For ENSEK · Data & Reporting · the 'deliver, don't apply' proof
Here's your work.
Here's the AI that does most of it.
You're hiring a Senior Data Analyst. We took the job description apart, built an agentic system that genuinely does the automatable two-thirds of it — live, in your browser, not slides — and we'll show you exactly which third still needs a person, and why. Then we'll tell you the headcount you'd actually hire.
01 · The work
A Data Analyst's job, decomposed.
Straight from the ENSEK Senior Data Analyst JD (Nottingham) and the standard energy-retail data-team remit. Twelve tasks, graded honestly: what an AI does end-to-end, what it drafts for a human to own, and what stays human.
| Task | Verdict | Why |
|---|---|---|
| Routine SQL pulls / ad-hoc data questions | automate | Deterministic, checkable. Demonstrated. |
| Dashboard drafting & refresh | automate | Spec→query→chart is mechanical. Demonstrated. |
| Billing-discrepancy / reconciliation audit | automate | Recompute the whole book each cycle. Demonstrated — 114 flagged live. |
| Estimated-read / data-quality monitoring | automate | Continuous cohort detection. Demonstrated. |
| Debt-book & early-arrears detection | automate | Billed-vs-paid arithmetic. Demonstrated. |
| Decision-ready report drafting | assist | AI drafts; analyst owns the interpretation. Demonstrated. |
| Metric-definition governance | assist | AI proposes & enforces one definition; human signs off. Demonstrated. |
| A/B test design & read-out | assist | AI computes; human owns validity & the ship call. |
| Stakeholder framing of ambiguous problems | human-only | Needs org context, politics, judgement. |
| Coaching / raising analytical standards | human-only | Mentorship is a human relationship. |
| Customer corrections, refunds, collections | human-only | Accountable to customer + Ofgem. AI detects; a person decides. |
| Pricing / tariff decisions | human-only | Regulatory, fairness, commercial weight. |
Five of the highest-volume tasks are demonstrated running live on the next page — over a synthetic ENSEK-shaped dataset, computed in your browser. Not a mock-up. Edit the SQL yourself.
02 · The system
An agentic backend pointed at the automatable work.
Ask in English
The agent translates the analyst's question into SQL — and exposes the SQL so it's auditable, not a black box.
Runs for real
Queries execute live against the data. Billing reconciliation re-derives every charge from tariff maths — across the whole book, every cycle.
Writes the read
Each result comes with a decision-ready narrative and a recommended action — drafted for a human to own.
Knows where to stop
Every output names its human control point. The system is built to hand off, not to overreach.
03 · The headcount you'd actually hire
The function doesn't go to zero. It changes shape.
Fewer hands turning the SQL crank. The people who remain are accountable owners and judgement roles — not query-writers. Post-automation, this is the team we'd argue for:
| Role | Heads | Why it survives automation |
|---|---|---|
| Analytics lead / owner | 1 | Owns metric governance, signs off canonical definitions, accountable for what the AI ships. |
| Billing / assurance analyst | 1–2 | Adjudicates flagged discrepancies, owns customer-facing corrections & regulator-facing accuracy. |
| Data engineer | 1 | Owns the pipelines & tariff/reference data the AI depends on. Garbage in still means garbage out. |
| Senior / principal analyst | 1 | Problem framing, experiment validity, exec partnership, coaching — the human-only column. |
The role that's absorbed: the 'write-the-SQL, refresh-the-dashboard' recurring-production headcount. You hire for accountability and judgement, not throughput. See the full human-core argument →
The provocation
You advertised one job.
Here's the system that does most of it — built before the interview.
This is what 'deliver, don't apply' looks like at org scale: real, sourced, dated, and running. The question isn't whether the work can be automated. It's which third you want a person accountable for.