Building the RGL Ecosystem红桥全球旅居 · 平台生态建设战略提案

Section 01Executive summary

执行摘要

Five thousand and seventy-six people have contacted this business since December. Thirty-five signed. Of the rest, 2,829 have no salesperson, no outcome and no reason recorded against them, so nobody can say whether we turned them down or simply lost them.

This proposal sets out how to build the RGL ecosystem — website and web application, booking, checkout and payment, and four portals covering staff, clients, customer relationships and content. It starts with that gap, because closing it costs ten weeks and buys back the only asset the business is currently throwing away.

This is not a proposal to build less than the Product Book sets out. We will build all of it. We will start with the parts the evidence says pay back first, because working down the list in the order it happens to be written in is how the money ends up in the wrong place.

The proposal in six points

提案要点六条

1. One platform, eight surfaces, two systems of record. The website, the web app, booking, checkout and the four portals should all be windows onto the same underlying system. One record of who the client is. One record of what they owe and have paid. Build each piece as its own product and they stop agreeing with each other. The client who paid in the checkout is a different client in the CRM, and reconciling the two becomes somebody's monthly job. That is the most common and most expensive way projects of this shape go wrong.

2. Start where the money is, not at the top of the list. The company's own lead data points to how leads are handled after they arrive. Of 5,076 leads in the register, 38.7% have a sales rep named against them and the register records 8 closes against the 35 the business reports. The unnamed majority were not all neglected. Sales screens out enquirers who don't fit, and that's a legitimate commercial judgement. But only 271 leads carry a refusal reason of any kind. A further 2,829, or 55.7% of the register, have no rep, no outcome and no reason recorded at all. So the business cannot tell a lead it correctly refused from one it simply dropped, and neither can anyone reading its numbers. Closing that gap is Phase 1, and it does not depend on a decision from anyone.

3. Then the money layer. Milestone billing for a A$220,000–270,000 engagement, in AUD and CNY, across bank transfer, PayID, WeChat Pay and Alipay, with the payment rails assigned by product tier rather than offered uniformly, because the wallets cannot carry six-figure stages.

4. Then commerce, portals and content. Booking, checkout, the client portal and the CMS follow, serving the products that the first two phases prove there is demand for.

5. Compliance has to be built into the system, not written into a policy. The eight boundaries set out in the Product Book each become a working control: an approval step before marketing claims are published, disclosure of related-party relationships, payee types the system will not accept, milestones described by the work delivered, and evidence recorded as that work is done. Including it now is cheap. Adding it afterwards, and being believed, is not.

6. Ten of these screens already run. The IT function built a staff and client portal for RedBridge Consulting and offered it for testing in March 2026. Pipeline, case milestones, document requests, the client portal and staff messaging are working today. Phase 1 is therefore ten weeks of work on a system that already runs. Routing, deduplication, response timing and reporting, plus the single change that matters most: connecting the portal to where leads actually arrive. About 1.3% of leads reach the website form. The rest arrive on Xiaohongshu. That gap, not the software, is the likeliest reason nobody tried the portal, and Section 8.5 addresses it.

1.1 The triage inbox

1.1 分诊收件箱

Sales already declines leads that don't fit a A$220,000 engagement, and that judgement is sound. What the register shows is that the judgement is almost never written down. 271 leads carry a refusal reason. 2,829 carry nothing at all, so a lead that was correctly refused and a lead that was simply dropped look identical on the record.

The fix is a triage inbox rather than an assignment queue. Imported leads land there scored, and management works the inbox instead of the spreadsheet. Two buttons: admit, which routes the lead by rule and starts the response clock, or decline, which asks for a reason from a short list and archives it. Declining takes one click and one reason. That single field turns 2,829 unexplained rows into a record the business can audit, count and learn from.

1.2 Scoring, and what it can honestly do

1.2 评分机制及其现实边界

Each lead arrives with a score built from what the register already captures well: whether the person is onshore (about 80% are), their visa type and expiry, years of experience and sector, and English test results where stated. The score orders the inbox. It does not decide anything.

That limit is deliberate and worth stating plainly to anyone who asks. The register records 8 closes against the 35 the business reports, so there are not enough recorded outcomes to fit weights to. Any weighting proposed today is judgement expressed as a number. Presenting it as calibration would not survive the first question asked of it.

So the weights start as a written-down version of how the team already decides, they are visible to management, and they are adjusted quarterly once admitted leads have produced enough outcomes to check them against. Recording the decline reasons is what generates that evidence. The scoring model comes out of Phase 1's data discipline. It cannot go in ahead of it.

Hard limits are separate from the score. A small number of conditions send a lead straight to triage flagged for review, with the reason shown, rather than being auto-declined. Anything touching visa history or eligibility is reviewed by the MARA Agent at Insight Idea. No rule resolves it, because that judgement sits with the licence.

Section 02Mandate and scope

委托事项与范围

2.1 What was asked

2.1 委托内容

To propose how we build the RGL platform ecosystem in accordance with the Product Book, covering — but not limited to — a website, a web application, a booking system, a checkout and payment system, a staff portal, a client portal, a CRM portal and a CMS portal, with payments in AUD and CNY via bank transfer / PayID, WeChat Pay and Alipay.

