Our AI tools
We advise regulated firms on how GenAI can transform their business. We started with ourselves.
We knew our use of AI had to be explainable, secure, well-governed, and built for regulated work – everything our clients would demand of us.
There was nothing on the market that fit the bill, so we built our own: two platforms, Workbench and Fleming, which are used across the business every day.
They are redefining how we work and what we can do: they’re enabling our teams to be smaller and more expert, and our discovery to go deeper.
Workbench
Workbench is the AI tool our whole team uses, every day.
Most organisations are still at the scattered-licences stage of AI: a few people with a chat interface, working independently of their colleagues. Workbench is what comes next.
It’s much more than an internal chatbot or a wrapper for an LLM. We’ve built a deeply-integrated platform that opens up our knowledge estate to our team, with an intentionally-designed governance layer that ensures every element is under our control.
It connects the knowledge spread across our business – in people's heads, in past projects, and in the systems we run on – and makes it usable at the point of need. Purpose-built skills capture good practice once and apply it everywhere. Live connections into our own systems – projects, emails, diaries, meeting transcripts – let AI act on real work.
Workbench’s governance layer is what turns it from a productivity toy into a serious business platform. We call it the ‘control tower’.
Workbench lets us monitor the health of every AI skill in use: who owns it, when it was last reviewed, whether two skills are about to collide, how each performs against measured standards. It lets us pin skills to the model they’ve been tested with and approved, rather than applying them to each new frontier model and hoping for the best.
It’s also designed for total control over where the data goes. The platform is not tied to any vendor, so we can ensure European data residency, and safely adopt frontier models faster than the firms we advise.
This is the part most financial institutions haven't built for themselves yet: central, auditable oversight of AI in daily use, rather than a policy document and a hope that nobody’s pasting client data into a US-based consumer tool.
Fleming
Fleming is our in-house discovery and research platform. When our teams need to understand how a business really works, Fleming does the heavy lifting.
It reads across an entire project at once – documents, internal notes, survey responses, meeting transcripts – and surfaces the patterns no single interview would reveal. Every finding, and every step taken to reach it, can be traced back to the source: who said it, where, and when. That traceability means our conclusions stand up in front of an executive committee, and it’s how we can go deeper on discovery and research every time.
Fleming is not a wrapper for an LLM with a clever prompt. Beneath the surface sits a deep, self-improving reasoning system. Clients tell us it’s not something they’d expect a design consultancy to have built.
Williamina Fleming (1857-1911)
Quality that’s measured, never assumed
A research platform is only as good as its worst answer, so Fleming grades every one. An automated judge scores each answer
for accuracy, for how well it’s grounded in the source material, and for staying in scope.
The judge itself is validated against our own experts, with the bar held by our Head of Research, Dr Alexa Haynes, to the same
standard she expects of her team.
And because the models keep changing, Fleming re-tests itself: when its agreement with our experts starts to drift, it recalibrates
until it passes.
The result is that a junior researcher can produce work close to the standard of our PhD-led team, so we field smaller teams that
reach better answers sooner. The groundwork is Fleming’s. The final call is always a person’s.
Fleming also features a powerful baselining tool.
When we need a picture of how a whole organisation really works, Fleming’s baselining tool gathers the evidence. Instead of interviewing a handful of people and inferring the rest, it lets us hold a structured, AI-led conversation with everyone – by voice or text, activity by activity – capturing what people actually do and how they do it.
Each answer builds a live picture of the business as the conversation goes, which the person can see and correct on the spot, and anything unusual is flagged for an expert to review. That produces three views of the same business, held at once: what the documents say, what leadership believes, and what people actually do. Most businesses never get to see where the three disagree, and that’s where the interesting and valuable work is.
It lets us deliver qualitative research at the depth of an interview, across a whole organisation, at a scale traditional interviews could never reach.
But the real value isn’t the map, it’s what that map lets you decide. Fleming shows where work can safely be automated, and – just as importantly – where human judgement has to be protected as more of the work becomes agentic. It shows who your people need to become in that future: the systems thinkers, the builders, and the governors a well-run AI operation depends on.
That’s the hardest question a regulated business faces right now. It's the one Fleming is built to answer
Built to the standard our clients expect
Fleming and Workbench both run on private models from Anthropic, inside infrastructure we control, with European data residency. That means sensitive client data is handled the way a regulated firm requires. Because we built them, we're not locked to any one vendor's platform or roadmap, and we control exactly what they can see, how they behave, and how we prove it.