Summary
Ad Performance Intelligence is a demonstration Marketing Analytics and BI product that organizes simulated Google Ads and Meta Ads metrics into an executive view of spend, efficiency, funnel, and attributed return.
Project case study
- Demonstration project
- Marketing Analytics and BI
- Simulated data
- Private repository
- Functional application
Ad Performance Intelligence
An analytics dashboard that consolidates a simulated paid-media scenario to compare channels, model metrics, and support budget decisions with explainable rules.
Independent demonstration project with no institutional affiliation or reproduction of confidential client data. Google Ads and Meta Ads are represented dataset channels; there are no active platform integrations.
- Deterministic synthetic data
- Cross-channel analysis
- Calculated metrics
- Rule-based recommendations
- Demonstration simulator
Private repository; audited reference on the main branch, commit 87fc21b.
Private source codeVerifiable technical evidence is documented in this case study.
Overview
Problem
Channel comparison loses clarity when spend, delivery, conversion, and return are read in isolation or with inconsistent denominators.
My role
Analytics scenario modeling, implementation of calculations and rules, web experience development, documentation of assumptions, and validation of technical contracts.
- Overall ROAS
- 18.5x
- Published demo / /api/kpis over 1,800 simulated records / 13 Oct 2026
- Analyzed conversions
- 12,798
- Published demo / /api/kpis over 1,800 simulated records / 13 Oct 2026
- Consolidated CTR
- 3.3%
- Published demo / /api/kpis over 1,800 simulated records / 13 Oct 2026
Data and assumptions
The data is synthetic, deterministic, and defined in versioned TypeScript modules. The published application does not query Google Ads, Meta Ads, a CRM, or a data warehouse.
The public scenario contains ten campaigns and 1,800 daily facts from December 3, 2025 through May 31, 2026. Values are illustrative and do not represent clients, accounts, or business results.
- Origin
- TypeScript modules under src/data
- Classification
- Synthetic and simulated data
- Grain
- Campaign by day
- Coverage
- 10 campaigns · 180 days · 1,800 facts
- Channels
- Google Ads and Meta Ads
- Persistence
- No database, upload, or writes
Data structure
Modeled entities
- Campaign: channel, objective, region, period, budget, and status.
- DailyPerformance: delivery, cost, leads, conversions, and attributed revenue by campaign and day.
- Lead: simulated source, score, quality, intent, and stage.
- Channel, Benchmark, and Recommendation: channel context, demonstration references, and heuristic rules.
Analytics architecture
Five layers reflect the public code and published application without describing the project as an official integration, automated ETL, or data warehouse.
1. Data source
TypeScript seeds define ten campaigns and fixed parameters for Google Ads and Meta Ads.
2. Deterministic loading
Local modules generate 180 days of facts from a fixed start date and reproducible mathematical functions.
3. Modeling
campaignId relates campaigns, daily facts, and leads; channel, objective, and region act as dimensions.
4. Calculations and rules
TypeScript functions calculate rates and costs, while heuristic rules generate recommendations and estimates.
5. Visualization and decisions
Next.js, Recharts, and TanStack Table present KPIs, comparisons, funnel, filters, and simulator.
Implemented metrics
The formulas below were verified in the public code. Divisions use safeDivide and return zero when the denominator is zero.
