The pilot
A six-week, read-only pilot on one agent workflow
Connect one agent runtime and the systems of record it writes to. In six weeks you'll know what the agent did, what each completed workflow cost, and which routing, caps and controls to put in place — measured against success criteria you set before we start.
- Duration
- Six weeks
- Scope
- One workflow · one agent runtime
- Systems
- Two or three systems of record
- Access
- Read-only. No changes to your agents
- Data
- Minimized in your network · EU-hosted
- Commercial
- Paid, credited to the annual contract
- Exit
- Criteria missed? You walk away
What's in scope
Narrow enough to set up quickly. Real enough to decide on.
A tight scope keeps security review short and the baseline clean. Everything runs alongside your agents; nothing sits in front of them.
A single agent workflow with a countable outcome — invoices posted, leads qualified, joiners onboarded, changes merged.
An in-house orchestrator or one vendor agent platform. Sub-agents it delegates to are included.
The systems the agent writes to — for example ERP, mailbox and approvals — read through their audit logs.
Detection, not enforcement. Nothing is blocked, rerouted or changed while the pilot runs.
- Changes to your agents, prompts or workflows
- Enforcement or blocking at any tool boundary
- A proxy in front of your agents
- Raw records or credentials leaving your network
Timeline
Connect, baseline, mandates, decide
Four milestones, two weeks apart. Each one ends with something you can hold — a scorecard, a mandate, a plan.
- Week 0Connect
The collector is installed in your network with read-only access to one agent runtime and two or three systems of record. Success criteria are signed off.
Deliverables- Signed success criteria
- Collector running read-only
- Data-minimization settings reviewed with security
- Week 2Baseline
Every run in the window is reconstructed end to end and linked to its evidence. The first scorecard goes to your sponsor.
Deliverables- Cost per completed workflow
- Straight-through and takeover rates
- Evidence coverage by source
- Week 4Mandates
Draft mandates for each agent are re-checked against the full history. The top exceptions are reviewed with their owners.
Deliverables- Draft mandate per agent
- Exception review with owners
- Audit trail of review decisions
- Week 6Decide
A routing, cap and control plan with projected savings, judged against the criteria you signed in week 0. Then go or no-go.
Deliverables- Routing, cap and control plan
- Projected savings, labeled as modeled
- Go / no-go against the criteria
Deliverables
What you walk away with — whatever you decide
Six artefacts, each traceable to the runs and records behind it. They're yours whether the answer at week 6 is go or no-go.
Cost per completed workflow, straight-through and takeover rates and evidence coverage for every agent in scope — with the metric definitions behind them.
Every run in the pilot window rebuilt end to end, each step linked to its system-of-record evidence with a method and confidence.
A versioned mandate per agent — tools, data, thresholds, spend caps, identity — already re-checked against the full history.
The highest-severity findings reviewed with their owners, each decision kept in an audit trail.
Which tasks go to which agent, where to set spend caps and which controls to enforce first — with projected savings labeled as modeled.
A written recommendation measured against the success criteria you signed off in week 0.
Success criteria
Agreed up front. If we miss them, you walk away.
Before any access is granted, we agree in writing what the pilot has to show — and your sponsor and CFO sign it. The week-6 decision is judged against those criteria, not against a demo.
- Agent spend at equal outputCost per completed workflow, at the same volume and task mixReduction you set
- Control findingsA defined set of exceptions found and reviewed with ownersFindings list agreed
- Evidence coverageShare of agent-claimed actions confirmed in a system of recordThreshold you set
What we need from you
Three people, read-only access and a few hours a week
Chief AI Officer or Head of AI
Owns the decision at week 6 and the agent portfolio it affects.
e.g. Finance Ops lead or VP Engineering
Knows the workflow, joins the weekly check-ins and reviews exceptions with owners.
CFO or finance leadership
Agrees the success criteria up front and signs the go / no-go.
To one agent runtime and two or three systems of record, set up by your IT or platform team in week 0. You hold the credentials.
Typically two to four hours of owner time per week — mostly reviewing exceptions and draft mandates with us.
Security during the pilot
Your data stays in your network. Evidence leaves minimized.
The pilot runs on the same two-plane architecture as production, so your security review covers what you'd deploy.
Security & trustRuns in your environment with read-only access. Your credentials stay with you, and you can revoke access at any time.
Identifiers such as IBANs, invoice numbers and message IDs are hashed; email bodies, subjects and cell values are dropped before export.
Frankfurt by default, with tenant isolation at every layer. Other regions on request.
Security documentation and a data processing agreement are available on request. The DPA is signed before any access is granted.
Commercial
Paid pilot, credited to the annual contract
A paid pilot keeps both sides focused on a decision. If you go ahead, the pilot fee is credited to your annual contract. If the success criteria aren't met, you walk away with no further commitment.
Book a DemoThe pilot rolls into an annual contract with the fee credited. You choose the next workflows and, when you're ready, where to add enforcement at tool boundaries.
You keep the deliverables, remove the collector and revoke access. Data handling after the pilot follows the DPA.
Recommended first workflows
Where pilots start
Each of these has a countable outcome and a system of record with an audit log — the two things a clean baseline needs.
Invoice intake, vendor master and payment agents across email, spreadsheets and ERP.
- Workflow
- AP invoice processing
- Systems of record
- ERP · AP mailbox · approvals
Lead qualification, CRM updates and service agents, measured per resolved case.
- Workflow
- Lead qualification or case handling
- Systems of record
- CRM · CPQ or service desk · mailbox
Onboarding and HRIS agents handling personal data under strict mandates.
- Workflow
- Employee onboarding
- Systems of record
- HRIS · identity provider · IT service desk
Coding agents measured by cost per merged change and revert rate.
- Workflow
- Agent-assisted changes for one to three teams
- Systems of record
- Git host · CI · usage and billing exports
Audit-ready agent evidence for DORA, the EU AI Act and internal audit.
- Workflow
- One operational workflow, e.g. payment exceptions
- Systems of record
- Core system · approvals · case management
If an agent writes to a system with an audit log, we can scope a pilot around it.
Connector availability for your systems is confirmed during pilot scoping.
FAQ
Questions about the pilot
What does the pilot cost?
Do you need write access or changes to our agents?
What happens if the success criteria aren't met?
How much of our team's time does it take?
Which workflow should we start with?
What data leaves our network during the pilot?
Book a Demo
Scope a pilot around your agents.
Book a 30-minute walkthrough. We'll reconstruct a workflow end to end, then agree the scope and success criteria for six weeks on yours.