tailor_cv / cover_letter / interview_prep system prompts (from poc/ollin-dossier-verbs) applied to the synthetic Lena→Helvetia fixtures. The demo's mock model returns a deterministic stub; this is the quality the wired oll-model path is built to produce. The person and both documents are fictional. Nothing here is inflated: every claim traces to Lena's profile, and where the job asks for something the profile doesn't support, the output leaves an explicit gap rather than invent it.A good send-worthiness test needs both genuine strengths to foreground and a genuine gap to handle honestly. This pairing has both — which is exactly where a dishonest tool would fabricate.
Data analyst with six years turning operational data into decisions the business acts on — strong SQL and Python, and end-to-end ownership from pipeline to stakeholder dashboard. Warehouse-and-dbt reporting experience across insurance and retail, the two domains closest to Helvetia's Non-Life analytics.
Data Analyst — Migros Genossenschaft, Zürich · 2021–present
Junior Data Analyst — Die Mobiliar, Bern · 2019–2021
Data Intern — Bernexpo AG, Bern · 2018–2019
MSc Statistics — University of Bern, 2018 · BSc Mathematics — University of Bern, 2016
SQL · Python (pandas, scikit-learn) · dbt · BigQuery · Tableau · Git · stakeholder reporting · A/B testing · data modelling
German (native) · English (C1) · French (B1)
[add: the role mentions mentoring two junior analysts / line-management — your profile doesn't state formal mentoring. If you've informally guided colleagues, add one concrete line; otherwise it's honestly a growth area, addressed in the letter.] — this is the groundedness rule working: it flags the shortfall instead of inventing a leadership title.Dear Hiring Team,
I'm applying for the Senior Data Analyst role in your Non-Life division. What draws me to it is that it's the work I've been moving toward: owning a reporting domain end to end, and doing it on a modern, version-controlled stack rather than a pile of spreadsheets.
At Migros I rebuilt the weekly sales-reporting pipeline in dbt and BigQuery, cutting refresh time from six hours to twenty-five minutes and removing three manual steps — the same shift toward tested, version-controlled transformations your posting describes. The Tableau demand dashboard I built there is used by more than forty regional managers, so I'm used to turning a model into something leadership relies on weekly.
The insurance context isn't new to me either: at Die Mobiliar I automated monthly claims reporting and reconciled a two-million-row policy dataset for the actuarial team. Together with an A/B test at Migros that lifted loyalty sign-ups by 12%, that covers most of what you're asking for — advanced SQL and Python, a modern warehouse with dbt, a dashboarding tool in production, and experiment design communicated to non-technical stakeholders.
One area I'd be growing into is formally mentoring two junior analysts; my experience there is informal rather than line-management, and I'd welcome that as part of the step up. I'd bring the same end-to-end ownership to it that I've brought to every pipeline I've built.
I'd be glad to talk about the Non-Life reporting suite and where I could take it.
Kind regards,
Lena Brunner
| They'll ask | Why it comes up | Your angle (from your real work) |
|---|---|---|
| Walk us through your first 90 days taking over the Non-Life reporting suite. | The role is end-to-end ownership of a business-critical domain — they want to see you can inherit and stabilise it. | Map the existing pipeline first; port fragile steps into tested dbt models (as you did at Migros: 6h→25min, 3 manual steps removed); prioritise the dashboards leadership uses weekly before anything cosmetic. |
| Tell us about raising the engineering bar on reporting. | The JD explicitly wants version-controlled, tested transformations. | The dbt + BigQuery rebuild at Migros — version control, tests, and the refresh-time win that made it stick. |
| You'd mentor two junior analysts — what's your experience leading others? | Gap probe. The JD wants mentoring / line-management; your profile doesn't state it. | Be honest: no formal line-management yet. Describe informal guidance and code-review habits, and say plainly you'd welcome growing into structured mentoring. Don't claim a title you haven't held — they'll test it. |
| How do you design an A/B test and explain it to non-technical leaders? | Explicit requirement — experiments on customer-facing flows, communicated upward. | The checkout-prompt test (+12% loyalty sign-ups): hypothesis, guardrail metrics, and how you framed the result for regional managers. |
| What's your experience with claims / policy data? | Domain fit for Non-Life insurance. | Die Mobiliar: automated monthly claims reporting and reconciled a 2M-row policy dataset for actuaries — real insurance-data ownership. |
| Where are you weakest against this role? | Self-awareness + they'll circle back to the mentoring / scale gap. | Name it: formal people-leadership. Commit to pairing and structured mentoring; anchor on the technical ownership that's clearly strong. |
POC-A's pass signal was: Sam judges ≥1 output good enough to send under his own name; every claim traces to a cited passage; no invented facts. That's now judgeable — this dossier is grounded strictly in Lena's profile, foregrounds her real matches, and handles the one genuine gap honestly in all three artifacts.
llm_client.py, no architecture change. Either way you're judging quality, not plumbing — which was the whole point of splitting it into a POC. Related: the demo · POC results · dev-test manual.