Campaign Abroad search · cheapest click
Campaign Bintent search · higher cost
- sources
- Aggregate journey events by source and stage (pseudonymous). Google Ads and Meta reporting, read-only. No individual application file.
- reasoning
- Campaign A applicants stall between submit and assessment at twice B’s rate and generate the most “what happens now?” support contacts.
- proposed
- Shift $1,400 of A’s weekly budget to B. Replace the landing entry with a five-step “what happens after you submit”. Suppress retargeting to anyone already in assessment.
- limit
- Total spend unchanged. No campaign paused. No bid or audience change outside A and B.
- duration
- Seven days, then review.
- stop if
- Complete applications fall more than 10%, or support contacts rise, or any compliance flag.
- cannot see
- Income, transactions, credit scores, decision reasons, names.
- nectar decides
- Whether this runs at all. Budget beyond the test. Anything touching criteria, pricing or eligibility.
// what Nectar’s systems send the component. Nothing more. journey: ntr_48391 state: clarification_required stage_began: 2026-08-23T09:41:12+12:00 reason_code: regular_payment_classification customer_action: required human_owner: assessment_team sharing_permission: prepare_only explanation_code: EXP-CLAR-07 // approved Nectar copy // not sent: income, transactions, credit score, decision reasoning
Codes and approved context, not a file. Enough to explain the issue and prepare a response.
the agent receives
- journey reference
- current state and when it began
- whether the customer must act
- approved explanation code
- missing-item code
- contact and sharing permission
- named team for the next step
the agent never receives
- income records
- bank transactions
- credit scores or bureau data
- decision reasoning
- affordability assessment
- authority to judge sufficiency
- any other customer’s data
- Nectar journey event
- → permission and data check
- → retrieve approved Nectar content (process explanations, document requirements, quote terms, total-cost wording, signing guidance, support routes, prohibited language, responsible-lending and brand rules)
- → agent prepares a structured output: explanation · customer_action · limitations · human_owner
- → policy and boundary check
- → customer reviews, edits or skips
- → named Nectar person or permitted workflow
- → evidence receipt
The customer can always keep waiting, leave and return, correct information, decline preparation, request a person, and see what will be shared before approving it.
// journey-performance table: pseudonymous, aggregate, never a file campaign_id · source_id · landing_id · experiment_id application_started · application_completed assessable_file_reached · clarification_required quote_viewed · signing_completed · funded_outcome time_between_stages · support_contact · journey_abandoned // never sent back to ad platforms: decision reasons, sensitive data
Which source produces complete applications. Where a campaign’s applicants get stuck. Which search terms bring volume but poor downstream quality. Whether people are retargeted after entering assessment.
assembl prepares or runs
- monitoring, anomaly detection, pacing
- search-term, audience and funnel analysis
- recommendations with evidence
- landing, suppression and keyword changes, prepared
- after approval: the exact change, by deterministic code
nectar keeps
- lending decisions and responsible-lending accountability
- compliance judgement
- criteria, pricing, eligibility
- major budget decisions, positioning, brand
- anything outside the agreed authority
- continue, change or stop
Find the real wait. Baseline, one journey state, the data boundary, the approved content library, permissions, privacy and compliance review, events instrumented, platforms connected read-only, synthetic journeys tested.
Shadow. Staff or a very small cohort. The operator observes and prepares. Nothing writes to advertising or customer systems. Compare its output with the team’s.
One live wait. Prepare-only. Measure comprehension, missing information, time to next step, skips, corrections, escalations.
One bounded action. A landing, suppression or keyword experiment with cohort, duration and rollback. Then a continue, change or stop recommendation with evidence.
The same system that operates growth also assembles a better journey around every customer.
This is not another marketing dashboard, and not a chatbot attached to a loan application. The performance role does not sit outside the funnel producing traffic and reports. It becomes part of the customer experience it is responsible for improving.
The customer agentinside the phone
Helps one person move through the experience: what is happening now, what to understand, what to prepare, whether anything is required, what remains a Nectar decision, who can help when a person is needed.
The growth operatorbehind the journey
Learns from aggregate journey patterns and improves what the next person encounters. Its governing question: which activity creates complete, assessable and funded customer journeys, not merely cheap applications?
A named Nectar personin control
Controls every consequential change. The operator prepares; Nectar decides. Every approved action is executed by deterministic code, not the model, and leaves a receipt.
Personalisation is limited to explanation, preparation, timing, preferences and support. It never changes lending criteria, eligibility, pricing or the responsible-lending decision.the rule the whole concept is built on
Every line of the brief becomes visible work, with its authority stated.
