AI Use Policy

1. Scope of AI use

AI is an optional component inside the internal admin workflows we build in Business Automation engagements. We recommend it only where it genuinely helps, and we agree its use with you during the Automation Assessment. Typical uses:

• Classifying and routing incoming requests — sorting incoming emails or form submissions so they reach the right person or queue
• Pre-filling and extracting data — pulling key details from documents or emails into a form or record for a person to check
• Summarising internal documents — producing short summaries of long internal documents or threads for staff
• Drafting for staff review — preparing draft text that a staff member reviews, edits and sends

PulseForge does not use AI to make regulated professional decisions, make or approve payments, change bank details, make binding commitments on your behalf, or perform any task that requires a professional licence or judgement.

2. Human approval and review

Every workflow that uses AI is designed so a person stays in control. The rules are agreed during the assessment and recorded in the runbook:

• A named person in your team approves anything important before it takes effect
• Items the workflow cannot classify confidently go to a named reviewer
• Complaints, sensitive matters and anything outside the defined workflow are handled by a person by default
• Every AI-assisted workflow has a documented disable procedure, so your team can switch the AI step off and handle items manually

3. What AI is never used for

Workflows configured by PulseForge do not, and are not permitted to, provide or decide:

• Legal advice — including contractual, family, immigration, employment or commercial legal matters
• Medical advice — diagnosis, treatment, medication, symptom interpretation
• Financial or tax advice — investment, tax, superannuation, retirement, insurance product recommendations
• Credit advice — loan suitability, credit product recommendations
• Migration advice — visa eligibility, application guidance
• Payments and bank details — approving or making payments, or changing supplier or customer bank details
• Binding commitments — accepting contracts, quotes or other commitments on your behalf

4. Regulated professions

For clients in accounting, legal, financial, health or other regulated professions, we apply additional controls:

• Professional judgement stays with your qualified staff; workflows only prepare or sort information for them
• Default human handling for anything outside the defined workflow
• Mandatory staff review of any AI-drafted content before it is used
• No collection, storage or routing of identity or KYC documents, tax file numbers, or health or participant clinical records by the workflows we build

We discuss compliance requirements during the assessment and may decline work where a workflow cannot be configured safely for the use case.

5. Automated-message disclosure

Our workflows are mainly internal. Where a workflow genuinely needs to send an automated message to people outside your business, the message identifies your business as the sender, includes a working unsubscribe or opt-out where it is a commercial electronic message (per ACMA guidance under the Spam Act 2003), and, where applicable, links to your privacy policy. Any AI-drafted message is reviewed by your staff before it is sent. See our Communication & Consent Policy.

6. Design and change approval

Before go-live, you approve the workflow design, including:

• Where AI is used in the workflow, and where it is not
• Which items need human approval, and who the named approver is
• How unclassified or low-confidence items are routed
• Any message templates the workflow uses
• The disable procedure

Changes after launch also require your sign-off, except minor corrections (such as typos), which we record in the runbook change log. For Care clients, changes are also recorded in the Care change log.

7. Data handling by AI tools

How data is handled depends on the AI provider and deployment agreed for your workflow. Key points:

• We agree the AI provider and deployment with you, and data is processed under that provider's terms and your settings
• Where supported by the provider, we configure settings to minimise data retention and to stop your data being used to train models
• Workflows send the AI step only the data it needs for the task
• We document the AI provider used for your workflow in the runbook

For more detail, see our Privacy Policy.

8. Monitoring and quality control

Every build includes monitoring and alerts, and a 30-day stabilisation period. For AI-assisted steps we:

• Test classification and extraction against your sample data before go-live
• Review a sample of outputs during the 30-day stabilisation period
• Track how many items go to manual review — too many or too few both indicate tuning is needed
• Under Care, monitor the workflow in business hours and fix faults
• Under Care · Improve, include AI-assisted steps in the quarterly improvement review

Issues we find are reported to you and fixed. We don't hide mistakes — we surface them and improve the configuration.

9. Known limitations of AI

We're upfront about what AI cannot do well:

• Classification and extraction can occasionally misread nuance or context — human review exists for this reason
• AI does not have human judgement; it follows the instructions and rules it is given
• AI-drafted text can contain errors and must be checked by a person
• AI providers occasionally have outages or service degradation outside our control
• Unusual documents, poor scans, non-English text or heavy jargon may reduce accuracy

Setting expectations correctly is part of every engagement. AI supports your team — it doesn't replace skilled people.

10. Client responsibility

Clients are responsible for:

• Approving the workflow design and any message templates before go-live
• Keeping a named approver in place and reviewing items routed to them
• Reviewing AI-drafted content before it is used or sent
• Holding any AI provider account or licence the workflow needs
• Obtaining any consent required for automated messages sent on your behalf
• Telling us promptly about process changes, service changes or compliance concerns

11. Updates to this policy

We review this policy at least annually and whenever our services change. Material updates are communicated to active clients by email. This page always reflects the current policy.

Questions about this policy: hello@pulseforge.com.au. This is a general policy and doesn't constitute legal advice. Specific industry compliance requirements should be reviewed with a qualified legal professional.