I learnt to separate accumulated experience from the permission to use its source material.

I joined the fictional Vorendell Data Services in July 1992, shortly after completing applied mathematics at Merravane College of Applied Mathematics. I began as an analyst in a seven-person group, without spending or client-contract authority. The work introduced me to the difference between a dataset that could be processed and an account that could be used responsibly. I learned to ask why the material had been collected and which part of the client's question it could actually answer. Those questions remained useful as my responsibility grew. I wanted the recipient to see the distinction between a method we understood and a conclusion the particular assignment had earned, even when that distinction made the first explanation less convenient.

Project leadership followed in April 1999. By 2005 the invented services firm had US$86 million annual revenue and 420 employees; I led fourteen colleagues through two direct reports with US$620000 annual delivery expense. I was responsible for making the analytical work understandable within an assignment. The company's sales and contracting decisions remained elsewhere. I became interested in what could be carried from one project to another: a method might be useful, while the client records and assumptions behind its first application could not simply become a general company asset.

In June 2006 I became Head of Analytics Delivery. The remit reached thirty-eight colleagues through four direct reports with US$3.1 million annual expense by 2012. Company revenue at that point was US$340 million, with 1650 employees. Delivery leadership exposed the friction between local success and a wider practice. An analysis that satisfied one engagement could depend on definitions, permissions and operating conditions absent in another. I wanted the firm to learn across projects, but I also wanted each proposed reuse to state what had changed rather than borrowing the authority of the earlier result.

My Vice-President of Data Practice appointment began in August 2013. In 2016 I presented a ticket classifier's 96% overall accuracy as though it supported automatic routing of severe incidents. Clara Fenwicke challenged that interpretation. Of 10000 test tickets, 9600 were routine and 400 severe. The classifier correctly identified 9400 routine tickets and 200 severe tickets. The overall count was 9600 correct, but severe recall was only 50%. I withdrew the readiness claim, withheld automatic routing and supported US$650000 review. The mistake was mine: I had allowed a strong aggregate to conceal the missed cases most relevant to the proposed use.

I became CDO in March 2019 and began executive board presentations. A client-exit programme that year covered 340 extracts, with 312 documented closures and twenty-eight awaiting client and legal decisions. I kept the pending material in a separately restricted state rather than describing the whole inventory as closed. A 2020 benchmark proposal then brought together twenty-seven million records from fourteen clients. Eight measured cycle time from order approval and six from submission. We retained two cohorts instead of presenting a single ranking. Neither a large dataset nor a completed inventory count settled the meaning of the result.

The 2023 reuse library examined 600 table definitions. We excluded 190 client-specific definitions from the proposed common set and treated 410 as technical candidates for further review. Seventy-two patterns were subsequently approved through the relevant owners and used in twelve projects by June 2024. The US$3.6 million programme did not establish blanket permission to reuse source values or a financial return. Owen Varleston had argued for a broader launch, and his concern about repeated project effort was reasonable. The selected response kept the review status attached to the pattern rather than treating technical similarity as sufficient authority.

I now lead 146 colleagues through seven direct reports with US$18.7 million delegated 2026 function expense. The latest example examines 60000 synthetic service examples that covered 540 known categories but omitted seventeen specified combinations of rare conditions. Further test work remained necessary. My public account holds no statutory director seat and offers no automatic claim to another institution's oversight role. Every person, employer and result here is fictional. The story demonstrates how an IT-services data leader can explain learning across assignments while keeping purpose, permission and evidential scope visible to the next reader.