Who built this, why, and what it is for
StepChangeAnalysis.com is built on fifty years of working with systems where getting the measurement wrong has real consequences — pharmaceutical-grade manufacturing, major hazards plant management, and clinical software for patients on high risk medications. The question that runs through all of it is the same: how do you know whether anything has actually changed?
Who built it
Twelve years as a process engineer and plant manager at a major chemicals company producing a high-specification product with impurity limits measured in parts per million — where a shift in process mean is not just an efficiency question but a product quality and patient safety question. That environment was the first encounter with CUSUM: a statistical tool that detects process shifts that standard control charts miss.
Plant manager on a major hazards installation — COMAH-regulated, where the consequences of not detecting a structural change in process behaviour are measured in more than financial terms. The discipline of major hazards management asks the same questions as good improvement work: what is the system actually doing, how do you know, and what level of intervention produces durable change rather than temporary compliance?
Forty-plus years managing a clinical software company helping healthcare professionals manage patients taking high risk medications — anticoagulants, cytotoxics, immunosuppressants, biologics, and biosimilars. The software supports the monitoring and clinical decision-making that keeps those patients safe. The data those systems generate is time-series data. The question of whether a patient’s medication control has structurally improved — or whether variation is noise around an unchanged mean — is precisely the question Bootstrap CUSUM is designed to answer.
StepChangeAnalysis.com is the convergence of those three strands: the statistical method from pharmaceutical manufacturing, the systems thinking from major hazards management, and the clinical context from forty years in healthcare software.
The career behind the site
The connection between pharmaceutical manufacturing, major hazards plant management, and NHS improvement data is not coincidental. They share a common structure: a system operating under tight specifications, a measurement challenge, and high stakes if the wrong conclusion is drawn from the data.
Provides decision support software and services supporting chronic disease patients taking potentially harmful medications, with the objectives of enhancing patient safety, quality of care and healthcare effectiveness. DAWN anticoagulation management software is used across the NHS and internationally. Over 40 years of building, improving and running a specialist clinical software business — navigating the Innovator’s Dilemma from the inside, applying every concept on this site in the actual conditions they were designed for.
Strategic data planning and database consultancy for major organisations including ICI Pharmaceuticals (now AstraZeneca), Unilever, and British Nuclear Fuels. Provided training and consultancy in structured data planning, database design, and systems performance improvement.
Planning systems for improved performance in chemical works. Member of a joint ICI / IBM study team for two years, working on the application of information systems to chemical plant operations and performance measurement. First systematic exposure to how data, decisions and operational systems interact in large continuous process environments.
Concurrent with plant management at ICI. Tutored the Open University’s systems performance course — a discipline covering human factors, system failure modes, and the gap between how systems are designed to work and how they actually behave under operational conditions. The course material shaped the diagnostic frameworks that inform this site: understanding failure as a property of systems, not of the people operating within them.
Plant manager on various large continuous process plants including a Major Hazards Installation (COMAH-regulated). The hydrogen plant case study on this site — the 17% efficiency gap, the residual CUSUM monitoring system, the optical pyrometer corroboration — is from this period. Managing continuous processes operating around the clock with tight specifications and high operational stakes was where CUSUM became a practical tool rather than a statistical concept.
Process quality control and process improvement using CUSUMs and Statistical Process Control. First direct application of CUSUM in an industrial setting — monitoring product quality and process performance against specification. The fundamental question was the same one that runs through every article on this site: how do you distinguish a genuine process shift from background noise, and how quickly can you detect it?
“The question I was trying to answer in 1968 — has this process genuinely changed, or is this variation just noise? — is the same question Bootstrap CUSUM answers in 2026. The data changes. The industries change. The computational tools change. The underlying question about what constitutes real evidence of change versus random variation is invariant. That question is what this site is about.”
Why the tool is free
The Bootstrap CUSUM tool costs nothing to use, requires no login, and uploads no data. Every calculation runs entirely in your browser. This is a deliberate choice, not an oversight.
The practitioners who most need a statistically honest answer to the question has my improvement programme actually worked? are not always in organisations with commercial analytics budgets. A QI lead in an NHS trust, a safety manager in a small manufacturer, a clinical pharmacist asking whether a new monitoring protocol has genuinely changed patient outcomes — these are exactly the people this tool is designed for, and charging for it would stop most of them using it.
“Give people better tools and a better mental model to ask better questions about whether things have improved through structural change — and to understand why, when they haven’t.”
Built for any sector, anywhere in the world
The NHS articles on this site are the most developed example set because that is where the clinical software background runs deepest — but Bootstrap CUSUM and the diagnostic method here apply equally to any organisation, in any sector, anywhere in the world where improvement is attempted and measurement matters.
The underlying question is always the same: has this system structurally changed, or are we seeing variation around an unchanged mean? That question does not belong to any single sector. It belongs to anyone who collects time-series data and needs an honest answer about whether an intervention has worked.
The clinical software company behind this site has customers in the UK, the United States, Canada, and around the world — and the tool has been applied to data from process manufacturing, environmental monitoring, economics, and public services as well as healthcare. The articles will expand to reflect that breadth as the site grows.
