Start with the problem you already have.

Leadership wants AI, but nobody can say which processes are stable enough to automate.

Which processes are well understood, which are disputed, and where data quality blocks automation.

An evidence-backed readiness map with prioritised, de-risked opportunities.

Documented processes no longer reflect how work is done.

The real sequence, actors, systems, workarounds and exceptions.

A living process model with status on every step.

Delays and rework are felt everywhere but owned by no one.

Where time is lost, where information is re-keyed, where decisions stall.

Diagnosed root causes and targeted interventions.

Systems were adopted one at a time and never designed to connect.

Which objects move between systems, how, and with which identifiers.

An integration map with linkage evidence and gaps.

A major platform change risks replicating today's problems in new software.

Current truth, exceptions and dependencies before requirements are written.

Requirements grounded in verified current state.

Automations are being built in isolation and break unexpectedly.

Shared patterns and where automation should, and should not, sit.

A coherent automation architecture traced to objectives.

Agents need context and permissions the organisation hasn't defined.

The business objects, operations and controls agents will need.

A governed Company MCP specification.

Two organisations, two ways of working, one target model to agree.

Where processes genuinely differ and where they only appear to.

Side-by-side current state with explicit contradictions.

Critical know-how lives with a few long-serving people.

Which decisions and workarounds depend on individual memory.

Captured, validated knowledge that outlasts any one person.

Improvement programmes lose momentum once the workshop ends.

Whether changes held, and where new drift is appearing.

A model that stays current, with monitors on what matters.