Systems, not seats.
Six builds, one ladder.
We build a small number of AI systems extremely well. Each moves a named metric, ships with an eval suite and approval gates, and runs on your infrastructure. 95% of AI pilots never reach the P&L — MIT, 2025 — because the work stops at the model. Ours stops when the metric moves.
How we engage
Diagnostic Sprint
Fixed fee · 1–2 weeks
We map where your operation leaks and what each fix is worth. You leave with a ranked build plan — yours regardless of what you decide next.
Build Engagement
Fixed scope · 4–8 weeks
We build the system, with an eval suite and human approval gates at every step. It ships into your operation, documented, and you own it.
Operating Partnership
Monthly retainer
Monitoring, evals, model upgrades, and a new build each quarter. The governance layer that keeps a system alive past launch — where over 40% of agentic projects are projected to fail (Gartner).
The builds
Document Intelligence Core
Invoice capture, 3-way match, AP reconciliation, and order-entry read straight from email and PDF into your ERP.
Built to move: Cost per document processed
Front-Office Resolution Agent
Shared-inbox triage, draft, route, and resolve with approval gates — the logistics WISMO machine. Identifies itself as AI.
Built to move: First-response time · % auto-resolved
Quote Engine
Parse the RFQ, pull specs and pricing, and draft a quote for sign-off in hours instead of days.
Built to move: Quote turnaround · win-rate on speed
System Glue
Agent middleware across legacy ERP, Excel, and email that kills re-keying. The anti-ERP play — your tools stay.
Built to move: Re-keying hours eliminated · error rate
Decision Layer
An automated weekly ops brief, month-end narrative, and live KPI panel assembled from your source systems.
Built to move: Reporting lag
Knowledge Vault
Capture a retiring expert's knowledge into a queryable internal assistant before it walks out the door.
Built to move: Onboarding time · single-person dependency
Two ways in
Pilot Rescue
A fixed-fee teardown and relaunch of a stalled internal AI project. 95% of pilots never reach the P&L (MIT, 2025); external builds deploy roughly 2× as often. We find why yours stalled and ship the part that lands.
AI Act Readiness Sprint
Inventory and risk-classify your AI, implement Article 50 disclosure and labeling, prepare deployer documentation, and run staff AI-literacy. The Act applies from 2 August 2026; exposure runs to €35M or 7% of global turnover.
What we don't sell
Refusal is a strategy. We say no to the work that doesn't move a P&L metric:
- Hourly staff augmentation
- Commodity website chatbots
- Marketing-content AI — budgets already over-favour marketing; the ROI is in operations (MIT, 2025)
- One-off bespoke builds outside the six, in year one
Founding clients: we carry the risk.
Pricing tied to one named P&L metric. You keep the data. The first three engagements only.