2.2 In scope

2.2 范围之内

2.3 Out of scope

2.3 范围之外

Legal drafting; accounting policy; the corporate restructure itself; media buying; and delivery operations in destination markets. Each is referenced where it constrains the build, but none is proposed here.

Section 03Evidence base

证据基础

This proposal draws on four sources. The Product Book V1.0; the current operating structure of RedBridge Consulting and Insight Idea; the inbound lead register, as at 13 August 2026; and the existing internal portal built for RedBridge Consulting and offered to the sales and CRM teams for testing in March 2026.

3.1 What the lead register shows

3.1 留资台账反映了什么

The front-end status column describes a five-step funnel (arrived, left details, spoke, added on WeChat, assigned) and the volume through it falls away sharply: 1,553, then 895, 590, 489, 389. Assignment itself works as designed at the bottom: 386 of the 389 rows marked assigned name a rep, and 473 of 489 marked added do too. The leak is higher up. Of the 1,553 leads marked only as arrived, 1,540 have no outcome recorded of any kind, and 8% ever reach a rep. Another 590 got as far as a conversation and then stopped, 587 of them with nothing recorded after.

Sales screening out an unsuitable enquirer is a commercial judgement, and a reasonable one. But it accounts for 271 leads. The 2,829 with nothing recorded at all are not screened, refused or lost. They are unresolved on paper, so nobody can say which is which. The gate isn't assignment. It's the disposition that never gets recorded, and it sits before assignment, not at it.

3.1a Where leads actually arrive

3.1a 线索实际从哪里进来

The website form has produced 64 enquiries between 2 May and 14 August 2026, plus 11 on the earlier site version from 18 March. Monthly: 35 in May, 8 in June, 17 in July, 4 to mid-August. Two pairs of records sit four and six seconds apart, which is one person submitting twice. A small duplicate source, but a real one even on the verified form.

Set against the inbound register, this is the single most important structural fact in the document. The website accounts for roughly 1.3% of total lead flow; something close to 98.7% arrives through Xiaohongshu and does not pass through the website form at all.

The existing portal was designed on the assumption that leads enter via the website. That assumption is correct in design and wrong in fact, and it is the likeliest explanation for why nobody tried the portal. A sales team will not open a system that contains one lead in eighty. Section 8.5 addresses it as a plumbing problem, because that is what it is.

3.2 What this implies

3.2 由此得出的推论

The Product Book's underlying instincts hold up well. Own the client relationship instead of brokering it. Build a data asset that gains value over time. Serve a whole lifecycle, not one transaction. Never promise a government outcome. What the data changes is where the emphasis should sit. The people contacting the business today have already decided to migrate and are already living in Australia. So the platform's first job is to respond to that demand faster and more reliably than it does now. Selling additional things to those people comes after that, not before it.

3.3 An asset that already exists

3.3 一项已经存在的资产

The IT function built a staff and client portal for RedBridge Consulting and told the sales and CRM teams it was ready for testing in March 2026. Neither team tried it. The portal was never evaluated, so nothing is known about whether it would have worked for them. Going through the code has changed this proposal quite a lot, because a good deal of what Sections 5 to 8 describe has in fact already been written and works.

That matters for what comes next. Phase 1 is roughly ten weeks of engineering: automatic assignment, deduplication, response timing, attribution, and the reporting that shows whether any of it is working. But it's ten weeks of extension on a system that already runs, not ten weeks of building from nothing. The code is the predictable half. Getting the thing used is not.

Section 04Design principles

设计原则

Seven principles sit behind every decision in the rest of this document. If a later section seems to pick one option over another without explaining itself, the reason is usually here.

One client, one record

一个客户，一份档案

Someone who buys a A$120 experience and someone who signs a quarter-million-dollar engagement are the same kind of record in the system. Everything else (the booking, the case, the invoice, the family member who paid) hangs off that one record.

Configuration over code

配置优先于代码

Products, prices, milestone schedules, fee stages and visa pathways should all be changeable without a developer. Australian visa rules move, prices vary from one client to the next, and the Group will cut the product list back at some point. If any of those changes needs code, the platform will feel out of date within a year.

Compliance as a mechanism, not a memo

合规是机制，不是通知

Each boundary in Section 08 of the Product Book becomes a control with someone accountable for it, a record of what happened, and the ability to stop an action that breaches it. A rule that only works when the person typing remembers it isn't a control.

Evidence captured as a by-product of work

证据随工作自然生成

Deliverables, approvals, disclosures and consents get recorded at the moment they happen, not pieced together afterwards. In three years, that is the difference between a record that holds and one that doesn't.

WeChat-first, not email-first

以微信为先，而非以邮件为先

Almost all the contact details in the register are WeChat; very few are email addresses. While Chinese-market clients remain the core revenue, notifications, follow-up and client identity all work on that basis, with email as the backup. If the Group takes the products global, email becomes the primary channel and the platform has to carry both without a rebuild.

Buy the commodity, build the differentiator

通用能力外采，差异化能力自建

Payment processing, document storage, e-signature, messaging and analytics are all things to buy. The lead engine, the case model, the milestone billing logic and the employer network are things to build, because they are what the business actually is.

