$0M
Internal efficiency delivered
0
Hours saved per year, projected
$0M
Annual RTB and CTB owned
0
Users rolled out to
0k
Records under management
0
Enterprise clients on platform
Titles for this are still settling down, so here are the areas instead. Hover a card, or tap it, for what I have done in each.
Head of Forward Deployed Engineering, FDE Lead, Solutions Engineering Lead
Technical Director, Field CTO, Principal Consultant
AI Practice Lead, Head of Applied AI, AI Delivery Lead
Head of AI Product, Principal Product Manager
Principal Engineer, Founding Engineer
Side projects and experiments rather than products with customers. The engineering decisions in them are real, and they are the part worth talking about.
Chunking, embedding and serving a user's own history back as context. The retrieval layer is written directly against Vectorize rather than pulled in from a framework, so I had to understand how it breaks rather than trusting that it would not.
Built in Looper
A small fast model decides what a capture is before anything larger sees it. That is a cost and latency decision as much as a quality one, It is also the reason the running cost does not climb straight up with usage.
Built in Looper
Segment Anything running in the browser through transformers.js, so a greenkeeper traces a fairway with one tap and the image never leaves the device. No inference bill, no upload, no wait.
Built in Course Map
Postgres row level security, so a club could only ever reach its own course. I wanted the isolation to sit in the database rather than in a filter somewhere in application code that someone forgets to apply.
Built in Course Map, Tee Time Traveler
Satellite imagery and LiDAR in, usable course geometry out. Multispectral GeoTIFF processed into NDVI overlays by an edge function, elevation pulled per course rather than shipped with the app.
Built in Course Map
A terrain mesh built from elevation data with the aerial imagery draped over it, rendered with react-three-fiber. Getting this to look deliberate rather than accidental took longer than getting it to work.
Built in Course Map
Typed end to end, migrations in version control, tests in CI. Nobody was making me do any of that on a personal project, which is the point. It is the habit I would bring to something that did have users.
Built in Everything here
I have led delivery inside a global consultancy and inside a company with two people in it. This is roughly how I would want to run a new team.
In enterprise the quickest way to lose users is to go quiet. I answer on our product forum as close to instantly as I can manage, every day, and my team does the same. Reading the sentiment across it tells me more than a workshop does, because that is where people say what they think when nobody senior is on the call.
A few loud voices are not a mandate to reverse. I will make space for the discussion and explain the reasoning as many times as it takes. But changing direction because three people are annoyed teaches everybody that volume works.
I am poor at notes. I know it, so I do not pretend otherwise and I do not try to be the person in the room capturing everything. I trust my business analysts implicitly. My job is to give them the context to be right, not to duplicate them.
I look for people who want to innovate and genuinely enjoy the technical side, rather than people waiting to be told what the ticket says. Then I give them room to use it. I run a second board carrying deliberately innovative stories, with an incentive to pick them up, which costs almost nothing and stops proof-of-concept work losing every argument to a backlog of defects. With AI shortening the build we estimate around 40% better story completion, so a $1.5M run budget now buys more than firefighting.
I will have the awkward conversation, and I will have it calmly. Emotion in those meetings is usually a way of avoiding the specifics. I am also happy to open a terminal myself, which helps when we are working out whether something is genuinely hard or just unfamiliar.
2022 to now
Mercer/Hybrid, UK
Global Benefits Management
Own the strategy, the backlog and the budget for a B2B benefits platform holding 100,000 records across 500 clients, and steer the engineering team that builds it.
$3M
efficiency delivered
2019 to 2022
Mercer/London
Global Benefits Management
Strategic account lead for 12 multinational clients, and the person a global network of brokers answered to when something went wrong.
2,000
plans in flight
2015 to 2019
Aon/Krakow, Poland
EMEA Flexible Benefits
Moved an entire client service function from the UK to Poland without dropping a single SLA. Around 160 EMEA clients, none of whom noticed.
160
clients served
2011 to 2015
Aon/Sheffield and Farnborough
Pensions Administration
Administrator to Team Manager in three years, with a team reporting to me. Everything I know about how a business runs day to day started here.
3 yrs
admin to manager
I took the product in late 2022, after three quarters of falling usage. It now runs at about six times that level, and close to five times its previous best.
Weekly platform events, indexed to the quarter of handover. The absolute figures belong to my employer, so this shows the multiple and nothing else.
Town halls in person with the key user groups, and a reset of how defects were handled. What counts as a bug, what counts as an enhancement, and who decides.
Went straight at a priority I already knew from the business side, getting users off the Excel workarounds they had built to cope.
A full-scale rebuild of the data entry processes. This is where the $3M in internal efficiency and the 35% lift in data completeness came from.
Heavy training across the user base, then retiring the legacy system it replaced.
Chat layers and document ingestion into the product itself, rather than alongside it.
Fifteen years of watching the same things go wrong. I am easy to work with about most of it, and hard to shift on these four.
I spent years in benefits administration watching business-critical data sit in workbooks with no validation, no audit trail and no owner. Someone sorts one column without the others and nobody finds out for a quarter. The process rebuild I led was worth about $3M in efficiency and lifted data completeness by 35%, and most of that was moving decisions out of spreadsheets and into something that could enforce a rule.
That is not a licence to overrule them. It means the job is establishing a collective understanding of the actual problem, and getting there takes longer than writing the spec does. I have run in-person workshops for exactly that reason and delivered 30+ rollout sessions off the back of them. The useful feedback never came from the first meeting. It is also how you notice that twenty separate requests are really one configurable feature.
The incentive in most product roles runs the wrong way. Hard-code it, and every future change needs a developer, and needing developers is how you justify your funding next year. I took an inherited database and made it admin-configurable so the people using it could serve themselves. It standardised data capture across clients and saved around $200K a year, and it reduced the work my own team could bill for. I would do it again. The alternative is charging the business for a ticket every time somebody wants a dropdown changed.
I am usually the person arguing for exploration, not the one blocking it. But a backlog full of defects wins every argument it is allowed to have, so the experimental work needs somewhere of its own to live, with a budget, a boundary and a date to report back. Left open ended you get a demo that impresses a room once. Left unprotected you get nothing at all.
An interview runs both ways. These are the six I would actually bring, and what I would be listening for.
What is running on a spreadsheet right now that should not be?
Listening forI would end up accountable for that risk, so I would rather see it early.
If I try something and it does not work, what actually happens to me?
Listening forInnovation dies quietly in places where a failed experiment is career damage.
How do you know your AI features still work six months after launch?
Listening forIf nobody can answer this, there are no evals.
When engineering and the business disagree, who decides?
Listening forIt will happen. I want to know where I sit when it does.
How close does the team actually get to the people using this?
Listening forDistance from users shows up in the roadmap within a quarter.
What do you want in twelve months that you have not got now?
Listening forIt tells me how much of this is already decided.
Next
Forward deployed engineering, or a product role with real build responsibility. If the job needs someone who can sit with the client and then go and write the code, that is the one I want.
So there is a version of my CV written for one. One column, no tables, no images, no headers or footers to trip over, and the dates in a consistent format. Paste the URL straight into a model if that is easier than downloading it.
