Illustrative example · subject to change
Illustrative assessment example
This is a fictional illustration. It shows one possible format for a workflow assessment report. The brokerage, staff, files, findings and figures are fictional. It is not a client result and does not describe any real organisation. An assessment is a service; it does not use the TrustShyft platform, which is in development. The format may change.
Loan application to lodgement
Where AI is used in this workflow, where people review its work today, and what we recommend so oversight is proportionate and can be shown later.
Residential home loan applications, from first client meeting to lodgement with the lender.
Post-settlement servicing, commercial lending, lender-side processes.
Three staff interviews, a walkthrough of two de-identified files, and your current procedures.
No AI used
Reviewed, not recorded
Reviewed, not recorded
No review identified
Review recorded
Reviewed, not recorded
Step 4: Best interests notes
The broker records why the recommended loan is in the client's best interests.
A general-purpose AI tool drafts the notes from the fact find and product comparison.
Drafts are pasted into the CRM. We did not find a step where someone checks them before they are saved.
- Best interests duty (NCCP Act)
- Record-keeping for credit assistance
- ASIC REP 798, a report on AI governance (context, not an obligation)
F1, R1
Best interests notes are saved without an identified review
These notes may later be relied on to show why the loan suited the client. If an AI draft is wrong, nothing currently catches it before it becomes the record.
Review of AI drafts happens but leaves no trace
Staff describe genuine checking. Because it is not recorded, the brokerage could not show later who checked what on a given file.
No written guidance on AI use
Which tools are approved, what may be entered into them and who checks the output are left to individual judgement, so practice varies between staff.
Client details entered into a general-purpose AI tool
The tool's data retention and training settings have not been checked. This is a privacy question to resolve, not a finding of a breach.
Proportionate to the risk at each step, listed in priority order. You decide which to adopt.
Add a named review before best interests notes are saved
Record what the broker checked and whether they accepted, changed, rejected or escalated the draft, with a brief rationale, before it goes into the client file.
Check the AI tool's data settings and limit client details
Confirm retention and training settings, switch to a business account if needed, and use matter references instead of client details where possible.
Set a proportionate minimum review note for higher-risk AI drafts at Steps 2, 3 and 6
The note covers what was checked and why. Higher-risk drafts include, for example, fact-find summaries with missing or conflicting information, serviceability drafts where the surplus is close to the lender's buffer, and client correspondence that explains a recommendation.
Write a one-page AI use guide
Approved tools, what information may be entered, and who checks the output at each step. Review it every six months.
Next step
We walk through the findings with you and issue the final report.
This report sets out our observations and recommendations for one workflow, within the agreed scope. It is not legal advice, and it does not determine whether Example Brokerage complies with any law, standard or regulatory guidance. References to obligations show what to consider, not a conclusion about compliance.
What an assessment gives you
- A map of one AI-assisted workflow and where review happens
- Findings on the conditions affecting that review: workload, tools, incentives and decision design
- Prioritised recommendations, proportionate to the risk at each step
An assessment does not determine whether your organisation complies with any law or standard. See our Disclaimer.
Talk to us about an assessment