Your TMD Needs Customers Who Do Not Exist.

Synthetic personas can expose a target market that is too broad, a distribution condition that fails at the edge and a review trigger that never fires. Use them as adversarial tests. Never mistake generated cases for evidence of real customer outcomes.

Your target market determination should survive customers who do not exist. Build fictional edge cases that sit just inside, just outside and directly across its boundaries. Then ask whether the product attributes, target-market description, distribution conditions and review triggers produce a defensible result.

This is a proposed testing technique, not an ASIC-endorsed form of evidence. It is product design assurance, not AI-driven personalisation or a distribution decision. Synthetic personas can reveal contradictions before a product reaches a customer. They cannot show how the product performs in real hands, replace complaints and claims data, or decide whether a product is suitable for an individual. Where AI is instead used to tailor the offer itself, the separate risks in DDO and AI-driven personalisation apply.

The strongest method is not a gallery of plausible profiles. It is a controlled set of counterfactual pairs. Change one relevant fact at a time, preserve the expected result and make a human product owner explain why the outcome changes.

What does the TMD actually have to prove?

Part 7.8A of the Corporations Act 2001, current compilation No. 147, registered as C2026C00339 and in force from 1 July 2026, contains the enforceable design and distribution obligations. The regime commenced on 5 October 2021, as recorded in ASIC's 18 July 2023 industry letter. Section 994B(5) requires a TMD to describe the class of retail clients comprising the target market, specify distribution conditions, identify review triggers, set review periods and state specified reporting arrangements. Section 994B(8) adds the appropriateness requirements. It must be reasonable to conclude that consumers acquiring in accordance with the distribution conditions are likely to be in the target market, and that the product is likely to be consistent with the likely objectives, financial situation and needs of consumers in that market.

ASIC Regulatory Guide 274, issued 10 September 2024, is guidance about ASIC's interpretation and administration of those obligations. RG 274.6 says DDO is not an individualised point-of-sale suitability test. RG 274.65 says the appropriateness requirements are objective and do not require knowledge of individual consumers. A synthetic persona therefore tests class-based TMD logic, not suitability for a real person.

RG 274.81 says the target market should use objective, tangible parameters. RG 274.84 warns against insufficient granularity, while RG 274.86 and RG 274.87 explain that identifying excluded classes may help define the boundary and inform distribution controls. At the product level, RG 274.88 to RG 274.90 expects the issuer to critically assess the key attributes and consider how the product is likely to perform in consumers' hands.

Use those requirements as test surfaces. A paired case can change only [PREMIUM_CAPACITY] to challenge an affordability boundary. Another can test whether conflicting answers, changed circumstances or an exclusion still allow an application to proceed.

The law also operates beyond the document. Sections 994E and 994F require reasonable steps directed at consistent distribution and complete and accurate records of TMD decisions and reasons. A synthetic test record can assist that evidence chain only as an input to the human decision.

How do you build a persona test that can fail?

Randomly asking a model for ten customer personas creates colour, not assurance. Build a boundary-pair suite instead. Each pair contains two fictional consumers who are identical except for one TMD-relevant variable. The controlled difference might be capacity to bear loss, access to an insured benefit, expected product use, premium affordability, investment timeframe or eligibility for an exclusion.

Assign each case an expected state before running it: INSIDE, OUTSIDE, BOUNDARY or UNRESOLVED. Then test four linked decisions:

• Product fit: are the key attributes likely to be consistent with the class member's likely objectives, financial situation and needs?
• Target-market boundary: does the written description place the case where the product team expected?
• Distribution response: do the channel, questionnaire, script and other conditions direct or stop the case as designed?
• Review sensitivity: would an adverse pattern involving this case activate a defined trigger or another escalation?

Use three kinds of pair. Together, they test design, routing and continuing appropriateness without pretending to predict actual customer behaviour.

Use this prompt to generate candidate boundary pairs from approved product material. A human product owner and compliance reviewer must verify every extracted attribute, set the expected state and approve cases before testing.

Using only [APPROVED_TMD], [PRODUCT_TERMS] and [DISTRIBUTION_RULES], draft adversarial fictional persona pairs for [PRODUCT_ID]. Do not use real customer data and do not decide compliance or individual suitability.

For each pair:
1. keep all facts identical except one TMD-relevant variable
2. name the changed variable and source passage
3. propose INSIDE, OUTSIDE, BOUNDARY or UNRESOLVED for human review
4. state the expected distribution response
5. identify the product attribute, TMD boundary, condition and review trigger tested
6. flag ambiguity, missing evidence and possible stereotyping

Use merge fields only. End with HUMAN APPROVAL REQUIRED.

Fictional worked example: [PRODUCT_ID] is an income-protection style insurance product used only to illustrate the method. Persona [PAIR_A] and persona [PAIR_B] have the same age, occupation, income, objective, coverage need and distribution channel. The only changed fact is [EMPLOYMENT_STATUS]. The approved TMD says the target market requires a continuing income stream that the benefit is designed to protect.

The human product team pre-classifies [PAIR_A] as INSIDE and [PAIR_B] as OUTSIDE. The digital journey nevertheless lets both reach a quote because its question captures occupation but not current employment status. That is not proof that section 994E has been breached. It is a test failure showing that the stated boundary and the implemented distribution response may not align. The team must investigate the product facts, law, TMD and live control before deciding whether to change the question, distribution arrangement or target-market wording.