Nothing ships without measurement

没有度量，不予上线

Before we build a phase, we agree which numbers it should move. The platform then reports them. A business whose own register understated its signed clients by a factor of four cannot afford to build something else it has no way of measuring.

Section 05Target architecture

目标架构

5.0 The stack, and why it is not off-the-shelf

5.0 技术栈，以及为何不采用现成产品

The stack below is the one the existing portal already runs on. Naming it here is not a proposal to adopt something new; it is a statement of what Phase 1 extends, and it is the reason a ten-week estimate is credible.

The migration is incremental by design. Rebuilding all ten existing screens before Phase 1 starts would delay the only work with a measurable return, and rebuilding them after would mean writing the Phase 1 features twice. So the new surfaces are built on the new stack from the first day, and the older screens move across as they are touched. Nothing stops working while it happens.

The case against a packaged CRM is narrower than it is usually put. Off-the-shelf tools handle pipelines well. What they do not handle is this business's specific shape: leads arriving from Xiaohongshu with no API, a WeChat-first contact model where email is the exception, engagements measured in years with milestone billing in two currencies, and a legal-matter boundary that has to hold at row level. Every one of those would be a workaround in a packaged product, and workarounds are what staff stop using.

5.1 The shape of the system

5.1 系统的整体形态

Eight surfaces sitting on one spine. The spine is two systems of record. One holds people and the relationships between them. The other holds money and what is owed. Every surface reads from those two and writes back to them.

5.1a The two systems of record

5.1a 两套记录系统

The client record holds people, households, organisations, relationships, consents and the full interaction history. It is the Product Book's 客户主档案 made concrete, and it is the asset the Product Book correctly identifies as the Group's most durable.

The money record holds quotes, contracts, milestone schedules, invoices, payments, payers, refunds and exposure. It is deliberately separate, because money has different retention, audit and access requirements from relationship data.

5.2 The complete map

5.2 全景地图

Every surface the Product Book asks for, and what state each is in today. Tags read: built already running, phase 1 in the first ten weeks, later a subsequent phase. Section 11 governs when the later ones enter, and none of them enters without evidence of demand.

Two things on this map are commonly conflated and behave differently. Products 1 to 11 book a product, which means dates, party size, payment and confirmation. Product 12 books a consultation, which means a calendar slot with no payment, before anything is bought. The engagement that follows is quoted, contracted and billed by milestone, so product 12 never passes through checkout at all. Section 6 keeps the two apart in the transaction model.

5.3 Build, buy and integrate

5.3 自建、外采与集成

5.4 E-signature: build or buy

5.4 电子签名：自建还是外采

Can it be built? Yes. The mechanics are not exotic. Capture a drawn or typed signature, hash the document, record signer identity, email verification, IP address, user agent and timestamp, then seal a PDF with an audit trail appended. This can be built in two to three weeks.

Is it legally effective? Under Australian electronic transactions legislation, an electronic signature is generally valid where the signer is identified, their intention is indicated, the method is as reliable as appropriate for the purpose, and the other party consents. A self-built system can satisfy that. So the question is not legality.

The question is evidentiary. If a A$220,000–270,000 engagement is ever disputed, the signature has to survive challenge, and the firm would be defending its own implementation instead of pointing at an industry-standard certificate and a vendor who produces a court-ready audit trail. Add key management, tamper-evidence, long-term validation and ongoing maintenance, and the cost of a build exceeds a subscription many times over.

Recommendation: paper for the Client Agreement, electronic for everything else. The portal runs the whole envelope either way. It fills the contract from the client record, sends it, tracks it, files the executed copy against the case and moves the stage forward. That automation is worth having, and none of it depends on how the signature itself is captured.

For the Client Agreement, the signature stays on paper. Of the eight signings the register records, six have a documented in-person meeting and seven are onshore, six of those in Melbourne. The firm also has a Chengdu branch, so face-to-face signing is available on both sides of the relationship. At roughly 35 signings a year, a signature given in person, scanned and filed against the record, is the strongest evidence available and costs nothing. Thirty-five sheets of paper a year is a cheap way to remove the argument entirely from the one document carrying a quarter of a million dollars.

Everything else signs electronically in the portal: engagement letters, variations, consents, acknowledgements, and any internal document where an audit trail is useful but a dispute is not plausible.

The Client Agreement should include a clause in which the client consents to electronic execution and accepts the portal's audit record as evidence. The legislation already requires consent. Obtaining it in writing, in advance, closes a line of argument before it opens.

Section 06Product-to-system mapping

产品与系统映射

The twelve products in the Product Book are not twelve versions of the same transaction. They come down to five different transaction types. Building five of those properly is achievable; building twelve separate products as twelve separate features is not.

Product 12 is the existing RedBridge Consulting business. The four-stage employer-sponsorship engagement described in the Product Book is what the firm already sells today. It isn't a new product to be designed. The Product Book adds a wrapper of settlement and relocation services around it; the engagement at its centre already exists, already converts, and already produces essentially all revenue.

Product 12 is worth dwelling on. It looks like a booking. It is case management with staged billing attached. Building it inside a reservations engine is the most expensive structural mistake on offer in this project, and it's where almost all the current revenue comes from.

