Data skills

Own CAUSAL, incremental marketing measurement, choose the model by a spend/data-maturity tier (small → lift tests; mid → multi-touch attribution checked against holdouts; large → Marketing Mix Modeling + geo-holdout incrementality), never defaulting to the fanciest, and separate a real mix shift from a tracking artifact via signal-integrity diagnostics (Consent Mode v2, server-side tagging, CAPI).

Produce the white-labeled monthly client report from one input (client name + connected accounts), pull GA/GSC/Meta/ads metrics, trend WoW/MoM, narrate wins/misses, set next-month priorities, render under the client's logo and brand. The #1 recurring retainer time sink, done in one pass.

Build a live, reopenable executive dashboard (HTML artifact) that pulls fresh numbers from connected tools on each open.

Configure and AUDIT Google Analytics 4, read via the official read-only analytics MCP, write config via the GA4 Admin API (mark/create key events + conversions, custom dimensions/metrics, enhanced measurement, data streams, retention), audit configured-vs-actually-firing events, and apply the 10/10 conversion blueprint for lead-gen/agency accounts.

Analyze a campaign, channel, or content set and report wins, misses, and prioritized optimizations.

Compose owned analytics (Umami for privacy web stats, PostHog for product/flags/replay, GlitchTip if Sentry residency hurts) one tag per job, licensed, on hostssh. Not GA4 admin (as-analytics-ga4-admin), not a monthly client report (as-analytics-client-monthly-report).

Produce the weekly business metrics report (acquisition, activation, revenue, retention, content/SEO, and pipeline) with trends and recommended actions.

Answer a data question end to end, from a quick lookup to a full investigation (explore → query → analyze → validate → visualize → communicate).

Rebuild a catalogue from the site still serving it, find the JSON API behind the listing UI instead of scraping the DOM, harvest every field, then recover the binary assets with a resumable parallel downloader. Quotes metadata and assets as two separate jobs because they differ by orders of magnitude.

Capture a company's data-warehouse tribal knowledge into a reusable reference, schemas, metric definitions, join patterns, gotchas, and the canonical query for each KPI.

Turn a scanned or photographed document into governed, structured JSON, layout-and-table-aware OCR for signed contracts, receipts, faxed forms, and flattened legacy PDFs, with per-field confidence, source bounding boxes, internal-consistency validation, and a low-confidence human-review queue. Never fabricates a field; PII is handled, never leaked.

Profile and explore a dataset before analysis, shape, grain, nulls, distributions, cardinality, duplicates, outliers, and data-quality issues.

Harvest structured information from repositories, documents, local files, APIs, websites, or web apps; normalize it into a provenance-rich, deduplicated package; and stage or import it into applications, sites, databases, documents, or Google Docs.

Data category hub: routes to the right leaf for answering a question (analyze, explore, SQL, stats), trusting the answer (validate, lineage and provenance), showing it (viz, JSON canvas), and standing up the sources (owned corpus, knowledge graph, document OCR extraction, warehouse context). Entry point for the data domain.

Turn an idea, structure, or set of notes into a visual JSON Canvas (.canvas) file, mind maps, flowcharts, project/kanban boards, and research canvases, valid to the JSON Canvas 1.0 open spec (nodes + edges), grid-aligned and grounded in real source.

Turn a pile of source documents (PDFs, markdown, transcripts, scrapes) into a governed, human-readable entity-relation knowledge graph in the Project Brain, extract → profile → dedup → index → dual-level query.

Trace every client-report number back through its pipeline (source feed → join → derived column → metric → report cell) and fingerprint each upstream feed so a silent shape change is caught before the report ships, not after the client asks. Keywords: lineage, provenance, source-of-record, shape drift, schema drift, fingerprint, blast radius, impact analysis, fanout, attribution-window.

Stand up and keep fresh the first-party corpus every content system runs on, credential-safe routines that pull call and meeting transcripts, YouTube and podcast transcripts, every email you have sent with its open and click stats, wins screenshots, and case-study interviews into one governed folder with stable filenames, incremental sync, and duplicate-recording dedupe.

Write correct, performant SQL for any warehouse dialect. Or fix, optimize, or translate an existing query.

Run statistical analysis (descriptive stats, trend/seasonality, outlier/anomaly detection, hypothesis tests, and correlation) with the right method and the misuse caught.

QA an analysis before it's shared, check methodology, accuracy, denominators, timezones, filter leakage, and bias.

Build a publication-quality chart: the right chart type for the question, accessible and honest-axed (matplotlib/plotly).

Per-engagement research operations desk: decompose the question, gather independent sources, cross-source triangulate, cite or flag every claim, then synthesize a verified dossier (company · market · competitors · regulatory · buying-signals) and refresh it on a schedule. Productizes the $5-15k research deliverable.