Events that shaped my career and principles.
External events shape a career as much as any job description — sometimes a great deal more.
I came into data through the business, not the technology
The web was arriving, and every newspaper in the country was trying to work out what it was for, and whether it could safely be ignored. Most of them couldn't. I was running IT and operations for a regional newspaper group — print production and distribution across multiple sites, on a schedule that doesn't negotiate with anybody. Miss the vans and the paper doesn't exist that day; there's no version of tomorrow where you catch up.
I built the group's first networked distribution system, on a platform by Atex called Matrix. Before it, the logistics ran on paper, phone calls and the memory of people who'd done it for years — it worked, in the way things work when good people compensate for bad systems. The errors we'd been living with had come to feel normal, and normal is a dangerous thing for an error to become. Once the system was in, the operation became legible. You could see it. Because you could see it, you could question it. And because you could question it, you could improve it — in that order, always in that order.
It was easier to teach the person who knew the business how to build the systems than to hire a systems person and teach them the business. That's the whole of my career in one sentence.
From 2000 I did it again through my own company, Datavation — the same Matrix system for a county newspaper title in the same group. Same platform, a different business, and by then I knew it inside out. That second build put me on the radar of Atex themselves, and led directly to everything that came next.
I lived Y2K from the inside, and I've never accepted the lazy version where it was a fuss over nothing. The reason nothing happened is that an enormous amount of quiet, unglamorous work happened first — finding the exposure, testing it, fixing it, planning seriously for a roll-over that might genuinely have fallen over. Thousands of people did months of invisible work so the world could wake up on the first of January and complain it had all been overhyped. It became a lens I've used ever since: the load-bearing work is usually the work that never shows up in the headlines, and never gets the credit.
Learning to do it repeatably, not once
The industry that had only just moved online was now watching its readers move somewhere else again — to platforms that published themselves. Every media group suddenly needed to see its own numbers faster than the market was changing around it.
Atex sent me out to do it properly. I pioneered the business intelligence implementation for CanWest in Canada, then replicated the approach in Sweden and again in the United States. Three countries, one engine, different data every time. The build was the easy half — and it is not the half people assume. What mattered was standing up analytics centres of excellence inside each organisation and training the people there to run the reporting themselves, so the capability stayed after I got on a plane home and the cost came down every year I was gone.
A well-designed, configurable system becomes whatever the data makes it. You build the engine once and deploy it into context after context.
I first met that principle on the print floor, refined it across three countries here — and I'm building on it again now, with AI.
Turning regulation into rules a machine can run
The sub-prime market turned, lending froze, and regulators who'd been comfortably light-touch for the better part of a decade became, almost overnight, the most demanding audience in the building. I moved into financial services — regulatory reporting and BI inside a major bank's group legal and compliance function, covering regimes like the UK Takeover Panel and the EU Transparency Directive. On paper, a long way from a print floor. In practice, the same job at considerably higher stakes.
What I did, over and over, was sit between the law and the machine. Regulation arrives as language — shades of grey, conditions, room for judgement, and lawyers paid to argue about all three. A monitoring system needs precisely the opposite: a one or a zero, a rule it can execute today, at volume, without asking anyone's opinion. Somebody has to translate one into the other without losing the meaning in between. Underneath it I was building the deeply unglamorous foundations — reporting processes, controls, auditability, data lineage — that everything afterwards would stand on.
Barclays' contested bid for ABN AMRO carried a severe regulatory test: prove we could onboard an enormous volume of Friday-close holdings and have them reconciled and reportable by Monday — a whole acquisition's positions over a single weekend, at real volume, with the regulator watching closely. We built a rapid-onboarding pipeline, iterated for months on masked data at genuine scale, and got it green-lit. Then a rival consortium outbid us and the deal was gone. The system was never used in anger — but the capability didn't die. It compounded, and it was sitting there ready when it was needed for real. The value often isn't where the deal is.
