Skip to content
SE
menu

$ method

Internal AI optimization in three phases.

We join to lay the work out properly — not to start coding. Diagnostics come first; engineering is a later, gated phase.

01 /  The phases
  1. Phase 1 Now · ~6 weeks

    Diagnose

    /0.1

    Workflow mapping and time-tracking baselines across delivery lines. Data-landscape mapping. Top ~15 automation candidates scored on annualised hours saved and ease of automation.

    our role: Senior PM to structure workstreams, baselines and decision material.

  2. Phase 2 Following · ~6 weeks

    Design

    /0.2

    Future-state, data-centric agentic workflows with human validation gates. Target agentification per workflow, MVP scoping, success metrics and scale criteria — all before a line of production code.

    our role: PM continuity plus fractional solution architecture.

  3. Phase 3 Gated · from ~month 4

    Build

    /0.3

    Two-week agent build cycles on an off-the-shelf stack (LLM APIs, orchestration, lightweight UI). Every agent has a measured baseline and delta; below ~20% productivity gain it is pivoted or killed. Throwaway before scale.

    our role: Senior engineering squads, opened only when the gates are passed.

Phase 1 tooling

The diagnostic runs on Baseline — our platform, deployed inside your estate in week one.

See the platform →
02 /  How we measure

baseline → delta → decision

"Metrics that work" is a deliverable, not a slide. The measurement mechanism is built before the build.

01 baseline

Time-tracked, per workflow, before any change.

02 delta

Measured gain after the two-week prototype and limited deployment.

03 decision

Below ~20% productivity gain, the agent is pivoted or killed. Above it, scale.

rule: if delta < ~20% → kill / pivot · else → scale
03 /  Design principles

Off-the-shelf first

LLM APIs, standard orchestration, vector stores. Custom only where curated data creates real defensibility.

Human-in-the-loop

Every workflow keeps a validation gate owned by a domain expert. The expert stays accountable.

Measured per agent

Each agent carries its own baseline and delta. No aggregate hand-waving.

Internal wins first

Prove value inside the organisation before anything gets productized for the market.

No AI demos, no platform commitments ahead of validated demand. Measurable workflow improvements only — each one either scales or is killed.
operating rule

See the method run on your workflows.

A paid two-week discovery sprint puts a senior PM inside your live diagnostic.

Book an intro call