| Definition | Unit | Calculation rule | Use | Limitation |
|---|---|---|---|---|
| CTRImpressions that resulted in a click. | % | clicks ÷ impressions × 100; a zero denominator returns 0. | Read top-of-funnel response. | Does not measure quality or downstream conversion. |
| CPCAverage cost per click. | BRL per click | spend ÷ clicks; a zero denominator returns 0. | Compare traffic acquisition. | Does not demonstrate traffic value or intent. |
| CPLAverage cost per generated lead. | BRL per lead | spend ÷ leads; a zero denominator returns 0. | Assess lead capture. | A simulated lead does not equal a sale. |
| CPAAverage cost per attributed conversion. | BRL per conversion | spend ÷ conversions; a zero denominator returns 0. | Compare final actions across campaigns. | Demonstration attribution, without incrementality. |
| Conversion rateClicks that resulted in an attributed conversion. | % | conversions ÷ clicks × 100; a zero denominator returns 0. | Relate traffic to final actions. | Does not isolate channel, offer, or sales-service effects. |
| ROASAttributed revenue divided by media spend. | multiplier | attributed revenue ÷ spend; a zero denominator returns 0. | Contextualize attributed return. | Not ROI, total business revenue, or profit. |
| Budget utilizationShare of campaign budget represented by spend. | % | spend ÷ budget × 100; a zero denominator returns 0. | Track simulated budget use. | Fixed budget without platform pacing. |
| Average lead qualityMean lead score in the selected view. | points | sum of lead scores ÷ lead count; an empty collection returns 0. | Add qualitative context to lead volume. | Synthetic assumption, not a trained model. |
| Campaign Efficiency ScoreScore used to rank efficiency signals. | 0 to 100 | Combines normalized CTR, CPC, CPL, conversion rate, ROAS, and lead quality. | Compare campaigns within the same scenario. | Heuristic weights; not a forecast or guarantee. |
CTR
Impressions that resulted in a click.
- Unit
- %
- Calculation rule
- clicks ÷ impressions × 100; a zero denominator returns 0.
- Use
- Read top-of-funnel response.
- Limitation
- Does not measure quality or downstream conversion.
CPC
Average cost per click.
- Unit
- BRL per click
- Calculation rule
- spend ÷ clicks; a zero denominator returns 0.
- Use
- Compare traffic acquisition.
- Limitation
- Does not demonstrate traffic value or intent.
CPL
Average cost per generated lead.
- Unit
- BRL per lead
- Calculation rule
- spend ÷ leads; a zero denominator returns 0.
- Use
- Assess lead capture.
- Limitation
- A simulated lead does not equal a sale.
CPA
Average cost per attributed conversion.
- Unit
- BRL per conversion
- Calculation rule
- spend ÷ conversions; a zero denominator returns 0.
- Use
- Compare final actions across campaigns.
- Limitation
- Demonstration attribution, without incrementality.
Conversion rate
Clicks that resulted in an attributed conversion.
- Unit
- %
- Calculation rule
- conversions ÷ clicks × 100; a zero denominator returns 0.
- Use
- Relate traffic to final actions.
- Limitation
- Does not isolate channel, offer, or sales-service effects.
ROAS
Attributed revenue divided by media spend.
- Unit
- multiplier
- Calculation rule
- attributed revenue ÷ spend; a zero denominator returns 0.
- Use
- Contextualize attributed return.
- Limitation
- Not ROI, total business revenue, or profit.
Budget utilization
Share of campaign budget represented by spend.
- Unit
- %
- Calculation rule
- spend ÷ budget × 100; a zero denominator returns 0.
- Use
- Track simulated budget use.
- Limitation
- Fixed budget without platform pacing.
Average lead quality
Mean lead score in the selected view.
- Unit
- points
- Calculation rule
- sum of lead scores ÷ lead count; an empty collection returns 0.
- Use
- Add qualitative context to lead volume.
- Limitation
- Synthetic assumption, not a trained model.
Campaign Efficiency Score
Score used to rank efficiency signals.
- Unit
- 0 to 100
- Calculation rule
- Combines normalized CTR, CPC, CPL, conversion rate, ROAS, and lead quality.
- Use
- Compare campaigns within the same scenario.
- Limitation
- Heuristic weights; not a forecast or guarantee.
Technical decisions
Deterministic synthetic dataset in TypeScript
- Reason
- Makes the demonstration reproducible without exposing client accounts, campaigns, or information.
- Consequence / trade-off
- The scenario does not capture real-platform seasonality, noise, and data quality.
Centralized formulas with safe division
- Reason
- Keeps CTR, CPC, CPL, CPA, conversion, and ROAS consistent across components and APIs.
- Consequence / trade-off
- Returning zero prevents invalid values but requires context to distinguish a missing base from real performance.
