Innovation
Technology as operating leverage.
We treat software and automation as shared infrastructure across the companies we own — useful in proportion to the manual work it removes, not to how it sounds.
Our position
A deliberately unexciting view.
There is a version of this page that promises a proprietary platform, an AI engine, and a transformation programme. We are not going to write it, because the honest version is more useful.
Most of the companies we are interested in are not held back by a lack of advanced technology. They are held back by work that a computer should have been doing for a decade: data re-keyed between systems, reports assembled by hand, decisions waiting on a spreadsheet only one person knows how to update.
Fixing that is not glamorous and it is not novel. It is also, reliably, where the return is. Our starting position in any company we own is to look for that work and remove it before considering anything more ambitious.
Focus areas
How we decide what to build.
These principles describe our approach to technology decisions across the group. They are how we evaluate a proposal, not a catalogue of systems currently in production.
Infrastructure, not a product line
Technology exists here to make the companies we own work better. We are not building software to sell to the market on the strength of a roadmap.
Automate the boring part first
The highest-value automation is almost never the impressive one. It is the recurring manual task somebody has quietly done every week for years.
Decisions need better inputs
Most operating problems are information problems. Shortening the distance between a question and a trustworthy answer changes what a team is able to do.
Buy before building
Existing software is usually cheaper and better than what we would write. We build only where nothing adequate exists or where the workflow is genuinely ours.
AI where it earns its place
Language models are useful for specific, bounded tasks — summarising, classifying, drafting, extracting. We apply them where the failure mode is acceptable and a human still checks the result.
Systems outlive enthusiasm
Anything we build has to keep working after the person who was excited about it moves on. Boring, documented, and maintained beats clever.
On artificial intelligence
Useful, bounded, and supervised.
We use language models the way we use any other tool: where the task is well defined, the cost of being wrong is manageable, and someone competent reviews the output before it matters.
That rules out a lot of what gets marketed. It leaves a genuinely useful set of applications — summarising long documents, drafting a first version of routine correspondence, classifying and routing incoming work, pulling structure out of unstructured records.
Where we would not put a model is anywhere its confident errors reach a customer or a financial record without a human in between. That is not caution about the technology so much as ordinary operating discipline.
Get in touch
Building something in this space?
We are interested in operators who have already seen a specific problem up close — particularly one they have tried and failed to solve with off-the-shelf tools.