Product 10 (CitySoul) is a user-generated content platform and carries content-moderation and platform-liability obligations that none of the other eleven do. It is treated in this proposal as a Phase 4 candidate requiring a separate decision.

Section 07Payments and money

支付与资金架构

7.1 Currencies

7.1 币种

Pricing and settlement in AUD, with CNY accepted at checkout. There is a material difference between pricing in CNY and collecting in CNY, and the Group has to pick one:

• Accept CNY, settle AUD. The provider converts; RGL's exposure is limited to the movement between quote and payment. Recommended default. 收人民币、结澳元。由服务商换汇；红桥的汇率敞口仅限于报价到付款之间的波动。建议采用此方案。
• Price in CNY. Places FX risk on RGL across engagements billed over months or years. Requires a quote validity window and a re-pricing rule. 以人民币定价。在跨月甚至跨年收费的委托中，将汇率风险转移到红桥身上，需设定报价有效期与重新定价规则。
• Collect CNY onshore in China. A structural question about entities and accounts, not a payment-provider question. Out of scope here; refer to the accountant. 在中国境内收取人民币。这是关于主体与账户的架构问题，不是支付服务商的问题。不在本提案范围内，应交由会计师处理。

7.2 Rails, assigned by product tier

7.2 支付通道按产品层级分配

The four required methods are not offered uniformly. Wallets are built for retail-scale transactions and cannot carry a six-figure milestone; conversely, a bank transfer is poor experience for a A$120 event ticket. The checkout must know which product it is selling before it offers a method.

7.3 Milestone billing

7.3 里程碑收费

The flagship engagement is A$220,000–270,000 excluding legal fees, billed across four stages. Because the total varies per client and the stage split is set per contract, the system stores the milestone schedule against each engagement and bills from the signed contract rather than from a global price list. This is why the stage split being unconfirmed does not block the build.

Three rules govern the design:

• Milestones are defined by work delivered. An event occurring does not release a payment. The invoice line, the contract schedule and the system's milestone definition all use the same wording. 里程碑以已交付的工作定义，而非以某事件的发生定义。发票行、合同附表与系统中的里程碑定义，三者使用完全相同的措辞。
• The system records deliverable evidence against each stage as the work happens — searches run, employers approached, screening completed, documents prepared. 各阶段的交付证据在工作发生时同步留存——已开展的搜寻、已接洽的雇主、已完成的筛选、已准备的材料。
• Legal fees never pass through the platform. They are invoiced by the firm under its own arrangements and are excluded from platform revenue entirely. 法律费用绝不经由本平台。由律所按其自身安排开票，完全排除在平台收入之外。

7.4 Payers

7.4 付款人

Payment by someone other than the client is existing practice and, at these values, will be common. The payer is therefore a first-class record, carrying identity, relationship to the client, and the authority under which they paid. Not a note in a comment field. Three consequences follow:

• Refunds return to the original payer, not to the client. 退款退回原付款人，而非退给客户。
• The invoice and the payment may name different parties; who the tax invoice addresses is an accounting decision we need before building receipting. 发票抬头与实际付款人可能不是同一方；税务发票开给谁，是在建设收据功能之前必须由会计给出的决定。
• An employer can never be a permitted payer of a client fee. The ledger blocks that payer type outright, rather than leaving it to judgement at data entry. 雇主在任何情况下都不得成为客户费用的付款人。这一点在账务系统中以禁止的付款人类型强制实现，而不是交由录入人员临场判断。

7.5 Quote and approval

7.5 报价与审批

Because price is agreed per engagement across a range, a quote step comes before the contract. Someone prices it, someone approves anything outside an agreed band, and the client accepts a specific schedule. The system records whatever moves the price within the range as a field against every quote, permission-restricted if commercially sensitive, so finance can analyse margin later.

Section 08Data, CRM and the client record

数据、客户关系与主档案

8.1 Core entities

8.1 核心实体

Person · Household · Organisation · Employer · Lead · Opportunity · Engagement · Case · Milestone · Deliverable · Quote · Contract · Order · Invoice · Payment · Payer · Refund · Supplier · Document · Consent · Approval · Audit event.

Two modelling choices matter disproportionately. First, person and household are distinct, because several products are bought by one family member and delivered to another. Second, lead and person are the same underlying record at different stages, so that a person who bought a ticket in 2024 and returns as an engagement client in 2027 keeps one history.

8.2 The lead engine

8.2 线索引擎

This is Phase 1 and the highest-value component in the proposal.

