day producing. SPRINT BRIEF: - Objective: subsun want to get into fun ai creative, look at some of the videos their competitors have in the database which are similar vibe, and make some examples for subsun of where to start - Client: SubSun - Client background: SubSun is a sunglasses brand. DTC ecommerce via Shopify. Runs Meta (Facebook/Instagram) ads. Key metrics: ROAS, CPA, AOV. - Vertical: fashion / sunglasses / DTC ecommerce - Context: None APPROVED SCOPE (user reviewed and approved this): - Deliverable: A competitive creative intelligence report identifying top-performing 'fun AI creative' video ads from sunglasses/fashion competitors, with curated examples, performance benchmarks, and specific creative direction recommendations for SubSun's entry into AI-generated video content - Methodology: 1. Filter competitor_ads for video creatives with AI/fun/creative indicators in copy 2. Cross-reference with competitor_ad_daily to identify high-performing ads (reach, longevity metrics) 3. Generate viewing URLs for top-performing video examples 4. Analyze copy patterns and messaging themes from high-performers 5. Create competitor benchmark matrix showing creative approach vs performance 6. Develop specific creative briefs for SubSun based on successful patterns - Key analyses: ["Video ad identification and performance ranking by reach and active duration", "Copy sentiment analysis focusing on 'fun', 'creative', 'AI' themed messaging", "Creative format distribution across sunglasses/fashion competitors", "Performance correlation between creative approach and engagement metrics", "Competitor creative positioning matrix (fun vs serious, AI vs traditional)", "Headline and CTA pattern analysis for top video performers"] - Limitations noted: ["Cannot analyze actual video content or visual style - relies on metadata and copy only", "Performance metrics limited to reach data, no conversion or ROAS data available", "AI creative identification based on copy analysis rather than visual AI detection", "No direct SubSun historical performance data for benchmarking", "Cannot determine actual AI usage - only messaging that suggests AI creative approach"] You MUST follow this scope. The user approved this deliverable — deliver exactly this. DATA AUDIT RESULTS: - Available sources: 4 confirmed accessible • competitor_ads: 22957 rows • competitor_ad_daily: 283338 rows • competitor-previews storage: N/A rows • competitor_pages: 25 rows - All requested sources available ✓ USER FEEDBACK ON SCOPE: u gotta analyse the actual video files to see which ones are clearly ai made a prime example would be the floyd f1 pitstop POV suitcase wheel one Adjust your plan to address this feedback. AVAILABLE DATA (with schemas): { "competitor_ads": { "table": "competitor_ads", "columns": [ "id", "ad_archive_id", "competitor_page_id", "page_name", "primary_text", "headline", "description", "creative_type", "platform_list", "ad_delivery_start_time", "ad_delivery_stop_time", "first_seen_date", "last_seen_date", "is_currently_active", "languages", "created_at", "updated_at", "eu_total_reach", "age_country_gender_reach", "target_ages", "target_gender", "target_locations", "ad_snapshot_url", "preview_url", "cta_type", "publisher_platforms" ], "row_count": 22957 }, "competitor_ad_daily": { "table": "competitor_ad_daily", "columns": [ "id", "ad_archive_id", "check_date", "is_active", "created_at", "eu_total_reach", "age_country_gender_reach" ], "row_count": 283338 }, "competitor_pages": { "table": "competitor_pages", "columns": [ "id", "page_id", "page_name", "category", "is_political", "country_code", "notes", "created_at", "updated_at" ], "row_count": 25 }, "image_classifications": { "table": "image_classifications", "columns": [ "filename", "shot_type", "excluded", "created_at", "updated_at" ], "row_count": 517 } } DATA CONTEXT — what each table/source actually contains: TABLES: - competitor_ads: Meta Ad Library scrape. Each row = one competitor ad with page_name, headline, primary_text (copy), description, creative_type (static/carousel/image/video), CTA type, start/stop dates, active status, ad_archive_id, ad_snapshot_url, preview_url, eu_total_reach, target_ages, target_gender, target_locations. ~1000 rows. - competitor_ad_daily: Daily check of each ad's status and EU reach estimate. Tracks ad longevity and reach over time. - competitor_pages: List of competitor Facebook pages we track (page_name, category, country_code). ~25 pages in eyewear/fashion. - competitor_copy_analysis: Pre-analysed copy patterns from competitor ads (may be empty). - image_classifications: Classifications of the CLIENT's own product photography (shot_type: