Toby Lowe Analysis you can check · Manchester

Business Analyst · MSc Data Science, University of Manchester · E.ON Next

I write down what I expect to find before I look.

Four years at E.ON Next, from answering phones during the energy crisis to building the reporting infrastructure the operation runs on. Cross-site BI pipelines used by the Bolton Operational Leadership Team for resourcing, cost estimation and seasonality planning. Regulated billing work accurate to the penny under Ofgem requirements. AI deployment across operations alongside the Chief of Staff team. Two consecutive 'Outstanding' performance ratings.

Alongside all of it, an MSc in Data Science at the University of Manchester, currently at a distinction average with the thesis outstanding. I have never treated study and work as separate phases.

What connects the two is a way of working. My dissertation estimates household responsiveness to dynamic electricity tariffs across 18.9 million half-hourly smart-meter readings. Its specification, hypotheses and robustness ladder were committed to a pre-analysis plan before the first household-level model was run. Every departure from that plan since is logged with a date and a reason.

That plan contained stopping rules. One of them fired: the pre-registered parallel-trends test failed, so the difference-in-differences it was gating was not built, and five analyses that depended on it were not run. The result I most wanted, that vulnerable households would show a smaller capacity to shift, is not supported by my own data, and the site says so in the same typeface as everything else.

I am not claiming this makes the work correct. I am claiming it makes the work checkable, which is a different and more useful property. It is also the habit I bring to operational reporting, where the cost of a number nobody can verify is paid by whoever acts on it.

Now
Business Analyst, E.ON Next
Studying
MSc Data Science, University of Manchester — distinction average, thesis September 2026
Before
BSc Data Science, First Class, Nottingham Trent
Based
Manchester, UK
Looking for
Analyst, reporting and data roles
Contact
tobyjohnlowe@gmail.com

Evidence One estimate, and everything that tried to break it

Estimates and falsification tests on one shared axisThe primary estimate is minus 4.16 per cent with a 95 per cent confidence interval from minus 4.83 to minus 3.48. An independent replication on a separate cohort gives minus 5.34. Four hundred placebo tariff calendars span minus 3.40 to plus 3.51, and none reaches the primary estimate. Two shifted-event placebos land on zero as intended. A falsification test on never-treated households returns plus 0.99 per cent when it should return zero; it is flagged as unresolved.−6−5−4−3−2−10+1+2+3+4% change in consumptionReplication, separate cohortSN 7857 · 1,025 households · own price series−5.34%Primary estimate1,117 households · 18,879,437 half-hours−4.16%400 placebo calendarsrandomisation inference · 0 of 400 reach the estimaten = 400Placebo · events +1 dayexpected ≈ 0−0.03%Placebo · events −1 dayexpected ≈ 0+0.04%Falsification, never-treated arm4,411 households · expected ≈ 0 · it is not+0.99%
Fig. 1 Every estimate and every falsification test from the dissertation, on one shared axis. Effect on household electricity consumption during high-price half-hours. Source: ERP/Full Code/outputs/ — stats_phase6_household.json, stats_t1_std_falsification.json, stats_t2_randomization_inference.json, stats_t6_sn_cohort.json, robustness_table.csv

Read it left to right. The estimate and its independent replication sit clear of zero. Four hundred placebo tariff calendars — the same model run against randomly shifted event dates — span the middle and none of them reaches the estimate. Two more placebos, built by moving every event a single day, land on zero exactly as they should. The last row should also have landed on zero and did not, and that is reported here rather than left in an appendix.

Work Indexed by the question, not the toolchain

  1. 01 Do households actually shift electricity consumption when prices change — and do the households least able to absorb a high price shift less than everyone else? Household responsiveness to dynamic electricity tariffs Consumption falls 4.16% under high prices [95% CI −4.83, −3.48]. The fairness gap I pre-registered is not there — −0.63pp, p = 0.57. Pre-registered MSc dissertation · 2026-09within-household fixed effects · randomisation inference · event study · pre-registered robustness ladder
  2. 02 Does the topological shape of gas-price returns carry regime information that GARCH-family volatility models cannot see? Persistence norms and conditional volatility in European gas markets 26 sustained regime signals at the crisis peak against GARCH's 0 — and 0 against 39–46 through the recovery. Complementary, not superior. MSc · 2026-05topological data analysis · Takens embedding · persistent homology · GARCH
  3. 03 Can a camera-based drowsiness detector built from off-the-shelf components actually work in a moving car, rather than only on a held-out test set? Real-time driver drowsiness detection 98.31% on the test set; near-total false positives in a moving car. The road test is how I found the data leakage. Award BSc dissertation · 2025-04convolutional neural networks · transfer learning · gradient boosting · stacked ensembles
  4. 04 What can a twenty-five-year record of UK banking and payments behaviour honestly support as a projection to 2034 — and where does the projection stop being defensible? UK banking and payments to 2034 Branch network down over 70% since 2000, r = −0.98 against remote banking. The 2034 figures are scenarios, and the assumptions dominate them. Group work MSc · 2026-05Bass diffusion · Fisher-Pry substitution · OLS · scenario forecasting