• Capture from paid and organic sources, preserving campaign and creative attribution 接收付费投放与自然流量的线索，并保留投放计划与创意的归因信息
• Deduplicate at entry on Xiaohongshu ID first, with WeChat as the secondary key — Xiaohongshu ID is present on 97% of rows and WeChat on 44%, and matching on it alone finds 875 duplicates, about 17% of handling load 在录入端优先按小红书 ID 去重，微信号作为次级匹配键——小红书 ID 在 97% 的行上有值，微信号仅 44%；单按小红书 ID 匹配即可识别 875 条重复，约占处理量的 17%
• Route automatically to a representative or team by rule, not by someone noticing 按规则自动分配给顾问或团队，而不是等人注意到
• First-response SLA timer, with escalation on breach — defined below 首次响应时限计时，超时自动升级——定义见下
• Follow-up scheduling and reminders, because a follow-up date field with no reminder is not used 跟进排期与提醒——因为没有提醒的下次跟进日期字段，等于没人填
• Required contact method at capture, rather than an optional field 留资时强制要求填写有效联系方式，而非设为选填字段
• Reporting on assignment rate, response time, duplicate rate, meetings and closes 报表：分配率、响应时长、重复率、面谈数与成交数

What a first-response SLA is

什么是首次响应时限

An internal commitment to how quickly a new lead receives its first human reply, enforced by the system instead of by good intentions. A clock starts when the lead is created and stops when the first outbound contact is logged. Thresholds trigger escalation automatically. At the first, the assigned representative is notified; at the second, their team leader; at the third, the lead returns to the pool for reassignment. Nobody has to notice.

It matters because in high-intent inbound sales, speed to first contact is the largest controllable influence on conversion, and right now nobody measures it. Two practical notes: measure the median and the 90th percentile, because averages hide the leads that were left for two days; and make the clock business-hours aware across both Australian and Chinese time zones, or every overnight lead will read as a breach.

8.3 The engagement file

8.3 委托档案

Client documents live in Google Docs today, as Insight Idea has always kept them. Passports, visa histories, English test results, testamurs and employment references sit in shared folders. It works, and it has worked for years.

What it does not do is form a file. A folder holds documents. It does not record which matter they belong to, when the firm asked for them, when they arrived, who opened them afterwards, or when they should be destroyed. Sharing links outlive the engagement that created them. For a A$274,000 matter handled under a MARA Agent's registration, that gap is the exposure worth naming.

The portal already produces the missing half. Document requests record what was asked for and when. Uploads record arrival. The case timeline records the stage each document supports. Row-level policies in the database record who may open what, and every access is logged. None of that requires new work. It requires the engagement to be run there.

The legal split then falls out of this rather than being imposed on it. Every case carries a tag: general CRM record, or legal matter. Legal matters open in full only to the legal role. CRM staff see that the matter exists and what stage it has reached, and nothing else. Access is logged in both directions, and the tag is a column in the database rather than a hidden menu, so the rule holds whoever is looking.

The rule covers records that already exist, not only new ones, so the historical CRM data needs a classification pass before we switch it on. Two things give that pass an end date. Anything ambiguous defaults to legal matter until someone with the legal role reclassifies it, so the safe state is the default state. And one named person from Insight Idea signs off the classified set, because a data-cleaning job with no end date does not get one.

Retention

留存期限

A file that is never destroyed is as much a liability as one that is destroyed too early. The platform closes a matter on the visa grant or the point the case is lost, holds the documents for the period the Code of Conduct requires from that date, then deletes them automatically and logs what it destroyed. Insight Idea supplies the retention period. The platform enforces it without anyone remembering to.

Two things keep it safe. A refusal under review is held until the review concludes, so nothing is destroyed while it may still be needed. And the deletion log is permanent, because the evidence of compliance is the log rather than the absence of the file.

Decision 6. Whether some matters should sit outside the platform entirely, in the firm's own system, is for Insight Idea to answer. What this document proposes is that whichever matters do sit inside it are recorded properly.

8.4 Migration of existing data

8.4 既有数据迁移

The existing pipeline records import with their stages preserved. Operating stages already in use win over the Product Book's six-stage lifecycle model, because they reflect how the business actually works. Where records move between separate legal entities, the consent and disclosure basis has to exist before any record moves. It is a prerequisite, and it belongs with the lawyers rather than in the engineering backlog.

8.5 Getting Xiaohongshu leads into the system

8.5 让小红书的线索进入系统

Around 98.7% of leads come from Xiaohongshu and never touch the website form. Xiaohongshu does not offer the kind of API access that would let a company of this size pull those leads out automatically. That access is reserved for very large advertisers. Since the leads never arrive in the portal, none of the routing, deduplication or reporting work can do anything for them, which is why this has to be solved before anything else in Phase 1.

The website form already works. Point it at the register

官网表单已经可用，只需把它接到台账上

RedBridge Consulting has a site with an eight-step consultation form on it. It collects visa status, occupation, years of experience and English level before anyone speaks to the enquirer, which is more than a Xiaohongshu lead arrives with. It works. It is not the problem.

The problem is where the submissions go. Repointing that form so it writes to the same lead register as the Xiaohongshu import, tagged by entry path, puts website enquiries into the same triage inbox as everything else. It takes two to three days and requires no change to the site.

Two entry paths, not one

两条进入路径，而非一条

Website enquiries and Xiaohongshu enquiries reach the business in completely different ways, and forcing them through one process is what would break it. Website leads arrive before anyone has spoken to them, so they run the full pipeline: qualification, discovery, strategy, proposal, contract. Xiaohongshu leads are screened live, inside the chat, in the first minutes of contact. By the time any export runs, the decision has already been made.

