AI Worklab 90

Ninety days from a work map to team capability

A joint working programme for companies that want more than a demo: their own safe way of working with AI.

Send a private application →See the 3 stages ↓
3 working stages

Every thirty days produces evidence and a decision

Ninety days is a frame for working together. Every stage can end in continuation, a scope change, or a stop decision.

01Day 1–30

Map the work and select

The team builds a shared view of where recurring work occurs and where human judgement must remain.

Activities
  • Interviews with the owner and people who perform the workflow.
  • Mapping inputs, systems, exceptions, and decision points.
  • Selecting one workflow by value, feasibility, and risk.
Outputs
  • A prioritised work map.
  • A workflow brief, baseline, and risk register.
Decision gate
Does the selected workflow have an owner, a safe test sample, and a measurable outcome?
02Day 31–60

Prototype and shadow mode

A narrow prototype works on real inputs without performing a production action.

Activities
  • Designing rules, permissions, and stop conditions.
  • Implementing the prototype in a restricted context.
  • Domain-reviewer scoring and collection of failure patterns.
Outputs
  • A restricted prototype with a working trace.
  • An evaluation scorecard and revised safety boundaries.
Decision gate
Is quality sufficient for a controlled pilot, or should the flow change or stop?
03Day 61–90

Controlled pilot and ownership

Validated steps are tested with clear approval, monitoring, and an owner on the team side.

Activities
  • Piloting the agreed scope and tracking metrics.
  • Rehearsing exceptions, incidents, stop conditions, and rollback.
  • Hands-on enablement for the owner and reviewers.
Outputs
  • An operating playbook and monitoring plan.
  • A decision to continue, iterate, or stop.
Decision gate
Can the team own the workflow safely, and has it decided what comes next?
Tangible deliverables

The programme does not end with workshop notes

The exact content adapts to the selected workflow; each artefact must be usable by its owner.

01

Work map

Prioritised workflows, systems, decision points, and owners.

02

Baseline and risk register

Current state, evaluation sample, sensitive data, and stop conditions.

03

Restricted prototype

A narrow workflow with least-privilege access and no hidden autonomy.

04

Evaluation scorecard

Agreed criteria, reviewer corrections, and failure patterns.

05

Operating playbook

Approval, monitoring, incident response, rollback, and change control.

06

Enabled owner

A named person understands the boundaries, metric, and stop decision.

Joint team

Internal ownership is built from day one

Programme sponsor

Removes organisational obstacles and keeps the programme connected to a real business need.

Workflow owner and reviewers

Bring work context, evaluate outputs, and take ownership of day-to-day decisions.

InovatorAI

Facilitates mapping, builds the prototype, sets up measurement, and prepares a safe handover.

Model starters

Three examples that can start the conversation

These are illustrative scenarios, not client outcomes or a promise that the same flow will suit your company.

Model scenario01 · Request intake

Requests ready for a decision

The agent reads the input, adds approved context, and prepares a draft of the next step.

  1. Email or form
  2. Category, owner, and context
  3. Draft waits for a person
Model scenario02 · Finance

Invoice discrepancies under control

The workflow extracts fields, compares them with the order, and flags a discrepancy for review.

  1. PDF or email
  2. Extraction and validation rules
  3. Exception goes to approval
Model scenario03 · Internal knowledge

Internal answers with traceable sources

The agent finds an answer in approved documents and attaches its source for verification.

  1. Employee question
  2. Approved knowledge retrieval
  3. Answer with a source
See all 9 model scenarios →
Safety boundaries

Adoption does not come from bypassing control

Boundaries are written into the workflow design, tested during the programme, and taken over by a named owner.

No production action first

Real inputs are evaluated in shadow mode before any production action.

Least privilege

The workflow receives only the data and tools needed for its approved scope.

A person at sensitive gates

Uncertain, regulated, or irreversible steps do not proceed without accountable approval.

Trace and a path back

Changes, outputs, and approvals are traceable; the owner can stop or roll back the flow.

FAQ

Questions about AI Worklab 90

Are ninety days a guarantee of an outcome?

No. It is a working frame with three gate decisions, not a guarantee of specific ROI or an obligation to proceed to deployment.

How many people need to take part?

The programme needs a sponsor, the selected workflow owner, and a small reviewer group familiar with real cases.

Do we need to expose all company data?

No. We begin with the smallest safe sample and expand access only when necessary and approved.

What remains with the team after the programme?

Tangible artefacts, an enabled owner, and a decision to operate, iterate, or stop the workflow.

Start with work context, not a tool purchase

In the private application, name the team and recurring workflow. Together we will assess whether the 90-day frame is suitable.

Send a private application →Back to the stages ↑