Capabilities

What I cover, and how deep it goes.

37 capabilities across 7 domains, graded against one rule: a capability only reaches the top tier if a published case study backs it. The gaps are on the map too — a capability map with no thin spots is a brochure.

Evidenced (9)

A published case study backs this.

Core (20)

Years of production use.

Working (5)

Used in production, less depth.

Developing (3)

Learning, or deliberately thin.

Explore it in 3D

Loads a ~670KB 3D renderer, so it is behind this button rather than in the page weight. Drag to rotate, scroll to zoom, hover a node for detail. Everything it shows is also in the table below.

The same data, readable

Every capability, by domain

Audit & data quality

Capability	Depth	What it means
Metric reconciliation	Evidenced	Source → model → report until the divergence point is provable, and classified as a real change or a bug. Evidence: A seven-figure block of revenue with no category attached, There was pressure to make the number match. The number was right
Root-cause analysis	Evidenced	Proving why a number moved, and separating a data artefact from a business change before anyone reacts. Evidence: The metric had not collapsed. Its attribution had, The aggregate metrics did not move. The harm was real anyway
Data quality framework	Evidenced	Nine dimensions with KPIs against each, plus the communications plan that makes quality visible outside the data team. Evidence: Data quality was ad hoc and invisible. It needed a definition first
Warehouse cost governance	Evidenced	Per-application service accounts so consumption is attributable, and documented monitoring so the whole team can see it. Evidence: The quota limit was the symptom. The absence of attribution was the problem
Observability (Elementary)	Core	Freshness, volume and schema monitoring on critical models, so degradation is detected rather than reported by a stakeholder.

Analytics engineering

Capability	Depth	What it means
dbt modelling	Core	Layered staging → intermediate → marts, so a change stays local instead of rippling.
Tests & CI	Core	Assertions on the things that would actually hurt, running on every pull request.
Metric definitions	Core	One definition per metric, in code, so two reports cannot quietly diverge.
SQL at depth	Core	Window functions, incremental logic, partition-aware rewrites, and query plans read rather than guessed at.
Lineage & documentation	Core	A dashboard number traceable back to the source column, documented where it is defined rather than afterwards.

Platform & pipelines

Capability	Depth	What it means
Backfills & partition repair	Evidenced	History corrected unsampled, then the structural follow-up so the same gap cannot reopen. Evidence: The metric had not collapsed. Its attribution had
BigQuery	Core	Partitioning, clustering, quota and slot behaviour, and large backfills that finish.
Airflow orchestration	Core	Dependency-aware scheduling with alerting that fires on the failure rather than the symptom.
CDC ingestion (Kafka / Debezium)	Core	Operational databases streamed into the warehouse, with completeness reconciled rather than assumed.
Python	Core	Pipeline glue, analysis, and the scripts that turn a one-off investigation into something repeatable.
Terraform / IaC	Developing	Developing, and stated as such. If a problem is primarily infrastructure-as-code, I am probably not the right person for it.

BI & reporting

Capability	Depth	What it means
BI migration	Evidenced	Side-by-side validation per report, with quality fixes folded in and any changed number made explicit. Evidence: A legacy reporting estate, moved without losing anyone’s trust
Metabase	Core	Self-serve models and dashboards built on governed metrics rather than bespoke SQL per report.
Looker Studio	Core	Executive and marketing reporting, wired to the modelled layer.
Self-serve design	Core	Datasets shaped for the questions people repeat, so routine questions stop becoming tickets.
Report estate rationalisation	Core	Usage data joined to cost, then a keep, merge or retire decision with an owner attached.

Product analytics

Capability	Depth	What it means
Funnel & cohort analysis	Evidenced	Cohorts isolated against a control group, with the substitution effects that hide harm in the aggregate. Evidence: The aggregate metrics did not move. The harm was real anyway
Statistical significance	Evidenced	Findings bounded by their standard error, so an effect is attributable rather than asserted. Evidence: The aggregate metrics did not move. The harm was real anyway
GA4 & raw export	Core	Event data modelled in the warehouse rather than trapped in the vendor UI.
Tracking plan design	Working	Events tied to decisions, with naming, properties and identity agreed once and applied everywhere.
Consent-aware measurement	Working	Default-denied signals, and an honest statement of what is not measurable as a result.

AI-assisted workflows

Capability	Depth	What it means
AI-assisted quality assessment	Evidenced	A documented, repeatable assessment any teammate can run and get comparable results from. Evidence: Data quality was ad hoc and invisible. It needed a definition first
Generated documentation	Working	Model and column docs produced as part of the pipeline rather than as a task that always loses.
Data-handling boundaries	Working	Written policy on which data may reach which service, and which outputs a human must confirm.
Retrieval over own data	Developing	Thin, and shown as thin. Grounding a model in a private corpus is adjacent to this work, not proven in it.
Evaluation harnesses	Developing	Measuring whether model output is getting better between releases. Developing.

Leadership & strategy

Capability	Depth	What it means
Data strategy & OKRs	Core	A roadmap tied to business priorities, reviewed against what actually happened.
Architecture decisions	Core	Decisions recorded with their reasoning, so a future team can revisit them rather than re-litigate them.
Estimation & planning	Core	Work sized in several vocabularies at once, with the biggest risk to the estimate named up front.
Cross-team coordination	Core	Pulling in the right owner instead of blocking silently, and holding scope boundaries openly.
Onboarding & enablement	Core	Runbooks, references and review that force-multiply a team rather than one person.
Hiring & role design	Working	Defining the role, screening technically, and designing the interview that tests for it.

Depth is self-assessed, with one external check: the top tier is only awarded where a case study on selected work backs the claim, and those describe in-house work over eight years rather than consulting engagements.

Need one of the deep ones?

The evidenced capabilities are the ones worth hiring for. Tell me which problem you have and I will tell you whether it is one of them.

Start a conversation