So the export serves as the ledger. Routing happens live on Xiaohongshu, before any file is downloaded. Someone downloads the report on a fixed schedule (daily is enough) and uploads it. Xiaohongshu rows land at their outcome, with the reason attached, and the sales representative adds the value on the ones that closed. From there they flow into the CRM pipeline like any other signed client.

Response time can still be measured

响应时长仍然可以度量

A daily import cannot chase a lead going cold in real time, and nothing on Xiaohongshu can. It can still measure the thing that matters. Xiaohongshu records when the person first messaged, and the portal records when the representative recorded an outcome. The gap between those two timestamps is first response, computed at import, reported the following morning. That gives an accurate median for managing the team, which is what the number is for. Live escalation while a lead is still warm is a separate capability, and it depends on the WeCom decision below, not on how often the report is uploaded.

It is not elegant, and it costs someone a short slot each day. But it turns a system holding one lead in eighty into a system holding all of them, and everything else in Phase 1 depends on that. Build it first, improve it later.

It is worth being clear that this removes work rather than adding it. The register holds 5,076 rows with conversation transcripts and written summaries typed into them by hand. That transcription is happening today, every day. The import populates those fields automatically. The only new action anyone performs is selecting a reason on the leads they declined, and even that can be done next morning from a short exception list rather than in the moment.

Three things make the difference between this working and it becoming another abandoned process:

• One named person, one time of day. If it's nobody's specific job, it won't happen. 一个指定的人，一个固定的时间。属于所有人的任务，等于不属于任何人。
• The upload must be forgiving. It should accept the report exactly as Xiaohongshu exports it, without anyone reformatting columns first. If the file needs preparing, the process will lapse within a fortnight. 上传功能必须足够宽容。要能直接接受小红书导出的原始报表，不需要任何人先去调整列格式。若上传前还要整理文件，这个流程两周之内就会中断。
• Re-uploading the same file must be harmless. Reports overlap, and people will upload twice. The importer should recognise rows it has already seen and ignore them. 重复上传同一文件必须无害。报表之间会有重叠，也一定会有人传两次。导入功能应能识别已处理过的记录并自动忽略。

What was ruled out, and why

已排除的方案及其原因

Conversations: keeping one source of truth

沟通记录：如何保持唯一真实来源

The same problem appears again with messaging. The portal already gives RedBridge Consulting and Insight Idea clients their own chat channel, and having every conversation attached to the right client, searchable, and still there when someone leaves, is plainly the right goal. The difficulty is that clients live in WeChat and will not move to a web chat simply because we would prefer it. A policy that fights that will lose, in the same way that running a spreadsheet alongside a CRM means the spreadsheet wins.

Two things make this workable in practice.

Pull the client back, don't push them. When a document is requested, a milestone completes or a staff member replies, the client should get a WeChat notification that opens the portal thread. The client stays in the app they already use, and the conversation of record stays in one place. Most of the traffic follows the notification.

Where the conversation genuinely has to happen on WeChat, run it through WeCom. WeCom is the corporate version of WeChat. Staff use a work account, clients message it from ordinary WeChat, and, unlike personal WeChat, it has a proper interface that lets the company retain conversation records and attach them to a customer. Each external contact carries a stable identifier, and that identifier becomes the key that ties a WeChat thread to the right client record in the CRM. Capturing it at first contact is what prevents orphaned conversations later.

Section 09The website, content and discovery

官网、内容与获客可见性

For this business, content is how customers are found in the first place. That makes the CMS an operational system, not a convenience for the marketing team.

9.1 What the website is, and what sits on it

9.1 官网的定位与结构

RGL is the parent property. RedBridge Consulting sits beneath it as the established service line, and the other products in the Product Book sit alongside as lines the Group may add, keep or retire without touching the brand above them. That structure decides the site, so state it before arguing about pages.

One consequence follows immediately. RedBridge is the only line with proven demand, so it cannot be presented as one of twelve equal offerings. It carries the site's credibility and produces the revenue that funds everything else. The other lines earn prominence as they demonstrate demand, rather than being granted it at launch.

Every page exists in English and Chinese, and the Chinese version is not a translation afterthought. Around 85% of leads arrive through Chinese-language Xiaohongshu accounts, so for most visitors the Chinese page is the site.

9.2 Content ownership

9.2 内容归属与责任

The build is not the constraint on this site. Content is. A page inventory of this size needs roughly sixty pieces of copy in two languages, and the review gate in 9.3 applies to every migration-related one of them.

Three roles need naming before Phase 3 begins: who writes, who approves migration content (the MARA Agent at Insight Idea, per Decision 7), and who owns translation quality. Without those three names the site launches with placeholder pages, and fewer complete pages would serve the business better.

9.3 The brand transition, and its cost

9.3 品牌迁移及其成本

Of 5,076 leads in the register, 4,332 arrived through Xiaohongshu accounts carrying the RedBridge name. The website and the WeChat official account together produced six. The acquisition channel is not merely branded RedBridge; it is a set of individually-named RedBridge accounts, each with its own following.

That matters if RGL becomes the front-facing brand. Followers do not transfer between accounts, and a renamed account on Xiaohongshu keeps its history without keeping its momentum. Whatever the eventual brand hierarchy, moving the acquisition channel onto it is a marketing project with a cost and a risk window. The same holds for the existing domain, where the accumulated search authority needs redirects and a transition period.

