Download this editable Markdown draft ↗

CyberG7: from WhatsApp enquiries to a measurable business system

Research date: 1 October 2026, Singapore time. Prepared for CyberG7 Technologies LLP.

Requested reference: the four-route comparison artifact, now supplied in full as the user's “PSG to EDGE: Jarvis Ad Playbook & EDGE Readiness Guide”. This report incorporates that text and independently verifies the rules. The updated GPT receptionist is the product basis.

Recommended direction: sell a narrow, dependable WhatsApp receptionist first; expand successful clients into a connected sales, operations and finance system operated through Jarvis. Choose funding after the workflow and economic case are defined.

Jarvis can participate in all four routes in the question, through different scopes. Its name, number of agents and “AI OS” branding do not determine eligibility. Three agents performing marketing tasks still perform one business function. A grant supports the SME applicant's approved project; it does not approve CyberG7's entire estate.

Companion documents: sample SME proposals, worked integrated ERP application and revised advertisement playbook.

1. The funding map, corrected

New EDGE applications opened on 30 September 2026. Enterprise Singapore's EDG, PSG and MRA closed to new applications after 29 September; existing projects continue under their original schemes. EDGE's overall annual support cap is S$100,000, refreshed on 1 April. Funding is reimbursed after completion and full payment. Enterprise Singapore: EDGE

Built Environment exception: BCA's PSG remains open, with transition to EDGE scheduled for December 2026. Its September 28 notice explicitly distinguishes it from Enterprise Singapore's PSG. Kim Pong's ACMV business merits an eligibility check against this route before deciding on EDGE, but its industry alone does not establish applicant or Jarvis package eligibility. BCA's digital route still requires a pre-approved solution. Do not use an unqualified “PSG is gone” advertisement for this audience. BCA Built Environment PSG

Claims calendar: EDGE accepts one final claim after completion and full payment. Claims open from 1 November 2026 for single-function digital solutions, integrated enterprise systems and pre-scoped equipment, and 1 December 2026 for other activities. A portal opening date does not mean every applicant becomes ready or receives payment that day. EDGE FAQ

Sources for the activity-specific figures: customer engagement, marketing content, integrated ERP, digital transformation, IRAS EIS.

The two single-function categories sit within the first of the user's four routes. They are different activities with different ceilings. Selected digital solutions, integrated systems and selected automation share a S$30,000 annual subcap. A separate application is required per activity; do not duplicate a cost across applications. The ERP page uses broader wording about automation/digitalisation caps than the general FAQ. Confirm the selected activity's treatment in BizSG and the Letter of Offer rather than assuming every transformation project has the same subcap. EDGE FAQ

How to use the EIS route honestly

IRAS includes qualifying AI-related development, deployment, consultancy, data, maintenance and training services, as well as AI systems. Hardware is excluded; mixed packages require allocation and government subsidies reduce the qualifying basis. This is broader than “any AI subscription”, but an AI label alone proves nothing. IRAS EIS

Illustrative company arithmetic: S$12,000 qualifying expenditure × 4 = S$48,000 deduction. At a 17% marginal rate, with sufficient taxable income and ignoring exemptions and rebates, the tax reduction would be S$8,160 in total, of which S$6,120 is additional to an ordinary 100% deduction. The supplier is still paid S$12,000. Confirm the actual tax effect and basis period with the accountant. A December-year-end company's YA 2027 normally relates to 2026, and YA 2028 to 2027. CyberG7 is an LLP; these company-rate illustrations must not be used for its partners' own taxes. IRAS corporate tax basics

2. What “integrated, without vendor pre-approval” means

The official ERP page describes one ERP platform, not a collection of unrelated apps with a chatbot in front. It permits a non-pre-approved vendor when the required functionality cannot be found in suitable pre-approved packages. The published application outcome times are one week with a pre-approved vendor and ten weeks with another vendor. These are assessment times, not project delivery or reimbursement dates. Integrated ERP

For Kim Pong, make the three functions concrete:

A marketing assistant, content writer and lead finder cannot stand in for these three functions. GHL plus WhatsApp alone does not establish an ERP. Jarvis Mission Control alone does not establish an ERP either. An underlying ERP such as ERPNext is a candidate for discovery: its documentation covers service sales, accounting, projects and maintenance workflows. This is a proposed foundation, not a CyberG7 deployment or a grant endorsement. ERPNext service organisations, maintenance schedule

CyberG7 should prepare a comparison of suitable approved packages. Document the required function, available package capability, missing requirement, evidence and operational consequence. A preference for CyberG7 or for an “AI OS” is insufficient justification. If an approved package meets the requirements, offer that route honestly, with separately priced CyberG7 services only where appropriate and supportable.

What a credible application contains

The ERP activity explicitly asks for the vendor scope and cost breakdown, a proposal covering at least three deployed/integrated modules for another vendor, and a one-year cost-saving projection. The accompanying worked application adds our recommended evidence: process diagrams, a package comparison, assumptions, acceptance tests, implementation milestones and accountable client owners. These additions make the case reviewable; they are not presented as extra statutory requirements. Integrated ERP application documents

Applicant checks include Singapore operations, at least 30% ultimate Singapore citizen/PR ownership, the SME group-size test, location and prior-support restrictions, sufficient funding, and vendor independence. The ERP route excludes home-based businesses. No project work, payment or deposit should precede submission. The SME submits, pays and claims through its own authorised personnel; CyberG7 supplies the proposal and evidence. Costs committed after submission remain at the SME's risk until approved. ERP eligibility, EDGE FAQ

