WORK

THE WORK, SHOWN WHOLE

Depth, not volume. Each engagement below is told whole — the situation, the friction, what was built, and what the client now runs without us.

Operating model

Anonymised engagement pattern

From AI Pilots to an Executive Operating Model

Context

A leadership team had several AI initiatives, each moving at its own pace. Nobody could say which decisions the company wanted AI to change first.

Challenge

The tools were not the problem. Every initiative gave a different answer to the same three questions: who owns this, what is it allowed to decide, and who checks it.

Intervention

We mapped the decisions that carried the most weight, gave each one an owner and a limit, and sequenced the first moves — starting where AI could assist a decision without owning it.

Deliverables

An executive AI operating model, signed off as one document
A decision map: every decision, its owner, its limit
A governance rhythm and a sequenced adoption plan

Impact

Conversations about AI stopped being about tools and started being about decisions. The leadership team now argues from the same map — and pilots have somewhere to graduate to.

What stays with the client

The model itself. When a new AI use case appears, the team can place it, assign it, and bound it without calling us.

Systems

Anonymised engagement pattern

An Agentic Prototype with Explicit Boundaries

Context

An operations team spent much of its week coordinating: chasing status, moving data between tools, deciding what escalates. That this was automatable was obvious. Which parts, safely, was not.

Challenge

Automating the wrong step would be worse than automating nothing. The team needed a line between work an agent could own and work where a human had to stay in the loop — drawn before building, not after an incident.

Intervention

We walked the workflow step by step, scored each handoff for how much autonomy it could bear, and designed the control model first: what the agent decides alone, what it proposes, what it must escalate. Only then did we scope the prototype.

Deliverables

A workflow map scored for automation readiness
A prototype design with explicit agent boundaries
An escalation model: who gets alerted, when, with what context

Impact

A broad ambition to automate operations became one bounded prototype with a named owner, an escalation path, and a list of evidence to produce before expanding.

What stays with the client

The evaluation method. Every future automation candidate gets the same test: score the autonomy, draw the boundary, then build.

Capability

Anonymised engagement pattern

A Capability System That Outlasts the Training

Context

Everyone had access to AI tools. Usage told a different story: a few enthusiasts ran ahead while most usage never became routine, and leadership literacy lagged behind.

Challenge

Training on its own fades. Adoption had to be built into how teams already work — with leaders able to question an AI result, not just approve it.

Intervention

We built the capability system around live work: sessions run on the team's real tasks, playbooks written from what actually held up, and a recurring rhythm where teams compare results and revise their methods.

Deliverables

Leadership sessions built on the team's own decisions
Playbooks written from live work, not generic prompts
A recurring rhythm for sharing results and revising practice

Impact

AI use stopped depending on individual enthusiasm. Teams now share a method, a place to improve it, and managers who know what to ask of an AI-assisted result.

What stays with the client

The rhythm and the playbooks — a system that keeps improving after we leave, because improving it is now part of the work.

Start with the decision that's stuck.

Contact