Left column: Nectar’s own words from the Seek listing (Nectar NZ Limited, Auckland CBD, hybrid, listed about 20 August 2026). Middle: what happens in the world. Right: who holds the authority.
| Responsibility (from the listing) | What happens in the world | Authority |
|---|---|---|
| Google Ads, Meta, Microsoft Ads“ownership of digital acquisition across Google Ads, Meta, Microsoft Ads and other channels” | Live streams show pacing, anomalies, spend and conversion movement. Read-only first. | agent-run monitoring |
| Keywords and search terms“digging into keywords, search terms, audiences, creative and campaign structure” | Wasteful queries physically break away; useful intent strengthens the journey path. Negative keywords prepared. | prepared or bounded |
| Audience strategy“building smarter audience” | Cohorts connect to aggregate journey quality and funded outcomes, never to an individual file. | agent-run analysis |
| Campaign structure“creative and campaign structure” | The operator assembles proposed groups, targeting and exclusions in space, for review. | approval-gated |
| Budgets against CAC“managing budgets against CAC” | Budget flows toward better downstream quality, inside a fixed total and a test limit. | approval-gated |
| Lead quality and funded outcomes“lead quality, approvals and funded outcomes” | Channels compared by complete and assessable journeys and funded yield, not form volume. | agent-run analysis |
| Landing-page and funnel conversion“identifying and improving landing-page and funnel conversion” | A friction point is exposed as a physical obstruction and a new entry experience is assembled. | experiment approval |
| Retargeting and suppression“retargeting and suppression strategies” | Messages disappear when journey state makes them unsuitable, for example anyone already in assessment. | governed rules |
| SEO / GEO“contributing to SEO/GEO” | Search opportunities assemble into briefs linked to genuine customer questions Nectar’s site leaves unanswered. | agent-prepared |
| Attribution and reporting“attribution and performance reporting” | Media spend visibly connects to application, assessment and funded outcome. A daily operating brief. | agent-run |
| Structured tests and creative“running structured tests across channels, audiences, creative and conversion journeys” | New creative and landing variants assembled from observed journey needs, with cohort, duration and rollback. | brand approval |
| Commercial recommendations and judgement“turning performance data into clear commercial recommendations” | Recommendations with evidence. Positioning, regulated judgement and commercial accountability stay with people. | human-owned |
Source: Seek job 94094122, “Performance Marketing Specialist”, Nectar NZ Limited, read 23 August 2026. The listing must be rechecked before this page is sent; it claims only responsibilities the listing actually names.
Before you hire the role, give assembl the workload for 30 days.
Not the job title. The workload. If it cannot cover enough, hire. If it can, Nectar has a different operating model, and a customer journey no longer built like an ordinary lending funnel.
What assembl would operateinside agreed authority
- daily paid-media monitoring, spend and CAC pacing, anomaly detection
- search-term, keyword and audience-quality analysis
- landing-page and funnel diagnostics; CRO experiments prepared
- SEO/GEO opportunity work
- retargeting and suppression recommendations
- attribution from spend to funded outcome; lead-quality reporting
- commercial recommendations and a daily operating brief
- bounded campaign and budget actions after authority is agreed
- one live customer-agent journey, application through funded
What Nectar retainsalways
- lending decisions and responsible-lending accountability
- compliance judgement under the FMA’s consumer-credit supervision (from 1 July 2026)
- major budget decisions
- positioning and strategy
- final brand approval
- any action outside the agreed authority
- the decision to continue, change or stop the pilot
What success is measured againstthe scorecard
- share of repetitive role responsibilities covered
- manual analysis and reporting hours removed
- time from issue detection to prepared action
- recommendations accepted, changed or rejected
- visibility from acquisition source to funded outcome
- journey comprehension; missing-information rates
- unnecessary customer contacts; irrelevant retargeting removed
- experiments launched; funded-outcome economics
- complaints, corrections and compliance guardrails
One product. Two tightly separated layers. The phone is the interface; underneath is event-driven, permissioned and auditable.
The cinematic phone above would become a native component inside Nectar’s own website or application, not a separate microsite. Nectar’s systems send it a handful of events. The growth operator works on a separate, aggregate layer and never sees an individual application.
Start with one real wait
Nectar publishes a roughly seven-minute application, a personalised quote, online signing and same-day funding after approval. Those are the landmarks. Nectar confirms where customers genuinely wait. The shell covers all six states; only one connects to real systems first. No delay is ever manufactured to create a moment.
A small component, fed by codes
Journey reference, state, when it began, whether the customer must act, an approved explanation code, a missing-item code, sharing permission and the named team. Enough to explain and prepare. Not income, transactions, credit scores or decision reasoning.
A controlled workflow, not a prompt
Event → permission and data check → approved Nectar content → structured output (explanation, customer action, limitations, human owner) → policy and boundary check → customer reviews, edits or skips → named person or permitted workflow → receipt.
Lending decisions stay completely separate
No authority to assess suitability or affordability, predict approval, recommend accepting, alter pricing or terms, interpret transactions freely, decide a clarification is sufficient, or route anyone toward borrowing more. Nectar’s engine and qualified people are authoritative.