Twelve months later the real crisis arrived. Lehman Brothers collapsed, and Barclays acquired its North American business over a single weekend in September. I was on the compliance and regulatory-reporting side, and the requirement was the exact shape of the thing we'd already built and proven on the deal we lost: onboard an entire failed bank's positions between Friday close and Monday open. Because that capability already existed, we could stand in front of the regulator and demonstrate it rather than promise it. There is an enormous difference between those two verbs when the world is on fire. There's a public footnote — a spreadsheet-filter error that pulled in contracts it shouldn't have — and the quiet part is that data-quality provisions of exactly the kind we'd built are what catch that class of error before it becomes a regulatory problem.
From group compliance to a global data operation
The crisis ended; the rulebook it left behind never did. Jurisdiction after jurisdiction, each wanting its own particular cut of the same underlying positions, in its own format, to its own deadline. I was promoted into the investment bank to build a global data capability from scratch and then run it — large-holdings monitoring and regulatory reporting across post-trade, more than ninety regulators worldwide, millions of trades every day. I'd started that decade covering three. Nobody builds a ninety-regulator operation from nothing; you build a three-regulator operation properly, and then find out it can carry weight.
Two results I'd happily put in front of any board. Automated data lineage and metadata frameworks that took validation errors from around twelve per cent to under three — a reduction of roughly three quarters, on reporting where an error isn't a rework ticket somebody picks up next sprint, it's a regulatory event. And governance embedded deeply enough into the daily process to mitigate multi-million-pound exposure in Germany and Sweden — fines avoided invisibly, by controls quietly doing their job before anybody had to draft an apology.
None of it was a solo build. I recruited and grew a global operation of data, governance and engineering specialists across multiple time zones, scaling functions to more than fifty people, with the goal every time of making the process survive the people running it. Including me.
I'm at my best building something from a fragmented or early-stage start. Once it runs well without me, I start looking for the next thing to build.
The whole business, not one specialism
Then the world sent everybody home. Lockdown achieved in about three weeks what cloud business cases had comprehensively failed to achieve in three years, and a firm running its reporting on-premise found out very quickly which of its numbers it couldn't reach from a kitchen table.
A deliberate change of scale and shape — out of a global bank and into a global law firm, from a PLC into an LLP, which is a genuinely different animal in how decisions get made and who has to be persuaded. I've described it as the difference between captaining an oil tanker and captaining a flotilla: at the bank I'd gone deep on one regulated specialism; here the remit was soup to nuts — HR, finance, the legal practices, and a data lab doing machine learning — across sixty-plus offices.
I built the firm's first enterprise-wide data and analytics capability from a standing start — hiring the data scientists, analysts, governance leads and engineers, and running training and data-literacy programmes so lawyers, finance and operations could actually use what we built. A capability nobody can use isn't a capability; it's an expense. I worked alongside Accenture on delivery and Gartner on strategy and market direction, moved the firm off on-premise reporting onto a multi-geo cloud platform, and took the data lab from proof-of-concept to production delivery of an AI-enabled legal analytics platform that cut claims-processing time by roughly eighty per cent while handling internal and client data safely.
Then ChatGPT arrived mid-programme and every partner in the firm suddenly had an opinion about AI — which, I'll admit, made the governance work I'd already done considerably easier to fund. Through all of it I supported the firm's largest merger in a decade: two firms' reporting, governance and data folded together without the numbers ever going dark.
My aim has never been to create more data. It's to make the data you already have more transparent, more enabled, and more used.
Consolidating data functions and wider digital policing strategy
Policing was told, in public and in considerable detail, that it was losing officer hours to process rather than to crime. The productivity review put numbers against it. Answering a problem like that is a data problem before it is ever a policing one — you cannot recover time you cannot measure.
I established a new shared-services data function for Met Business Services: governance, reporting and quality management in one place, with a target operating model that pushed data ownership and controls out into everyday business process rather than parking them with a central team nobody remembers to call. I worked with Deloitte on the wider digital policing strategy it sat inside. I structured the function, recruited into it, and brought staff across through internal transfer — defining the roles, responsibilities and practical standards so it would still be running long after my fixed term ended.
I took it deliberately alongside completing an executive MBA, finishing with distinction. Two kinds of learning running at once, and each one kept the other honest.