This file is generative eval scaffolding for testing capture and extraction, not knowledge. It holds invented NZ income-protection client personas and sample intake dialogues, each paired with the structured fields an extractor should produce, so they can seed an extraction golden set (strategy §6) and exercise multilingual capture (strategy §7). The extracted fields shown here are the human-labelled expected outputs, not anything the model priced or decided. Bob never invents premiums or numbers.
Header: everything below is fictional test data
Every name, employer, age, income figure, address, and quote in this file is synthetic and made up for testing. There are no real people here. All identifiers are obvious fake placeholders (FAKE-PERSONA-01, "[fake employer]", "$XX,XXX" style ranges, no real IRD or NHI numbers, no real addresses or phone numbers). Income figures are coarse round numbers used only to test extraction, never real earnings and never used to price anything. If any value in this file ever resembles a real person, that is coincidence and it should be edited out. Do not ingest this file as NZ product or advice knowledge; it is structure and test data only.
The expected-fields schema each persona is labelled against:
reason: why they are enquiring about income protection (the trigger), free text normalised.dependents: count and short note (children, partner, others relying on the income).occupation: job title or trade, normalised.employment: one of: employee (PAYE) / self-employed / contractor / business owner / casual / not working.income: approximate annual income as a coarse range; a plain hyphen if not stated.language_preference: preferred language for the conversation (transcripts here are all English; the preference is metadata for routing to native operative text and voice).
Source: Bob knowledge-base data strategy §6, §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Persona FAKE-PERSONA-01: young tradie, first cover, casual speech
A 26-year-old self-employed builder, sole trader, no staff, recently went out on his own. Talks fast and casual, drops articles, uses trade slang ("on the tools"). Worried because a mate hurt his back and could not work for months. No income if he cannot swing a hammer. Single, no kids. Prefers English.
Expected fields:
reason: Income stops entirely if injured and unable to do physical work; prompted by a peer's injury.dependents: 0.occupation: Builder (carpenter).employment: self-employed (sole trader).income: approx $70,000-$90,000.language_preference: English.
Source: Bob knowledge-base data strategy §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Persona FAKE-PERSONA-02: mid-career nurse, PAYE, two kids
A 41-year-old registered nurse at a [fake DHB], PAYE employee, works shifts. Married, partner works part-time, two children (ages roughly 8 and 11). Methodical, asks about how cover interacts with ACC and sick leave. Main earner. Prefers English.
Expected fields:
reason: Wants protection beyond ACC and employer sick leave because she is the main earner for a family of four.dependents: 2 children plus a partner on part-time income (note: partial reliance, not full).occupation: Registered nurse.employment: employee (PAYE).income: approx $80,000-$95,000.language_preference: English.
Source: Bob knowledge-base data strategy §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Persona FAKE-PERSONA-03: Mandarin-preferring restaurant owner (transcript in English)
A 53-year-old who owns a small restaurant in [fake suburb]. Would prefer to talk in Mandarin; this test transcript is captured in English to test the case where preference is set but the turn is English (mixed-language routing). Business owner with a few casual staff. Married, one adult child no longer dependent, supports elderly parent overseas. English is a second language so sentences are short and sometimes drop tense. Income varies month to month.
Expected fields:
reason: If she cannot run the restaurant, the business income drops and she also sends money to an elderly parent.dependents: 1 (elderly parent supported overseas; adult child not dependent).occupation: Restaurant owner / operator.employment: business owner.income: approx $90,000-$110,000 (variable, owner drawings; the seed dialogue narrows it to about $90,000-$100,000).language_preference: Mandarin (transcript captured in English).
Source: Bob knowledge-base data strategy §6, §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Persona FAKE-PERSONA-04: Cantonese-preferring IT contractor (transcript in English)
A 34-year-old software contractor on day rates through their own company, working at a [fake telco]. Prefers Cantonese; transcript here is English. Sharp, precise, asks detailed questions, gives exact figures. Single, no dependents, but has a large mortgage. Cares about whether cover follows them between contracts.
Expected fields:
reason: A gap between contracts plus illness would leave a large mortgage unfunded; wants cover that is not tied to one client.dependents: 0.occupation: Software developer / IT contractor.employment: contractor (own company).income: approx $130,000-$160,000 (day rate, variable).language_preference: Cantonese (transcript captured in English).
Source: Bob knowledge-base data strategy §6, §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Persona FAKE-PERSONA-05: Korean-preferring early-career professional (transcript in English)
A 29-year-old marketing coordinator at a [fake agency], PAYE. Prefers Korean; transcript captured in English. Polite, slightly unsure of insurance terms, asks Bob to explain things simply. Recently married, partner studying (so partner has little income), no children yet. New to dealing with financial products in NZ.
Expected fields:
reason: Newly the household's main earner while partner studies; wants a safety net if she cannot work.dependents: 1 (partner studying, low income, partial reliance).occupation: Marketing coordinator.employment: employee (PAYE).income: approx $60,000-$70,000.language_preference: Korean (transcript captured in English).