9.4 The CMS

9.4 内容管理系统

Requirements

具体要求

• Draft → review → publish, with a named approver. Migration content is the highest-risk content the Group publishes, and a CMS where anything can go live unreviewed is the mechanism by which copy drifts into outcome language. 草稿 → 审核 → 发布，并指定审核人。移民类内容是集团对外发布的内容中风险最高的一类；任何内容都能未经审核直接上线的 CMS，正是文案逐步滑向承诺结果表述的那个机制。
• Bilingual authoring with translation status per article, matching the Product Book's own bilingual discipline. 中英双语撰写，逐篇标记翻译状态，与《产品册》自身的双语规范保持一致。
• Structured content — questions, answers, eligibility explanations, pathway comparisons — because both search engines and generative answer engines reward precise, well-sourced, citable material. This works with the compliance requirement instead of against it: accurate, carefully-scoped content both performs better and carries less risk than the outcome-promising kind. 结构化内容——问题、解答、资格说明、路径对比——因为搜索引擎与生成式问答引擎都更青睐准确、有出处、可被引用的材料。这一点与合规要求是一致的，而非冲突的：准确、界定清晰的内容，既比承诺结果式的内容效果更好，风险也更低。
• Evidence record per published claim, linked from the article, satisfying the Product Book's own standard that claims be substantiated. 每条对外主张配备证据记录，从文章中可直接链接查看，以满足《产品册》自身主张须有依据的标准。
• Search-term library — the register already contains hundreds of distinct queries in customers' own words, ranked by volume, including trust queries about the brand. This is a ready-made content plan and costs nothing to act on. 搜索词库——台账中已有数百个以客户自己的措辞表达、并按热度排序的搜索词，其中包括对品牌可信度的查询。这本身就是一份现成的内容规划，落地不需要任何额外成本。

No creator accounts, attribution or revenue share are proposed for the initial release. Content is authored in-house, so CMS roles are sufficient and the creator backend is deferred.

Section 10Compliance by design

合规内嵌设计

The Product Book sets eight risk boundaries. Each becomes a mechanism.

Section 11Delivery plan

交付计划

11.0 Three ways forward

11.0 三条可选路径

Before the phases, the alternative. Doing nothing is a real option here and not a hypothetical one, because it is what the Group does today. Its consequences are visible in the register rather than modelled.

11.0a Stop conditions

11.0a 停止条件

Phases are only gates if something can fail them. Three conditions, each with a stated consequence, govern the next twelve months.

These only work if the Chairman fixes the numbers in advance and names someone to check them on a date. Set after the fact, they become commentary.

11.1 Adoption is a deliverable, not an outcome

11.1 被使用是交付物，不是自然结果

The existing portal was offered for testing and nobody tried it. That is a more useful fact than a rejection would have been, because it rules out the software as the cause. Section 3.1a sets out the likelier reason: the portal held about one lead in eighty, because the leads were arriving somewhere else. The wider lesson is worth saying plainly. Building the software is the easier half of this problem. A system nobody uses produces the same numbers as no system.

This proposal therefore treats adoption as scoped work with named owners, not as something expected to happen once the build is finished:

• One workflow, one team, first. Lead assignment and first response for one sales team. Not every feature for everyone at once. 先做一个流程、一个团队。先在单一销售团队上跑通线索分配与首次响应，而不是一次性向所有人上线所有功能。
• Migrate the real data before go-live. Staff won't use a system that's emptier than the spreadsheet they already have. The existing register imports first. 上线前先迁入真实数据。如果新系统比员工手上的表格还空，他们不会用它。现有台账必须先导入。
• Make the new path the shortest path. If a lead arrives in the portal already routed and already deduplicated, using it is easier than not using it. Most adoption failures come down to friction rather than unwillingness. 让新路径成为最省事的路径。如果线索进入门户时已经完成分配与去重，那么用它就比不用它更省事。推行失败通常源于摩擦，而非抵触。
• A named operations owner who sets routing rules and response thresholds, and whose numbers those become. 指定一位运营负责人，由其设定分配规则与响应时限，并对这些数字负责。
• Report the gate weekly — assignment rate and median first response, per team, visible to the Chairman from week one. 每周汇报闸门指标——分配率与首次响应中位时长，按团队呈现，自第一周起对董事长可见。
• Retire the alternative. Run it alongside a spreadsheet and the spreadsheet wins. Set a date on which the portal becomes the record. 停用替代方案。与表格并行运行，结果必然是表格胜出。须设定一个日期，自该日起门户即为唯一记录。

Deliberately deferred

有意推迟的部分

Supplier and external-professional logins; creator and city-curator backend with attribution and revenue share; multi-country tenancy; UGC platform. Each one is written down as excluded, so anyone adding it back has to say so out loud.

What each phase must prove

各阶段必须验证的指标

Section 12Resourcing

资源与人力

This plan assumes one developer. That is enough to do Phase 1 properly and then Phase 2. It is not enough to run Phases 1 through 4 at the same time, and trying to do so tends to deliver none of them.