A separate aggregate growth layer
A journey-performance table of pseudonymous events by campaign, source, landing and experiment: started, completed, assessable, clarification, quote viewed, signed, funded, time between stages, support contact, abandoned. Decision reasons never go back to advertising platforms.
Platforms connected read-only first
Reporting access to the platforms Nectar actually uses: Google Ads reporting down to search terms, Microsoft Advertising reporting, Meta. Retrieve, reconcile with journey events, detect anomalies, prepare recommendations, generate the daily brief, show evidence, wait for approval. No write access in phase one.
The first action is reversible
A controlled landing-page variant through Nectar’s feature flags, an approved negative keyword, suppression of advertising to people already in an application, or a non-consequential explanation change. Prepared with evidence, expected effect, audience, limits, duration, approver, stop condition and rollback. Executed by deterministic code, never by the model directly.
Approval alters the customer journey
The operator finds a campaign whose applicants misunderstand what follows submission, prepares a clearer entry, Symon approves a seven-day experiment, the feature flag activates it for the cohort, the phone renders the new opening, the operator monitors completion, clarification, support contact and assessable-file quality, and the system retains, changes or rolls back.
A production runtime, not this demo’s backend
The concept agent on this page is fine for a concept. A live lending journey needs single sign-on, role-based permissions, per-customer and per-operator authorisation, persistent journey state, a versioned content library, deterministic policy checks, encrypted evidence, rate limiting, model and prompt version records, safety checks, monitoring, retention and deletion rules, correction workflows, tested fallbacks and rollback. The runtime lives inside Nectar’s environment or an isolated, contractually approved assembl one.
Privacy and risk work first
Before any customer pilot: a Privacy Impact Assessment, a data-processing agreement, confirmed model-provider retention and training terms, a data-flow diagram, security review, lending and compliance review, accessibility testing, Māori and affected-customer considerations, customer-facing explanation and consent, incident and complaint procedures.
Nectar providesa small working group
- Symon or another commercial sponsor
- product owner
- marketing-platform owner
- an engineer who knows the application journey
- data and analytics owner
- lending and compliance representative
- privacy and security representative
- customer-service owner
- a named approver for customer content and operator actions
What Nectar receives at day 30kept, whatever you decide
- the working Nectar journey component and one connected customer wait
- an approved-content and policy library
- a read-only growth operator and an advertising-to-funded-outcome view
- an approval and evidence system; one completed controlled experiment
- a coverage assessment against the role description
- a quantified recommendation on what assembl can operate, and a roadmap
Regulatory context: consumer-credit regulation transferred from the Commerce Commission to the FMA on 1 July 2026. Privacy guidance: Office of the Privacy Commissioner on AI and generative AI. Platform APIs: Google Ads reporting, Microsoft Advertising. Each of these is a public source, not a claim about Nectar’s systems.
What this will never do
It does not decide. It does not assess suitability or affordability, predict approval, recommend borrowing, or touch criteria, pricing or eligibility. It does not read income, transactions or credit files. It does not write to an advertising platform, move a budget, change a bid or publish a variant until a named Nectar person approves that exact change, and then deterministic code performs it, not the model. It does not retarget anyone a journey state says should be left alone. It does not pretend to be a person: every explanation says it was prepared by an agent inside a Nectar journey.
Declined applicants never become an advertising surface for another lender. Hardship goes to a person, not a pathway.
No production access is requested by this concept.
Scope
One journey state (assessment or clarification), the six-state shell, Google Ads and Meta read-only, thirty days after a three-to-four-week preparation. One reversible experiment in the last ten days. Not the whole funnel, not write access, not customer data.
Access
Journey stage events as codes, an approved content library, read-only platform reporting, and a named approver. No lending data, no decision reasoning, no spend authority beyond the agreed test limit.
Scorecard
Share of the listing’s responsibilities genuinely covered. Reporting hours removed. Detection-to-prepared-action time. Recommendations accepted. Source-to-funded visibility. Comprehension and missing-information rates. Complaints, corrections and guardrails. And the one that matters: would Nectar be annoyed if you switched it off?
Fail any line of the scorecard and we change the design or stop.
A 90-minute workshop. Not access to production.
From it, assembl produces the event contract, the data boundary, the live-wait specification, an implementation estimate and a fixed 30-day pilot scope. It has to answer:
- Where do customers genuinely wait?
- Which state creates the most confusion or incomplete information?
- What events already exist?
- Which systems own those events?
- What advertising and analytics platforms are active?
- What data must never enter the assembl runtime?
- Who approves customer explanations and growth actions?
- Which reversible experiment would be safe enough to test?
Pick a verb.
Not “book a demo”. Any of these is a real next step, and the last one is a perfectly good answer.
What is the one constraint we have got wrong?
Every concept is built from the outside. There is always something about how Nectar actually runs, the assessment flow, the events that exist, the thing compliance would never allow, that we could not see. One line is enough.
Opens your mail app to assembl@assembl.co.nz. Nothing is collected by this page.