Source: Bob knowledge-base data strategy §6, §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Persona FAKE-PERSONA-06: older self-employed farmer, rambling speech
A 58-year-old who runs a small dairy operation with their spouse, self-employed, near [fake rural town]. Talks in long looping sentences, lots of context before the point, mentions weather and stock prices. Spouse works the farm too, two adult children moved away. Sceptical of "city insurance talk", values plain answers. Wants to know what happens if his health stops him farming before he can wind down.
Expected fields:
reason: A health event could stop him farming years before planned wind-down, with no salary to fall back on.dependents: 0 dependent children; spouse works the farm (joint income, not a dependent).occupation: Dairy farmer.employment: self-employed.income: approx $70,000-$110,000 (variable, farm income).language_preference: English.
Source: Bob knowledge-base data strategy §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Persona FAKE-PERSONA-07: part-time / casual worker, uncertain employment, single parent
A 36-year-old single parent working casual retail shifts at [fake retailer] plus occasional cleaning work, so income is irregular and split across two jobs. Hesitant, unsure whether she even qualifies, worried about cost. Two children (ages roughly 4 and 7), no partner. Prefers English. This persona tests the messy "employment is not a clean single value" case and a low, variable income.
Expected fields:
reason: As a sole earner for two young children with no second income, any time off work is a crisis.dependents: 2 children, sole provider.occupation: Retail assistant and casual cleaner (two jobs).employment: casual (multiple casual roles; income irregular).income: approx $35,000-$45,000 (variable, combined).language_preference: English.
Source: Bob knowledge-base data strategy §6, §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Sample intake dialogue 01: tradie, messy utterance to clean capture (golden-set seed)
Messy user utterances (verbatim style, for FAKE-PERSONA-01):
- User: "yeah nah so im a builder, just went out on me own this year, no boss now"
- User: "mate of mine did his back, off work like 3 months, scared me cause if im not on the tools theres no money coming in eh"
- User: "nah just me, no missus no kids"
- User: "dunno exactly, clear maybe 80k ish before the gst and that"
Clean normalised capture (expected extractor output):
reason: Self-employed and earns nothing if injured and off the tools; prompted by a peer's 3-month back injury.dependents: 0.occupation: Builder (carpenter).employment: self-employed (sole trader, started this year).income: approx $80,000 (self-reported, pre-tax, approximate).language_preference: English.
Notes for graders: "out on me own" → self-employed; "no money coming in" + injury → reason; "80k ish before the gst" → coarse income range, never treated as a quote. The extractor must not invent a premium or imply a price.
Source: Bob knowledge-base data strategy §6 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Sample intake dialogue 02: Mandarin-preferring owner, short ESL utterances (golden-set seed)
Messy user utterances (verbatim style, for FAKE-PERSONA-03; preference Mandarin, turn in English):
- User: "i have restaurant, small one, in [fake suburb]"
- User: "if i sick, no work, restaurant no money. and i send money my mother, she old, in [overseas]"
- User: "my son big now, work himself. just my mother i help"
- User: "income not same every month. maybe ninety, hundred thousand, depend"
Clean normalised capture (expected extractor output):
reason: If illness stops her running the restaurant the business income falls, and she financially supports an elderly parent overseas.dependents: 1 (elderly parent supported overseas; adult son not dependent).occupation: Restaurant owner / operator.employment: business owner.income: approx $90,000-$100,000 (variable, owner drawings).language_preference: Mandarin (this turn captured in English).
Notes for graders: short ESL phrasing must not lower extraction confidence on the clear facts; "i send money my mother, she old" → one dependent (the parent), not the adult son. Stricter bar applies to non-English preference routing even when the turn is English. The operative duty-of-disclosure and underwriting wording would be served in Mandarin from reviewed native text, never machine-translated.
Source: Bob knowledge-base data strategy §6, §7 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft
Sample intake dialogue 03: casual single parent, ambiguous employment (golden-set seed)
Messy user utterances (verbatim style, for FAKE-PERSONA-07):
- User: "im not sure i even qualify? i do shifts at [fake retailer], casual, and i clean a couple houses on the side"
- User: "its just me and the two littlies, 4 and 7, no one else"
- User: "money's all over the place honestly, some weeks good some weeks nothing, maybe forty grand if i add it all up"
- User: "worried what happens if i get crook and cant do the shifts"
Clean normalised capture (expected extractor output):
reason: Sole earner for two young children with irregular income; any time off work due to illness is a financial crisis.dependents: 2 children, sole provider.occupation: Retail assistant and casual cleaner (two roles).employment: casual (multiple casual roles; income irregular).income: approx $40,000 (variable, combined, self-reported).language_preference: English.
Notes for graders: two income sources collapse to a single employment: casual with a note, not two records; "the two littlies, 4 and 7" → 2 dependents, sole provider. The eligibility question ("i even qualify?") is for the adviser path, not something the model decides or prices.
Source: Bob knowledge-base data strategy §6 (https://bob.so/strategy/knowledge-base-data-strategy) · retrieved 2026-06-18 · rights: own-summary · drives: information · status: draft