Also

  1. 05 Where does patient delay in a hospital Clinical Pharmacy Unit actually accumulate, and which staffing or automation changes would reduce it most? Simulating an NHS hospital pharmacy unit 97.7% of a 38-hour patient pathway is queueing, not care. Every recommendation is a simulation projection, not a measured outcome. Group work MSc · 2026-05discrete-event simulation · distribution fitting · queueing analysis · scenario testing
  2. 06 Can six weeks of daily sales be forecast across the 1,115 Rossmann Germany stores in the dataset from store metadata, promotion schedules and calendar effects, and which model class handles that tabular structure best? Forecasting six weeks of Rossmann store sales XGBoost at RMSPE 0.1867 against a 0.4383 linear baseline — one holdout split, no intervals, no significance test. Group work MSc · 2026-01MICE imputation · missingness-mechanism diagnosis · feature engineering · exploratory data analysis
  3. 07 How much of the clinical normal/abnormal distinction in the vertebral column biomechanical dataset is recoverable from geometry alone, and how much does supervision add? Supervised versus unsupervised classification of vertebral column data Logistic regression 80.65% against k-means at 65.5% — a single-split point estimate I would not defend as significant. Group work MSc · 2026-03exploratory data analysis · k-means clustering · principal component analysis · logistic regression
  4. 08 How much of a board game's rules can be enforced by relational schema constraints and triggers rather than by application code? University Tycoon: a board game modelled in SQLite 8 tables, 11 triggers, 1 view. Movement, transfers and ownership are enforced by the schema; several other rules are not. MSc · 2025-11entity-relationship-modelling · isa-inheritance · third-normal-form · sql-triggers

CV Experience and education

E.ON Next August 2021 — present

Descriptions here are limited to role, dates and general responsibilities. No internal figures, dashboards, customer data or work products appear on this site.

Business Analyst

November 2024 — present

Building, testing and reporting on custom AI agents on Google's Gemini AI platform, automating repeatable tasks across customer operations.

Partnering with the Chief of Staff team on AI deployment across operations, and leading the Summer '26 AI Roadshow — an in-person programme delivered across the Bolton, Nottingham and Leicester sites.

Led the internal AI Working Group, identifying and progressing AI business cases with senior operational leadership.

Business Readiness Lead

February 2024 — November 2024

Expanded a cross-site task distribution and BI reporting system across Bolton, Leicester and Nottingham, standardising inconsistent back-office processes into a single operational solution. Automated pipelines on Amazon Athena, Google Cloud APIs, Apps Script and Tableau, used by the Bolton Operational Leadership Team for resourcing allocation, cost estimation and seasonality planning.

Built the prepayment component of a manual final billing solution, billing vulnerable prepayment customers to penny-level accuracy across years of price, meter and tariff changes under Ofgem requirements.

Delivered a prepayment demand handover system after identifying a compliance gap, routing ringfenced vulnerable-customer cases to a specialist team with automated allocation and volume reporting.

PAYG Analyst

November 2022 — February 2024

Built Tableau dashboards, alerting and statistical reporting across billing and prepayment operations, replacing manual reporting with SQL/Athena-driven workflows.

Led root-cause exploratory analysis for a billing improvement programme: defined the methodology, treated outliers and quantified the intervention's impact, which directly informed the decision to continue the work.

Energy Specialist

August 2021 — November 2022

Supported vulnerable customers through the national energy crisis. Led a repeat-demand analysis project that contributed directly to promotion into an analyst role.

Recognition

Two consecutive 'Outstanding' performance ratings.

MSc Data Science with Business & Management

The University of Manchester · 2025 — 2026

Current average: distinction. Only the thesis remains to be graded.

Thesis (in progress, submission September 2026): Quantifying Household Responsiveness to Dynamic Electricity Tariffs — an event-study and randomisation-inference analysis of the Low Carbon London smart-meter trial, including fairness analysis across customer segments. Supervised by Dr Fanlin Meng.

Modules: Statistics & Machine Learning I–II · Applied Data Science · Simulation & Risk Analysis · Topological Data Analysis · Understanding Data · Understanding Databases

BSc (Hons) Data Science — First Class

Nottingham Trent University · 2022 — 2025

Dissertation: real-time driver-drowsiness detection using a custom CNN, with a fine-tuned ResNet-18 comparison and a road-tested live alert dashboard. Supervised by Dr Archontis Giannakidis.

1st place, 2025 Nottingham Trent University Mathematics Degree Showcase Poster Competition, awarded by external employers.

NTU Digital Employability Award — Gold.

A Levels

New English International School, Kuwait

Business Studies · Mathematics · Physics · Economics

Technical

Languages — Python (pandas, NumPy, scikit-learn, XGBoost, PyTorch, statsmodels, matplotlib) · R (statistical computing, Monte Carlo simulation) · SQL (Amazon Athena, SQLite)

Methods — Panel fixed effects · randomisation inference · event studies · pre-registered analysis and robustness ladders · time-series and volatility modelling · topological data analysis · discrete-event simulation (Simul8)

Data & BI — Tableau · Power BI · Looker · Google Cloud & Apps Script

Delivery — Agile (Scrum) · Member, Association for Project Management · BPMN process mapping

Analytical roles — business analysis, reporting, data — where the question is genuinely open, and the job is to establish what is true and what the evidence will not carry, rather than to move a metric that has already been chosen.