Product Angle, Packaging, Product Hero, Model Editorial, etc). This is SubSun's own image library, NOT competitors. ~517 rows. - creative_analysis_feedback: Human feedback on creative analysis quality (may be empty). MEDIA STORAGE (Supabase Storage bucket: "competitor-previews"): - ~3,400 actual video files (.mp4) and ~6,400 images (.jpg) of competitor ad creative. - Organised by competitor folder (lowercase page name). Files named by ad_archive_id. - Joins to competitor_ads table via ad_archive_id column. - Accessible via public URL: {supabase_url}/storage/v1/object/public/competitor-previews/{folder}/{ad_archive_id}.mp4 (or .jpg) - Top competitors by video count: bound (658), the kooples (349), ace (297), ace & tate (266), chimi (206), le specs (194), mutimer (172), jmm (155), toast (154), vuarnet (124) - The sprint engine can list files in these folders and generate public URLs for viewing/download. - IMPORTANT: We cannot run computer vision on videos in the pipeline, but we CAN: list what exists, correlate file counts with ad metadata, generate viewing links, analyse ad metadata of ads that HAVE video creative, and compare patterns between video vs static ads. WHAT YOU CAN DEEPLY ANALYSE: - Copy patterns (headlines, primary text, descriptions, CTAs, messaging themes, tone) - Creative format strategy (who uses video vs static vs carousel, and when) - Ad longevity (which ads run longest, which get pulled quickly) - Timing patterns (launch dates, seasonal trends, campaign cadence) - Competitive share of voice (ad volume by competitor over time) - Targeting patterns (ages, genders, locations where available) - Reach estimates (EU reach from ad_daily data) - Performance proxies (longevity × reach as engagement proxy) - Cross-reference metadata with actual stored creative files WHAT YOU CANNOT DO IN THE PIPELINE: - Computer vision / video analysis (no frame extraction or visual classification) - Access video/image pixel content programmatically - But you CAN reference which files exist and generate viewable URLs for the team to review. PRIOR KNOWLEDGE FROM PREVIOUS SPRINTS: - [analysis] # Video Creative Investment Analysis: Eyewear Category Competitive Intelligence ## EXECUTIVE SUMMARY The analysis of 3,194 video files across 18 eyewear competitors reveals a significant opportunity for SubSun to differentiate through strategic video investment. **Monphell leads the category with - [analysis] # Fun AI Creative Video Strategy for SubSun: Competitor Analysis Report ## EXECUTIVE SUMMARY Analysis of 3,194 video files from 25 competitor brands reveals significant opportunities for SubSun to enter the fun AI creative video space. Top eyewear competitors like mutimer.co (58.5% video adoption) - [analysis] # Fun AI Creative Video Analysis for SubSun *Digital Marketing Analytics Sprint - Final Report* ## EXECUTIVE SUMMARY Analysis of 3,194 competitor video ads from 25 eyewear brands reveals that **Sun Buddies leads in fun AI creative content** with 7 high-scoring videos averaging 2.57 points, followe - [analysis] # Strategic Video Advertising Analysis for SubSun ## Competitor Creative Intelligence & AI Implementation Roadmap --- ## 1. EXECUTIVE SUMMARY The eyewear market shows surprisingly low video adoption with only **1.6% of competitor ads using video format** (16 out of 1,000 total ads analyzed). **Me - [analysis] ### EXECUTIVE SUMMARY SubSun can enhance its marketing strategy by adopting creative elements seen in successful campaigns of competitors like Moscot, Jimmy Fairly, and Etudes Studio. By integrating these strategies, SubSun can increase engagement and align more closely with current trends. ### KE PYTHON LIBRARIES AVAILABLE: pandas, numpy, scipy, sklearn, statsmodels, xgboost, lightgbm, matplotlib, seaborn, plotly, collections, re, datetime, math Your job: create a structured execution plan. Each step must be one of these types: 1. **data_fetch** — Fetch specific data from a database table. Parameters: table (str), columns (list), filters (dict), limit (int) The variable name in code will be the table name (e.g. "competitor_ads" → DataFrame called `competitor_ads`) 1b. **storage_list** — List actual media files (videos/images) from the competitor-previews storage bucket. Parameters: all_competitors (bool — list ALL competitors), folder (str — specific competitor folder e.g. "chimi/"), video_only (bool) Returns a DataFrame called `competitor_media` (or `media_{folder}`) with columns: filename, folder, mimetype, size_bytes, is_video, is_image, ad_archive_id, public_url, created_at. The ad_archive_id column joins to competitor_ads table for cross-referencing metadata with actual creative files. USE THIS to discover what actual video/image creative exists. This is the real data — ~3,400 videos and ~6,400 images. 