.png?fit=max&auto=format&n=Frm2GFbmok4D-yJA&q=85&s=93c3ebd47542af65d1cd06d8563a7f6e)
Capabilities
Creative performance analysis
Creative performance analysis
Pull top-performing creatives ranked by spend, ROAS, CPA, or CTR. Group by creative asset, ad name, headline, copy text, or landing page. Results render as visual cards in the chat with images and key metrics.Tools used:
get_creative_performance, get_creative_insightsA/B test comparison
A/B test comparison
Compare creative variants side by side. Filter to specific ad names or groups and see how they stack up across all metrics. Useful for evaluating test results and deciding what to scale.Tools used:
analyze_creative_ab_testVisual creative analysis
Visual creative analysis
When you share an image of a creative, Space Bob analyzes it visually:
- Visual hook — what grabs attention in the first 1-3 seconds
- Emotional appeal — what emotion the creative triggers
- CTA clarity — whether the call-to-action is clear and compelling
- Audience signals — who the creative targets and how
- Product presentation — how the product/service is showcased
- Design elements — color psychology, typography, composition
Ad copywriting
Ad copywriting
Generate high-performing ad copy using the Cassidy AI writing assistant. Provide an ad name, performance context, and desired tone — Space Bob sends it to Cassidy and returns polished copy with references.Tools used:
write_ad_copyRequires Cassidy AI configuration (assistant ID and API key) in ChatBot Settings.
Creative reports
Creative reports
Save, list, and reload creative analyses as named reports. All filters, date ranges, and groupings are preserved. Loading a saved report re-runs the analysis with the exact saved parameters.Tools used:
list_creative_reports, load_creative_report, save_creative_reportCreative Studio
Creative Studio
After pulling creative performance data, Space Bob includes an Open in Creative Studio link. This navigates to the full Creative Reporting page with filters pre-applied and auto-opens the Creative Studio dialog for:
- Exporting cards as PNG, JPEG, or animated GIF
- Customizing grid layout, colors, and typography
- Selecting which metrics to highlight on each card
- Adjusting scale and background for presentations
creative_studio_link (included in all creative tool responses)Guided setup flow
Guided setup flow
When you ask Space Bob to analyze creatives, it walks you through setup with clickable buttons:
- Which platform? — Facebook Ads, Google Ads, TikTok Ads
- What time period? — 7, 14, 30, or 90 days
- How to group? — Ad Name, Creative, Headline, Copy, Landing Page
- What to focus on? — ROAS, Spend, CPA, CTR
- Confirm and run
gather_creative_paramsVisualizations
Visualizations
Generate charts from creative data — compare spend across creatives, visualize ROAS distribution, etc.Tools used:
create_visualizationDeep-dive creative analysis (v2)
Deep-dive creative analysis (v2)
Backed by the Creative Analytics pipeline, Bob has 8 deep-dive tools that operate on the persisted analysis layer — vision tags, embeddings, exhaustion snapshots, named concept clusters, and the daily Creative Brief.Use these when the user asks “what should I make next?”, “is my creative tired?”, “find ads like this one”, or “what concepts am I testing?”Tools used:
find_similar_creatives, search_creatives_semantic, analyze_hook_exhaustion, creative_diversity_score, creative_concept_clusters, compare_creatives, top_creatives_by_archetype, recommend_next_creative_briefCreative deep-dive tools
These 8 tools sit on top of the Creative Analytics pipeline — every ingested creative is auto-analyzed (multi-frame vision + Whisper transcript), tagged across an 8-category taxonomy, embedded into pgvector, and rolled into hook-exhaustion snapshots and named concept clusters.| Tool | Description |
|---|---|
find_similar_creatives(creative_id, kind?) | Cosine kNN over CreativeEmbedding. kind='analysis_text' (default) for concept similarity, kind='image' for visual twins / near-duplicates. |
search_creatives_semantic(query, kind?) | Free-text search. Same kind toggle as above. |
analyze_hook_exhaustion(client_id?) | Latest HookExhaustionSnapshot per archetype — spend share, ROAS slope, net-new ratio, exhaustion score, recommended alternatives. See Hook Exhaustion. |
creative_diversity_score(client_id?, since_days?) | 0–1 score of how diverse the live creatives are (entropy over hook archetypes). Low score → library is repeating itself. |
creative_concept_clusters(client_id?) | Named clusters from the weekly k-means + GPT naming pass. Returns name, summary, member count, 30-day spend + ROAS. See Concept Clusters. |
compare_creatives(creative_id_a, creative_id_b) | Side-by-side: shared tags, distinct tags, cosine similarity score. |
top_creatives_by_archetype(archetype_slug, client_id?, limit?) | Top creatives in an archetype by 60-day ROAS. |
recommend_next_creative_brief(client_id?, force_regenerate?) | Today’s persisted Creative Brief — recommended archetype / format / audience / messaging theme + hook concept + evidence-backed rationale + risks + inspirations. The same brief shown on the analytics page and posted to Slack. |
When to call which
User asks "what should I make next?"
