Plan a safe pilot

Adoption / governed from day one

Safe AI startsbefore the first
connector.

Startwell AI gives your IT team a controlled route from “should we?” to “this is how we run it”—with permissions, evidence and human ownership at every gate.

See the adoption path
DEPLOYMENT / PRE-CONNECTORGATE 00
00
CURRENT STATENothing connected

Startwell has no access to company data.

✓

Deployment ownerNamed and accountable

1

Security baselineWaiting for IT

2

Data boundaryNot yet defined

3

Connector approvalLocked

DEFAULT: DENYAccess is granted connector by connector, purpose by purpose.

READINESS FIRSTLEAST PRIVILEGEEVIDENCE ALWAYS
002 / THE SIX GATES

Access is earned.One gate at a time.

The first workshop does not connect a thing. It establishes whether the organisation, the use case and the environment are ready.

00
DISCOVER

Establish the security baseline.

+

Architecture, accreditations, policies, network boundaries, identity provider, data locations and accountable owners.

OUTPUTReadiness record + decision to proceed
01
BOUND

Define what AI may know and do.

+

Agree the first use case, permitted data classes, prohibited data, retention, model boundary and measurable success criteria.

OUTPUTApproved scope + data flow map
02
CONNECT

Approve each connector separately.

+

Read access before write access. Minimum scopes, service accounts, owner approval and a documented purpose for every system touched.

OUTPUTConnector register + access matrix
03
PILOT

Prove it inside a controlled group.

+

Begin with synthetic, historical or low-risk data where practical. Test answer quality, failure modes, permissions and human approval.

OUTPUTTest evidence + exception log
04
ADOPT

Train the team on their real work.

+

Role-based onboarding, clear listening indicators, safe-use rules, feedback routes and named champions. Adoption is measured, not assumed.

OUTPUTPlaybooks + adoption measures
05
EXPAND

Open the next boundary deliberately.

+

Only after the first scope is safe and useful do we add a new team, connector, action or model. Every expansion returns through the gates.

OUTPUTGo / hold / rollback decision
003 / THE CONNECTOR GATE

Connected does not meanuncontrolled.

Every connector has an owner, a purpose, a minimum permission set and a revocation route. Nothing inherits trust merely because it is available.

OMicrosoft OutlookEmail + calendar · read onlyAPPROVED
XXeroInvoices + bills · read onlyAPPROVED
TMicrosoft TeamsMeetings only · owner reviewREVIEW
SSharePointApproved sites only · no access yetLOCKED
MCP / API REQUEST
POLICY GATECan this user
do this action?
  • Identity
  • Purpose
  • Scope
  • Permission
SOURCE-LINKED RESPONSE
DEDICATED PRIVATE COMPUTE
NVIDIADGX SPARK

Company neural net

Models, memory and permissions stay inside the approved boundary.

004 / GUARDRAILS THAT OPERATE

Not a policy PDF.Controls in the path.

The safe route should also be the easiest route. Controls are built into how people ask, approve and act.

01

Default deny

No system, dataset or action is available until an owner approves it.

02

Least privilege

Access is limited to the smallest scope needed for the agreed job.

03

Read before write

New connectors begin read-only. Actions are a separate approval.

04

Human checkpoints

Material actions wait for the right person—not merely any person—to approve.

05

Traceable answers

Users can see the source, time and permission context behind important outputs.

06

Stop and revoke

IT can suspend a connector, user, action or the service without dismantling the system.

005 / THE ASSURANCE PACK

Trust should leavean audit trail.

Your IT team receives the evidence needed to operate, review and challenge the deployment after Startwell leaves.

STARTWELL AI

Deployment
assurance pack

Controlled copy / example structure

01—08
  1. 01

    System architectureComponents, boundaries and deployment topology

  2. 02

    Data flow and retention mapWhat moves, where it is processed and for how long

  3. 03

    Connector registerOwner, purpose, scopes, credentials and revocation

  4. 04

    Role and access matrixUsers, teams, memory boundaries and approvers

  5. 05

    Risk and exception logKnown limitations, decisions and accountable owners

  6. 06

    Test and evaluation resultsPermission, failure mode and answer-quality evidence

  7. 07

    Operations runbookMonitoring, update, incident and shutdown procedures

  8. 08

    Adoption measuresUsage, value, confidence and support signals

006 / RECOGNISED FOUNDATIONS

Bring the framework
you already trust.

Cyber Essentials gives UK organisations a practical baseline around five technical controls. ISO/IEC 27001 defines requirements for managing information-security risk through an ISMS. Startwell uses your applicable framework to shape the deployment gates and evidence—not as a logo exercise.

START WITH THE PEOPLE WHO WILL RUN IT

Bring your IT lead.
Bring your hardest questions.

The first conversation is about the boundary, not the demo.

Plan a safe pilot↗