Hi, I'm Riccardo. I'm building this portfolio around a simple question: how can marketing and customer data help a team make a better decision? I enjoy working across campaigns, CRM and customer experience, and I try to make every project easy to review—not just technically correct. I'm looking for an internship or junior role where I can keep learning by working on real business questions.
| If you are hiring for | Open first | The business question |
|---|---|---|
| CRM / Marketing Automation · Lifecycle & Email | Lifecycle Automation Lab | Can an analysis model actually drive an ESP—segment, flow, send—without ever emailing someone who did not consent? |
| Performance / Digital Marketing · E-commerce | Digital Campaign Performance Dashboard | Which campaigns, audiences and customer segments deserve attention—and how should that change budget and CRM actions? |
| Customer Experience / Customer Insights | E-commerce CX Analytics | What is most closely associated with a poor Olist customer experience, and where should a CX team act first? |
| AI governance / responsible AI | Agentic Audit Reporting · live report | How can an AI-assisted workflow remain traceable, reviewable and subject to human approval? |
I wanted a portfolio that shows the work, not just a list of tools. I use public data so another person can inspect the assumptions, rerun the code and challenge the result. Where realistic public data does not exist, I label the simulated parts instead of presenting them as client or production data.
Three of these projects analyse. One of them operates: the lifecycle work runs on a real Klaviyo account connected to a real Shopify store, because knowing what a flow should do and getting an ESP to actually do it are different skills.
The projects are not meant to pretend I have already seen every real company setup. They are how I practise turning an open-ended question into a useful report, while being honest about what the data can and cannot prove.
Consent-first lifecycle automation, built and then actually run: three Klaviyo flows—welcome with double opt-in, abandoned cart, win-back—on a Shopify store, plus the Python bridge that turns RFM/lifecycle output into the segment property the win-back triggers on. All three have been entered by real events and have delivered. One subscriber and three sends: a working mechanism, not a measurement, and the write-up says which is which.
The parts worth reviewing:
- the results—what ran, and the ten things that broke on the way. The one worth opening it for: Smart Sending silently skipped the first abandoned-cart email because a welcome had gone out five hours earlier. Not a bug, the feature working and costing a conversion. Nothing in the API said so—no send event, no skip event—and one screen in the ESP names the reason. It was left switched on and reported rather than switched off;
- the consent design—double opt-in enforced at list level, and a sync script structurally unable to grant consent, only to refuse to act without it. A subscriber who arrived through Shopify's own newsletter field,
SUBSCRIBED86 ms after the webhook with no confirmation, is documented as the leak it is; - the flow specification, which records why each delay was chosen and what test would replace it—including the one claim still unproven, and the experiment that would settle it.
A marketing and CRM case study connecting campaign performance to budget, customer lifecycle and retention decisions. The CRM work uses the real UCI Online Retail II dataset; campaign, attribution and consent examples are clearly labelled deterministic simulations because an equivalent permissive public dataset is not available.
What to open first:
- the dashboard preview;
- the executive summary;
- the five-minute recruiter route.
A customer-experience analysis of about 95,000 real reviews from the Olist Brazilian e-commerce dataset. The clearest finding is practical: late delivery is strongly associated with lower satisfaction. The repository also examines other drivers, review themes and differences across customer groups without claiming that observational relationships are causal.
What to open first:
- the CX driver analysis;
- the stakeholder brief;
- the data provenance.
A small working integrity and reporting layer for AI-assisted workflows. It records a verifiable audit trail, preserves approval decisions and produces an EU AI Act Articles 12/13/14 report. It is intentionally scoped: this is not presented as a complete production agent platform.
Open the live compliance report
This detail is here for reviewers who want to inspect methods and implementation after the business overview.
| Project | Methods and evidence | Main tools |
|---|---|---|
| Lifecycle Automation | Event-triggered flows with exit conditions re-evaluated before every send, double opt-in as a list-level gate, suppression logic, RFM→ESP segment sync behind a hard consent check, sending-domain authentication (NS delegation, DKIM, DMARC), UTM taxonomy across all flow links | Klaviyo, Shopify, Python standard library, unittest, DNS |
| Digital Campaign Performance | RFM, cohort retention, historical CLV, A/B uplift, multi-touch attribution including Markov removal effect, saturation-aware budget scenarios, GA4 event planning and UTM/consent governance | Python, SQL, Excel, Power BI/Tableau/Looker specifications, pytest, GitHub Actions |
| E-commerce CX Analytics | Disclosed NPS proxy, two-proportion tests, Cohen's h, confidence intervals, delivery-SLA cohorts, multivariate logistic regression and Portuguese comment themes | Python standard library, real Olist review data, unit tests, GitHub Actions |
| Agentic Audit Reporting | SHA-256 hash-chained JSONL events, replay and verification, human approval records, HTML compliance reporting | Python standard library, pytest-compatible tests, GitHub Actions, GitHub Pages |
The analytics projects use a real public dataset with source and license information, or clearly labelled deterministic simulation where suitable public data is not available.
The lifecycle project is different and is labelled as such in its own README: the Klaviyo account, the Shopify store and the authenticated sending domain are real, but the store is a development store and the orders in it were seeded for the build. Every flow is in draft, nothing has been sent, and the list has no subscribers. The mechanism is real; the volume is not.
Nothing in this portfolio claims access to client, advertising-platform, production CRM or real customer data.
Bologna, Italy · CET — open to remote work across EU time zones.
I'm looking for an internship or junior role in marketing automation and CRM.
- LinkedIn: riccardo-capanna
- Email: errer441122@gmail.com