Built a specialist subsea contractor's central operating layer, Mission Control, as a Lovable + Claude + MCP stack.

A bounded 5-week Phase 1 build replacing fragmented tools with a single platform: Lovable as system of record, Claude as the AI brain, MCP for external services, Terms for governance. Two users (Owner + GM) before team rollout in Phase 2.

Case at a glance— 01

Problem type

Turn AI into operating capability

The mandate

Scope and deliver a central operating platform, data model, ingested historic tenders, pipeline management, voice-based field updates, Claude query layer, inside a fixed 5-week window, with a clean permissions story from day one.

Workstreams

Distinct parts of the intervention

Headline result

5 weeks

Phase 1 delivery

Measured outcomes— 02

5 weeks

Phase 1 delivery

Fixed scope

Discovery-gated build

HubSpot contract retired

Phase 2

Team rollout planned

The situation— 03

Context

What was happening.

A specialist subsea contractor with a small high-value client base and a lean team had data scattered across laptops, no CRM that fit, and a HubSpot contract that did not pay back. The opportunity was to build the foundation properly from scratch rather than layering AI on top of fragmented systems.

Mandate

What had to change.

What changed

The moves that made the difference.

— 04

The intervention, broken into the moves that mattered.

Highlights

5 moves

Reconstructed the client's operating model from a workshop with the leadership team, mapping Key Partners, Activities, Value Props, Customer Segments, Resources and Channels onto a Business Model Canvas, with green-ink baseline vs. red-ink augmentation to flag gaps.
Designed an AI-led operating stack with a clear separation of concerns: a single system of record, an AI layer for query and action, external integrations isolated behind a service boundary, and a central permissions and governance policy.
Sequenced the 5-week plan across Discovery, Design, Build, and Launch & Iterate, with a Discovery gate fixing Must / Could / Won't to control scope creep.
Replaced the HubSpot contract with a pipeline management capability sized to the client base, ingested historic tender and client records, and shipped voice-based meeting capture for field use.
Scoped Phase 2 separately, team-wide rollout, workflow digitisation per individual, and Claude Business integration so every Claude user defaults to Mission Control as context.

Recognise the problem?

Tell me what you are trying to change.

Start a conversation →

Related work— Next

Workday2024–2025

Stood up Workday's AI Centre of Excellence and a new Asia innovation centre.

What it demonstrates

AI capability scales when evaluation, guardrails and delivery patterns become shared infrastructure.

Read the case →

Clarity Diagnostic2025–Present

Fractional CAIO across a portfolio of SMB AI engagements.

What it demonstrates

AI strategy becomes useful when it turns into named work, ownership, controls and sequencing.

Read the case →

View all case studies →