2. **python_compute** — Run Python code for analysis/transformation. Parameters: code (str — valid Python), description (str) The code has access to: DataFrames from ALL previous steps (by table name), pd, np, json, re, datetime, collections, Counter, defaultdict, math, statistics. Use `save_chart(fig, "filename.png")` to save matplotlib/plotly figures. CHART_DIR variable has the save path. Use `print()` for key findings — stdout is captured and passed to the final synthesis. Assign results to named variables — they'll be captured automatically. IMPORTANT: Write COMPLETE, WORKING Python code. Import what you need. Don't use `...` or pseudocode. CRITICAL CODE QUALITY RULES: - Write DEFENSIVE code. Previous steps may have failed. Always wrap main logic in try/except and print fallback info. - When doing groupby().agg(), ONLY aggregate numeric columns. Use `numeric_only=True` or select specific numeric columns first. NEVER agg on object/string columns with min/max/mean. - When merging DataFrames, always check both exist and are non-empty first. - Convert date columns with `pd.to_datetime(col, errors='coerce')` before date operations. - When accessing a column that might not exist, use `df.get('col')` or check `if 'col' in df.columns`. - Assign key results to clearly named variables (e.g. `top_competitors`, `video_summary`) so later steps can use them. - Always `print()` key findings — this is what feeds the final synthesis. 3. **llm_synthesis** — Ask a LOCAL LLM (Ollama) to interpret/narrate computed results. Parameters: prompt_template (str with {{step_id.stdout}} or {{step_id.variables.name}} placeholders for prior step outputs) NOTE: This runs on a local 7B model — keep prompts focused and provide structured data to interpret. Do NOT specify a model parameter — the system uses the default local model. 4. **chart** — Generate a specific visualization. Parameters: code (str — Python code using matplotlib/plotly), description (str) matplotlib and seaborn are pre-imported. Dark theme is pre-configured (bg=#0a0a0a). You MUST call `save_chart(fig, "filename.png")` at the end or the chart won't be saved. Use `plt.close()` after saving to free memory. Write defensive code — check that data exists before plotting. If prior steps failed, print a message instead of crashing. PLANNING RULES: - This sprint should have **10-20 steps** and represent 15-45 minutes of compute time. - Start by fetching ALL relevant data tables — cast a wide net. - Then do MULTIPLE rounds of analysis: aggregation → statistical testing → pattern detection → segmentation → trend analysis. - Each python_compute step should do ONE focused computation. Do not cram everything into one step. - Include 1-2 llm_synthesis steps max (they take ~60s each on local models). Place them strategically to interpret key intermediate findings. - Include 2-4 chart steps throughout (not all at the end). - Build analysis progressively: basic stats → cross-tabulation → statistical tests → clustering/segmentation → temporal patterns → synthesis. - Make the analysis genuinely useful — not a summary a human could do in a spreadsheet. - Be specific: reference exact column names from the schemas above. - If the brief asks about something the data can't directly answer, plan the CLOSEST useful analysis and note the limitation. - The FINAL step must be an llm_synthesis that creates a comprehensive written deliverable from all prior findings. Return ONLY valid JSON: { "plan_name": "Short descriptive name", "estimated_duration_minutes": 30, "methodology_notes": "Brief explanation of analytical approach and any data limitations", "steps": [ { "id": "step_1", "type": "data_fetch", "name": "Fetch all competitor ads", "depends_on": [], "params": { "table": "competitor_ads", "columns": ["page_name", "headline", "primary_text", "creative_type", "cta_type", "ad_delivery_start_time", "ad_delivery_stop_time", "is_currently_active"], "filters": {}, "limit": 5000 } }, { "id": "step_2", "type": "python_compute", "name": "Compute creative format distribution by competitor", "depends_on": ["step_1"], "params": { "description": "Group ads by page_name and creative_type, compute counts and percentages", "code": "import pandas as pd\ncounts = competitor_ads.groupby(['page_name', 'creative_type']).size().reset_index(name='count')\ntotal_by_page = competitor_ads.groupby('page_name').size().reset_index(name='total')\ncounts = counts.merge(total_by_page, on='page_name')\ncounts['pct'] = (counts['count'] / counts['total'] * 100).round(1)\nprint(counts.to_string())" } } ] }