Call
recommend_next_creative_brief first — that’s the canonical answer. It already aggregates exhaustion + diversity + top performers into a structured recommendation.Optionally enrich with analyze_hook_exhaustion + creative_concept_clusters if the user wants the supporting context.User asks "is my creative tired?"
Call
analyze_hook_exhaustion and surface any is_exhausted=True rows with their recommended_alternatives. If everything is healthy, call creative_diversity_score — concentration without exhaustion is still a signal.User shares a creative_id and asks for context
Call
find_similar_creatives with kind='analysis_text' (concept similarity). Optionally also kind='image' if the user wants to find visual twins specifically.User asks for free-text search
Call
search_creatives_semantic with the user’s phrase. Default kind='analysis_text' for “ads about X”; switch to kind='image' for “ads that look like Y”.Tenant scoping
Every tool reads fromRunnableConfig.client_id (propagated from the JWT) and resolves accessible clients via shared.auth.get_accessible_client_ids. Cross-tenant queries are impossible — the source-creative lookup, the kNN query, and any aggregation all filter on client_id__in=accessible.
Tool reference
Creative analysis tools
| Tool | Description |
|---|---|
get_creative_performance | Top-performing creatives ranked by performance metrics. Returns visual cards with images. |
analyze_creative_ab_test | Compare creative variants side by side with full metric breakdown |
get_creative_insights | Strategic insights — top/bottom performers, best ROAS, best CTR, lowest CPA |
gather_creative_params | Guided setup — returns platform options, date ranges, and grouping choices with buttons |
Creative report tools
| Tool | Description |
|---|---|
list_creative_reports | List all saved creative reports with their configuration |
load_creative_report | Load a saved report and re-run the analysis with all saved filters. Supports date range overrides. |
save_creative_report | Save the current analysis as a named creative report |
Copywriting tools
| Tool | Description |
|---|---|
write_ad_copy | Generate ad copy via Cassidy AI with performance context and tone control |
Visualization tools
| Tool | Description |
|---|---|
create_visualization | Generate charts from creative performance data |
Creative insight rules
Space Bob follows specific rules when analyzing creatives:Connect visuals to metrics
Every visual observation is tied to performance data. “The bold red CTA button likely drives the 2.3% CTR, significantly above the 1.1% average.”
Explain why top performers work
Don’t just rank — explain what creative attributes drive performance.
Suggest A/B tests
Recommend test ideas based on what’s working. “Consider testing a version with a lifestyle image vs. this product-only shot.”
Available metrics
Creative analysis includes these performance metrics:| Metric | Description |
|---|---|
spend | Total ad spend |
revenue | Total revenue attributed |
roas | Return on Ad Spend (revenue / spend) |
cpa | Cost Per Acquisition (spend / purchases) |
ctr | Click-Through Rate (clicks / impressions) |
cpc | Cost Per Click |
cpm | Cost Per 1,000 Impressions |
purchases | Total conversions/purchases |
impressions | Total ad impressions |
clicks | Total clicks |
reach | Unique users reached |
frequency | Average times each user saw the ad |
aov | Average Order Value |
atc | Add-to-Cart events |