Services

Governance comes before automation

AI enters a company through whoever adopts it first, and by the time anyone asks who approved what, the answer no longer exists. We start from the other end: decision rights, risk classes and a verifiable trail first; automation second.

Request the diagnostic

Four engagements, with the duration stated

You come in through the diagnostic and go further only if the numbers hold. No subscription to software of ours: what stays in the company runs on your own tools.

011-2 weeks

AI Operations Diagnostic

Where AI can take real work off your back office, what it must never touch, and what the change is worth.

Manufacturing SMEs are told that AI will change their operations, and never told where, or at what price. The diagnostic works on the real back office and separates three things: the work AI can take today, the work it can take once the process is cleaned up, and the work it must never be allowed near. It ends with a first draft of the decision-rights table and a business case with estimated cost and return, so the next decision is an investment decision and not an act of faith.

What gets delivered

Prioritised map of the back-office work AI can take over
Exclusion list: what AI must never touch
Decision-rights table, first draft
Business case with estimated cost and return
Findings presentation for ownership and management

024-6 weeks

AI Operating Model & Governance

The decision rights, risk classes and approval levels that let AI run operations under ISO 9001, ISO 55001 and the EU AI Act.

We build the operating model before the automation: who decides what, which uses are high risk and which are routine, which actions require a human approval and at what level, and what trace every action leaves so an auditor or a customer can reconstruct it afterwards. Everything is written against the standards the company is already measured on (ISO 9001, and ISO 55001 where assets matter) and against the EU AI Act obligations that reach a manufacturer through its own products and its customers. It closes with training for the leadership group: governance nobody at the top understands is governance nobody enforces.

What gets delivered

Decision-rights table, approved
AI risk classification model
Approval levels and escalation rules
Traceability and audit requirements
ISO 9001, ISO 55001 and EU AI Act alignment map
Leadership training sessions

034-8 weeks

Single-Process Pilot

One back-office process in production with agents, approval gates and a verifiable log, built on the tools you already own.

A pilot is where governance stops being a document. One back-office process is taken end to end (a quotation flow, order handling, a technical documentation chain, supplier communication, a quality non-conformity) and rebuilt with agents doing the repetitive work, human approval gates where the decision-rights table says a person decides, and a log that records what was done, by what, and on whose approval. It is built on the systems you already have: the ERP, the mail, the file storage, the assistants already licensed. We install no software of ours and there is nothing to subscribe to afterwards. The engagement ends with a runbook and with your own people running the process without us.

What gets delivered

One back-office process in production, end to end
Agent configuration on the tools you already own
Human approval gates per the decision-rights table
Verifiable action log
Process runbook
Training for your operators

04Monthly, continuing engagement

AI Operations Retainer

Keeping AI operations governed month after month: continuous oversight, incremental automation, an escalation path and a quarterly review.

What is governed in a project drifts once the project ends: models change, people change, a supplier changes a format and an agent quietly starts producing nonsense. The retainer keeps the operating model current, extends automation one process at a time instead of through another large programme, gives you an escalation path with a name on it when an agent misbehaves or an approval gate blocks the business, and forces a quarterly review where the numbers, the incidents and the next candidates are put on the table. It is the only retainer we keep on the price list.

What gets delivered

Continuous maintenance of the operating model
Incremental automation of the next processes
Escalation path with a named owner
Quarterly review with ownership and management

Where it applies

Governance and decision rights

Who decides what, which AI uses are high risk and which are routine, which actions require a human approval and at what level. Written against the standards the company is already measured on.

What it covers

Decision-rights table: who signs what
AI risk classification
Approval levels and escalation rules
Traceability and audit requirements
Alignment with ISO 9001, ISO 55001 and the EU AI Act

Back-office processes

The administrative work that holds a manufacturing SME together and that nobody ever designed: quotations, order handling, technical documentation, supplier communication, quality non-conformities.

What it covers

Prioritised map of the work AI can take over
Exclusion list: what it must never touch
Process redesign before automation
Business case with estimated cost and return

Agents on your own systems

One process in production with agents doing the repetitive work, human approvals where the table says a person decides, and an action log. On the systems you already have, with no software of ours to install.

What it covers

Agent configuration on your own tools
Human approval gates on the steps that matter
Verifiable action log
Runbook and training for your operators
Training on common AI assistants: effective use, limits and confidentiality

The path we propose, in five steps

These are the steps that, in our experience, carry an engagement through. You enter where it makes sense and stop when the objective is met; if you have already covered a step, we start from the next one.

Diagnostic

Where the margin is and where it is not. Current state, work that can be taken away, work that must be left where it is, interventions scored for impact and feasibility, and numbers that have to hold.

Prioritised opportunity map
Exclusion list
Business case

Architecture

The system, not the patch. Processes, data model, roles, decision authority and the points where a person must be able to say no.

Processes per ISO and BPMN 2.0
Decision-rights table
Risk classes and approval gates

Integration

Building and connecting to the systems you already have. Automations and, where they earn it, agents enter here, inside processes that by now genuinely exist.

Automations in operation
Documented integrations
Testing and validation

Operation

The part usually missing. Monitoring, spend ceilings, approvals that cannot be forged, and a way to notice when something is stuck.

Monitoring and alerts
Approval log
Cost dashboard

Enablement

Your team runs the process without us. Training on the tools you will actually use (your own ERP and the assistants you license), documentation and a periodic review for as long as it helps.

Training materials
Runbook
Scheduled reviews

One case, in full

A process redesign and operational excellence case, not a demonstration of agents: what was done, what came out of it, and what was out of scope.

Operational excellence · Process redesign

Marcos Felix

Redesign of the core operational processes of a technical services organisation: from fragmented workflows to structured, measurable operations.

Operational excellenceProcess designKPIsChange management

We mapped the current state of three core processes, identified waste and failure points, redesigned each process with clear ownership and decision points, defined KPIs with practical measurement methods, and produced implementation plans that respected the organisation’s culture and capacity for change.

Outputs

Current-state maps for 3 core operational processes
Target-state process designs (BPMN 2.0)
Standard operating procedures for each process
KPI framework with 12 operational metrics
Implementation plan with phased rollout

Not sure where to start?

It is the right question, and the answer is the diagnostic: one or two weeks to know where AI pays in your back office, where it would be reckless, and what the first step costs.

Request the diagnostic How we work