Healthcare — globally
A US hospital monitoring readmission rates, a Canadian healthcare organisation tracking medication safety outcomes, an Australian aged care provider measuring quality indicators, a WHO vaccination programme evaluating coverage — all face the same measurement challenge as an NHS trust. Regulatory frameworks differ. The Bootstrap CUSUM question does not. The clinical software background on this site includes working with healthcare organisations in the UK, US, and Canada — where the clinical risk is the same wherever the patient is and the question of whether a patient’s treatment control has structurally improved is identical in every jurisdiction.
Process industries and manufacturing
This is where the career behind the site began. High-specification chemical manufacturing with ppm-level impurity control is where CUSUM was first encountered as a practical tool — not in a textbook but on a plant where detecting a small, sustained shift in process mean was the difference between a batch that met specification and one that did not. That discipline applies directly to pharmaceutical process validation, batch release data analysis, continuous process verification (CPV), and any manufacturing context where a process shift needs to be detected early, with statistical confidence, before it becomes a quality failure or a safety incident. Major hazards installations, COMAH-regulated sites, safety-critical process environments — all share the same need for honest, statistically robust change detection.
Environment and public policy
Did the clean air zone actually change air quality — or did traffic patterns shift before the zone launched? Did the emissions reduction policy produce structural change in the data — or did energy prices do the work? Bootstrap CUSUM answers these questions without the need for a control group, a randomised trial, or a regression model. It simply asks: did the process permanently shift to a new mean, and if so when? That question is as useful to an environmental regulator as to an NHS trust. The ULEZ and UK emissions articles on this site demonstrate the method applied to environmental data — and the findings are often more nuanced than the policy narrative suggests.
Education, local government, and public services
School outcome data, pupil attendance rates, council service performance, benefit claim processing times, housing allocation rates — all generate time-series data and all prompt the same question after an intervention: has anything actually changed? The Bootstrap CUSUM gives a statistically honest answer that a simple before-and-after comparison cannot. It is particularly useful in public service contexts where political pressure to declare success arrives well before the lag time required for a structural change to become statistically detectable.
Safety-critical industries
Aviation incident rates, nuclear plant performance indicators, rail safety metrics, offshore process safety data — all share the COMAH discipline of asking whether the safety barrier is holding or degrading. In safety-critical industries, detecting a structural shift in a safety indicator early — before it becomes a significant incident — is the entire purpose of process safety monitoring. Bootstrap CUSUM detects that shift with statistical confidence, distinguishing genuine degradation from noise at a level that standard SPC charts cannot match.
The intellectual framework
Bootstrap CUSUM answers one question well: has the process structurally shifted to a new mean? But knowing whether something has changed is only useful if you also understand why it has not changed, and what kind of intervention would actually produce the step change you need. The thinking on this site draws on four bodies of work:
| Framework | Core insight | Where it appears |
|---|---|---|
| W. Edwards Deming | 94% of problems are caused by the system, not the people in it. Targets without methods produce distortion, not improvement. Variation is the diagnostic clue — look hardest at where outcomes vary most. | Every article. The tampering trap. The by what method question. Start here. |
| Brian Joiner | Three levels of intervention: fix the output (Level 1), fix the process (Level 2), fix the system (Level 3). Only Level 3 produces structural change that Bootstrap CUSUM will confirm. The COMAH hierarchy maps directly onto Joiner’s levels. | Why nothing changes. The A&E, dementia, and never events articles. |
| Eliyahu Goldratt | The constraint determines throughput. You cannot improve a system by optimising anything other than the constraint. Subordinate everything else to the constraint before investing in elevating it. | The dementia article (memory clinic capacity as the constraint). Why nothing changes. |
| Senge & Meadows | Systems have reinforcing loops that make poor performance self-sustaining. High variation in outcomes is the fingerprint of a system dependent on people rather than design. Structural change requires intervening at the right node in the loop. | The A&E and dementia articles (reinforcing loops mapped). Why nothing changes. |
| Langley et al. | The three questions of the Model for Improvement: what are we trying to accomplish, how will we know a change is an improvement, what change can we make. Bootstrap CUSUM sits in the Study phase — the answer to Question 2. | Start here — the three questions. |
None of these frameworks is new. What is less common is applying all of them together to the same dataset, with a statistically honest measurement tool that confirms or denies whether the intervention worked. That combination — honest measurement plus structural diagnosis plus the right level of intervention — is what this site is trying to make more accessible.
The Bootstrap CUSUM method
Bootstrap CUSUM is a distribution-free method for detecting structural changes in time series data. Unlike standard SPC charts, it does not assume normality — making it particularly suited to healthcare safety data, incident counts, and pharmaceutical process data where normal distribution assumptions produce unreliable control limits.
The method computes the cumulative sum of deviations from the overall mean, then uses bootstrap resampling to establish confidence limits for each detected change point. A change point is only declared when the bootstrap confidence exceeds the threshold you set — typically 95% for standard analytical use, 99.7% for governance reports where false positives carry significant cost.
Feedback, corrections, and collaboration
Every article on this site uses publicly available data, documented methodology, and cited sources. If you find an error in the analysis, a misidentified data source, or a statistical interpretation you disagree with, please say so. The feedback form on the tool page is the best way to make contact.
If you work in clinical software, pharmaceutical manufacturing, process safety, or any sector where detecting a structural change in a process matters — and you would like to discuss applying Bootstrap CUSUM to your own data — that conversation is welcome.