Counterfactual pairs also reduce demographic theatre. Do not add age, postcode, disability, cultural background or family status merely for realism. Include a characteristic only where the approved design connects it to the product, boundary or control. Human reviewers must challenge proxies and discriminatory assumptions.

What belongs in the human decision record?

A passed synthetic test is not a clean bill of health. Keep a persona-to-control trace showing exactly what was tested and what the result can, and cannot, support. This trace is a proposed internal assurance artefact, not a prescribed ASIC template.

For every case, retain the source versions, fictional inputs, controlled variable, expected state, actual response, affected TMD passage, product attribute, condition, trigger and reviewer disposition. Preserve the model and prompt version if AI assisted. Name the human owner and approver.

ASIC Report 795, published 10 September 2024, says pre-launch questionnaire testing can identify flaws but cannot guarantee intended performance. It says you cannot know whether a questionnaire works until it is live and viewed alongside cancellations, complaints and consumer outcomes. REP 795 also reported limited outcome monitoring among reviewed issuers. Generated profiles cannot supply that evidence.

ASIC's 18 July 2023 review of more than 100 general and life insurance TMDs likewise found weaknesses including target markets and review triggers expressed without enough detail. Its letter to the Insurance Council of Australia identified objective and tangible parameters, affordability considerations and clear negative target markets among observed good practices. It also said insurers should use data such as claims ratios, policies sold, lapse and cancellation rates, claims durations and outcomes, and complaints when identifying review triggers.

The evidence model should therefore contain two lanes. The synthetic lane challenges whether the design can distinguish known edge cases. The outcome lane tests whether real distribution and product performance support the TMD after launch. A product governance forum can compare both, but must never use a generated pass to cancel a contradictory real-world signal.

Use this prompt to challenge a completed test pack. The human DDO owner must resolve the gaps and decide whether any TMD, product, distribution or monitoring change is required.

Review [PERSONA_TEST_PACK] against [APPROVED_TMD] and [LIVE_OUTCOME_EVIDENCE]. Do not treat synthetic results as customer-outcome evidence and do not make a legal or suitability decision.

Return:
1. untested target-market boundaries and key attributes
2. failed or ambiguous distribution responses
3. review triggers with no measurable signal or threshold
4. conflicts between synthetic results and live outcomes
5. unsupported assumptions, proxy risks and missing source passages
6. candidate actions for human decision, with owner and evidence required

Separate design findings from actual outcome findings. Mark every conclusion HUMAN REVIEW REQUIRED.

Do this Monday

1. Choose one TMD boundary. Select a product with an approved TMD and one disputed or operationally difficult inclusion, exclusion or distribution condition. Do not begin with the whole portfolio.
2. Build three counterfactual pairs. Create one entry pair, one control pair and one drift pair. Change a single relevant fact within each pair and set the expected state before testing.
3. Run the real journey safely. Test in an approved non-production environment using fictional merge-field data. Capture every question, branch, message and stop outcome.
4. Create the persona-to-control trace. Link each result to the exact product term, TMD passage, condition, trigger, source version, human reviewer and decision.
5. Reconcile with live evidence. Compare the synthetic findings with complaints, distribution, claims, lapse, cancellation and other relevant outcome data your organisation lawfully holds.
6. Send decisions to the forum. Have the authorised product governance body decide whether to change the product, TMD, distribution control, monitoring metric or test suite. AI should prepare the issue, not approve the answer.

Bottom line

Synthetic personas are useful when they are designed to break a TMD, not decorate it. Counterfactual pairs can expose weak boundaries, control gaps and review triggers before customers bear the cost. They remain proposed test data, never evidence of actual consumer outcomes or a decision about an individual's suitability. Make AI generate and organise the challenge; make accountable people approve the cases, interpret the failures and decide what changes.

This article is general information and education only. It is not legal, compliance, financial or professional advice. Obligations vary by organisation and circumstance. Verify current requirements against the primary sources cited and seek advice specific to your situation.

References

1. Federal Register of Legislation, Corporations Act 2001, current compilation No. 147, C2026C00339, in force from 1 July 2026, Part 7.8A. https://www.legislation.gov.au/C2004A00818/latest
2. Australian Securities and Investments Commission, Regulatory Guide 274 Product design and distribution obligations, issued 10 September 2024. https://download.asic.gov.au/media/etgm1amc/rg274-published-10-september-2024.pdf
3. Australian Securities and Investments Commission, Report 795 Design and distribution obligations: Compliance with the reasonable steps obligation, released 10 September 2024. https://download.asic.gov.au/media/clqh43yt/rep-795-published-10-september-2024.pdf
4. Australian Securities and Investments Commission, ASIC review of insurance target market determinations, 18 July 2023. https://www.asic.gov.au/about-asic/news-centre/news-items/asic-review-of-insurance-target-market-determinations/
5. Australian Securities and Investments Commission, Letter to Insurance Council of Australia: ASIC review of insurance target market determinations, 18 July 2023. https://download.asic.gov.au/media/kargc3iy/ica-ddo-letter-released-18-july-2023.pdf