The business supplies three roles, not engineering, and none of them is full-time:

• A decision owner who approves scope, settles a question within a week, and breaks a tie. 决策负责人：可批准范围、在一周内给出答复、并在意见相持时拍板。
• An operations owner who defines routing rules, SLA thresholds and pipeline stages — these are business rules and the business sets them. 运营负责人：定义分配规则、响应时限与管道阶段——这些是业务规则，不是技术问题。
• A content approver from Insight Idea, being the MARA Agent, for the publication gate. 内容审核人：由 Insight Idea 的 MARA 持牌代理担任，负责发布关卡。

To compress the timeline, the constraint is engineering capacity. There are three ways out: a second developer, a delivery partner for Phase 3, or a longer schedule. Better to have that conversation now than at month nine.

Section 13Investment model

投入与回报

13.1 Cost structure

13.1 成本结构

Costs fall into internal engineering time, third-party software subscriptions (payment processing, document storage, e-signature, messaging, hosting, monitoring), and professional fees wherever the build depends on legal or accounting input. Two of the three are already known. Phase 1 runs entirely on the free tiers of the current hosting and database providers, so it adds no subscription cost at all; paid tiers become necessary only when RGL exists as an operating entity with real volume, and at that point the two providers together sit between roughly US$45 and US$800 a month depending on tier. Internal engineering time is set out in Section 11: about ten weeks for Phase 1 and about thirty-two developer-weeks across Phases 1 to 3, on existing capacity. Only the internal cost rate and the professional fees remain open, because the rate and the corporate structure are not settled. The structure below is set up for finance to populate.

13.1a What it costs beyond money

13.1a 金钱之外的代价

Four costs do not appear in a budget line and should be weighed against the return.

• The daily export. Roughly one staff-hour a day, borne by operations rather than sales. It replaces manual transcription that happens today, so the net is negative, but the hour moves to a different person. 每日导出：约占用一名员工每天一小时，由运营而非销售承担。它替代的是目前正在发生的人工转录，因此净工时是减少的——但这一小时换了人来出。
• The transition period. Between go-live and the cutover date, two systems hold the same data and neither is fully trusted. Section 11.1 keeps that window short by naming a date; it does not remove it. 过渡期：从上线到切换日之间，两套系统同时保有相同数据，且哪一套都不完全可信。第 11.1 节通过明确切换日期把这个窗口压到最短，但无法消除它。
• Ten weeks of the only developer. No other IT work proceeds during Phase 1. Anything that arrives in that window waits or displaces the phase. 唯一开发人力的十周：第一阶段期间不进行其他 IT 工作。该窗口内出现的任何需求，要么等待，要么挤占本阶段。
• A new task for management. The triage inbox is work that did not exist before. It is short, and it can be batched, but it is real. 管理层的新增工作：分诊收件箱是此前不存在的工作量。它耗时很短、也可以批量处理，但确实是新增的。

13.2 The return case

13.2 回报测算

The clearest return is in Phase 1, though not for the reason an earlier draft of this section gave. Routing more leads to reps is not worth much on its own, because a large share of the unrouted ones were refused deliberately and would convert at nothing like the rate of the rest.

The return sits in the 2,829 leads with no disposition at all, and in the 1,553 that never moved past "arrived". Three things follow from closing that gap. Media spend gets cheaper: 93% of August leads were paid for, campaign and creative IDs are already captured on 38% of rows, and once refusals carry a reason those reasons tell the media buyer what to stop buying. Contact quality gets fixable: 206 leads went nowhere because the person never replied or the details were unusable, on a register where phone is filled 3% of the time and WeChat 44%. And the closes stop going missing. The register shows 8 against 35 advised, so every ratio the business quotes about itself is wrong by roughly four times.

Put a number on it after the sample described below, not before. Anything quoted now would be a guess wearing a forecast's clothes.

13.3 Unit economics

13.3 单位经济模型

At roughly 145 leads per signed client and an average engagement in the mid-A$200,000s, media cost remains a modest share of revenue across a wide band of plausible cost-per-lead values. Confirming the actual figure refines the budget without changing whether the plan is viable. Confirm it anyway, because it sets how hard the Group can push acquisition.

Section 14Risks

风险

Section 15Decisions required

需要决策的事项

Four block Phase 2. One blocks Phase 1. That's Decision 10, and it's about adoption, not engineering.

Appendix AThe twelve adviser questions

顾问提问的情景推演

Section 09 of the Product Book puts twelve questions to advisers, asking not for endorsement of the vision but for go/no-go conditions, resource sequencing, capital structure and twelve-month priorities. They remain open. None of them blocks Phase 1.

A scenario analysis of all twelve — the plausible options, what each would imply, and a suggested default where the evidence supports one — is in the companion paper The Twelve Adviser Questions, issued alongside this document and not for decision.

Appendix BGlossary and document control

附录 B · 术语与文件控制

This document is ordered as an argument rather than by case, because the Chairman reads it front to back and the argument is what carries the decision. The table above is for anyone reviewing it against the framework — an adviser, an investment committee, or a funder's due diligence — and it names the gaps rather than leaving them to be found.

Prepared for internal decision-making. Not legal, financial or migration advice. Contains no personal data.
供内部决策使用。不构成法律、财务或移民意见。不含个人信息。