One local source for UI and Route Handlers
- Reason
- Makes KPIs and internal responses verifiable without infrastructure or external credentials.
- Consequence / trade-off
- There is no automatic refresh, persistence, or user isolation.
Additional technical decisions
Explainable heuristic recommendations and simulation
- Reason
- Rules make assumptions, signals, and suggested actions auditable in the code.
- Consequence / trade-off
- Outputs do not demonstrate causality, statistical optimization, or a machine-learning model.
Public evidence
Application verified without sign-in; code paths checked in the private repository, main branch, commit 87fc21b.
Public application
Published application
Functional Portuguese dashboard with KPIs, filters, comparisons, recommendations, and a demonstration simulator.
Open evidencePrivate source code
Repository at commit 87fc21b
Private snapshot used as the reference for this audit.
Public documentation
Simulated-data strategy
ADR records deterministic data, lack of synchronization, and demonstration-only use.
Code referencedocs/adr/002-data-strategy.md
Code reference
Deterministic daily series
Generates 180 dates and campaign facts from fixed parameters.
Code referencesrc/data/daily-performance.ts
Code reference
Metric functions
Implementation of CTR, CPC, CPL, CPA, conversion, ROAS, utilization, and score.
Code referencesrc/lib/analytics/metrics.ts
Implementation references
Public endpoint
KPI endpoint
Route Handler that aggregates the local dataset and identifies the response as simulated.
Code referencesrc/app/api/kpis/route.ts
Code reference
Recommendation Engine
Heuristic rules with metrics, reasons, suggested actions, and trade-offs.
Code referencesrc/lib/analytics/recommendation-engine.ts
Code reference
Budget simulator
Estimates derived from budget, simulated historical efficiency, and lead quality.
Code referencesrc/lib/analytics/simulator.ts
Code reference
Metric tests
Reproducible cases for CTR, CPC, CPL, ROAS, and efficiency score.
Code referencesrc/tests/metrics.test.ts
Public documentation
Architecture document
Separation across presentation, analytics, validation, data, types, and Route Handlers.
Code referencedocs/architecture.md
Published demonstration scenario
- Simulated data
- BRL 398,713.16
- Aggregated spend
- Sum of cost across 1,800 simulated daily facts.
- Simulated data
- 3.33%
- Consolidated CTR
- Clicks divided by impressions across the full scenario.
- Simulated data
- 18.46x
- Demonstration ROAS
- Simulated attributed revenue divided by spend.
All three values belong to the application's simulated dataset. They do not represent a media account, client, business improvement, profit, or enterprise impact.
Values reproduced through the public /api/kpis endpoint and the formulas versioned in the audited reference.
Demonstrated capabilities
- Marketing Analytics and Business Intelligence in a cross-channel scenario.
- Metric modeling and documentation.
- Simulated-data quality, consistency, and traceability.
- Data visualization with Recharts and analytical tables.
- Development and testing with Next.js, React, TypeScript, and Vitest.
Limitations and next steps
Confirmed limitations
- The data is synthetic and contains no real client accounts, campaigns, or results.
- There is no active integration, automatic refresh, database, persistence, upload, or authentication.
- Attributed revenue and ROAS are demonstrative; they do not measure profit, ROI, or incrementality.
- Recommendations, benchmarks, and simulator estimates are heuristic; the application is Portuguese-only.
Possible evolutions
- Integrate official APIs and a persistent historical layer with secure credentials and incremental refresh.
- Model attribution and monitor data quality, reconciliation, and freshness.
- Validate recommendations through experiments before any operational use.
Gallery
Captures of the Portuguese-only published application; all numbers belong to the simulated scenario.
Optional technical details
How to run
- The repository is private; with access, use the main branch at reference commit 87fc21b.
- Run npm install and npm run dev to start the Next.js application.
- Use the Route Handlers or the interface to reproduce metrics over the simulated dataset.
- Run npm run test to validate the existing formulas and rules.
Next project
Real-Time Hand Gesture Recognition
Recognition of 11 hand gestures from the camera, in the browser, with explainable rules and a CPU fallback.
View full case