AI & Innovation
Regulators, customers, and employees all want to know the same thing: who is accountable when software starts making decisions you used to make alone.
Field notes
The goal is not to be innovative on a slide. The goal is to be explainable in a crisis.
The AI and innovation stream is not a catalog of product launches. It is a running commentary on the governance, documentation, and economic reality of using models in a regulated or safety-critical business. We write for people who are tired of the word ‘responsible’ with no back office to match.
You will read about control design, not model architecture for its own sake: what happens when a model drifts, when you need a human in the loop, and how to evidence that the loop is real. We also write about the boring parts that will sink programmes: data rights, record retention, and the quiet politics of which function owns a decision when the org chart is ambiguous.
Because we are also accountants, we care about the P&L and balance sheet: cap-ex rules for training spend, and how to talk about intangibles when your competitive edge is partly an algorithm. If you are looking for a prompt library, you will not find it here. If you are looking for a defensible file, you will.
Before work begins, we clarify the operating context, governance expectations, and commercial pressures behind the brief. That gives the engagement a clear purpose before technical analysis starts.
The result is a more complete advisory view: what matters now, where risk may surface next, and how recommendations can be implemented without creating unnecessary hand-offs or ambiguity.
Scope
Clarify the decision, deadline, stakeholders, and evidence standard before work begins.
Delivery
Combine partner judgement, technical review, and practical implementation planning in one workstream.
Follow-through
Convert findings into owners, actions, and next steps that leadership can track after the session.
Threads we keep returning to
Regulated AI in FS
Model inventory, model risk, and what supervisors actually ask in practice.
IP and data
Contracts, consents, and the grey zone of synthetic data.
Operating metrics
How to know if an AI bet is compounding or just burning cash.
When we say ‘human in the loop’, we mean a name
Accountability is not a committee. The articles in this stream name the function that owns release decisions, the evidence pack that proves review happened, and the chain that will survive a file inspection. We have seen every fashionable acronym; we have also seen the difference between a process that is designed and a process that is narrated. We write to close that gap.
AI in the enterprise: FAQs
There is, but it depends on your sector and the blast radius of a wrong answer. We publish sector-specific minimums, not a one-size list.
“If you cannot explain a model in a way your audit partner can re-perform, you are not ready to bet the franchise on it.”
Latest in AI Insights
Articles tagged in Sanity for this stream. Add or reorder pieces in the CMS; layouts here rotate by stream for visual variety.
Commission a short briefing for your board
We can stand up a 30-minute read or a 10-slide pack on a cross-cutting topic, with named authors and a clear scope, usually inside two weeks for existing clients and select new relationships.