For ERP, the activity page identifies software costs for up to twelve months and excludes maintenance and GST. The FAQ allows meaningful professional services within assessed solution packages. Consequently, list migration, integration, training, support and software separately and obtain their assessment; do not assume every line is funded or disguise maintenance as licensing. Use the final Letter of Offer as the controlling scope. The applicant must retain deployment identifiers, completion evidence, invoices and payment records. ERP costs and claims, professional services FAQ

A real comparator, and the limit of the evidence

Enterprise Singapore's February 2026 food manufacturing release describes Toh Thye San Farm using AI to capture WhatsApp orders into its platform, alongside integrated accounting, inventory and sales. It reports a 20% improvement in order accuracy and month-end reporting reduced from ten days to four. The same release describes KH Roberts integrating finance, procurement, sales, inventory and CRM. These are strong examples of connected operational outcomes. The release does not disclose an approved non-pre-approved EDGE application, its quotation, assessment or Letter of Offer. Enterprise Singapore and IMDA, company profiles, pages 7–9

I found no verified public successful application under the new ten-week EDGE route. EDGE opened yesterday relative to this research date. The worked packet is therefore an application model, not an approved case or assurance of success.

3. What the CyberG7 estate actually supports

The supplied estate map is a 27 August 2026 snapshot of 54 hostnames. Its counts and domain faults are historical. This review sampled the principal offer pages, three named businesses, relevant local repositories and the current receptionist health endpoint; it did not rescan all 54 properties or perform a production systems audit.

The live /healthz observation at wa.cyberg7.com.sg on 1 October showed tenant_answers: 0, delivered: 5, and an active gpt-5.4-nano composer. Global counters do not establish five customers, production reliability or per-tenant answer coverage. The local September 30 handoff and marketing-site documentation still say there are no onboarded clients. Older READMEs describing a wholly deterministic runtime are stale: the current code includes model composition after safety checks.

The public receptionist page correctly labels Rosewood Nails fictional. Keep that separation. Kim Pong, Lynn Beauty and GetSGPR are described as real client businesses in the estate; their public sites support the business descriptions. This review does not verify their contractual relationship, SME eligibility or AI deployments.

Three claims to reconcile before wider advertising

1. Approval behaviour. The Jarvis page promises review before sends. The receptionist sends permitted replies automatically. Specify that owners approve the knowledge and automation policy in advance; fresh quotations, discounts, refund decisions, payments and publications require the appropriate approval. Do not use one blanket promise for both modes.
2. Data location. The receptionist composer calls an external OpenAI endpoint with customer text and configured facts. That cannot be described as “nothing leaves your machines”. Offer a clearly documented hosted deployment, or a separately scoped local-model deployment, with the actual processing chain disclosed. Do not equate Singapore database hosting with wholly local processing.
3. Outcome proof. Launch pages still display “3,000+ Singapore SMEs launched”; the receptionist has no documented clients. These may refer to different products, but the website count needs its own substantiation. The receptionist's absolute “0 facts it invents” also requires reconsideration with a generative composer. Prefer a testable statement: approved facts, guarded composition and human escalation, with error monitoring. Jarvis public page, LaunchFast, LaunchNow

4. The connected product story

Owner-facing proposition: “Keep customer enquiries moving, without spending your day repeating answers or chasing the next step.”

Suggested product structure:

Recommended technical relationship:

Ads / search / website / referral
              ↓
Business-specific WhatsApp channel
              ↓
Receptionist: approved knowledge + guarded composition + human takeover
              ↓
Customer / opportunity / consent records
              ↓
Jarvis: propose → validate → obtain required approval → execute → record
              ↓
ERP: sales ↔ service operations ↔ finance
              ↓
Owner report: unresolved enquiries, next actions, time, bookings and cash status

Keep the public WhatsApp service limited to the SME's customers, orders, bookings and support. Do not expose a general-purpose Jarvis AI OS through the public number. WhatsApp's current terms restrict AI providers where AI itself is the primary functionality; application to a particular deployment remains Meta's determination. This supports a business-specific service wrapper and a protected internal Jarvis console. WhatsApp Business Solution Terms

Recommended integration controls: one persistent customer ID, one order/job ID, tenant-scoped access, signed webhooks, duplicate protection, approval receipts tied to the actual action, retry/error queues, recoverable history and monitoring. Tool access must be limited to the scoped business workflow. Customer messages must not grant system privileges. These are delivery recommendations, not claims that all are implemented today. Map processing purposes, subprocessors, access and retention with each client. PDPC AI guidance

5. Attracting the owner: problem, hook, demo and system

These are draft hooks, not performance claims or published advertisements. Replace forecasts with client evidence once earned. A useful grant line inside a proposal is: “We will assess the appropriate EDGE activity or qualifying AI tax treatment for this scope.” Avoid “guaranteed grant”, “400% cashback”, or suggesting ACRA registration/SME Centre contact means IMDA approval.

Use the receptionist page as the first campaign destination because it shows the actual narrow product and labels its demo. Keep Jarvis as the expansion explanation. Do not funnel this campaign through a cheap website price unless the offer is explicitly a website purchase. Unify naming, pricing basis and the single next step across the ad and destination.

The proposed sales journey is: industry-specific ad or AEO article → labelled demo → owner supplies an anonymised workflow → baseline review → scoped proposal → grant/self-funded decision → client application if relevant → controlled deployment → measured result → client-authorised case study → expansion.

Track the actual journey in the CRM: source, creative, conversation ID, business identity, discovery booked/attended, qualified opportunity, proposed scope, funding route, applicant readiness, won/lost reason and realised gross profit. Cost per conversation is a diagnostic; cost per qualified opportunity and acquired customer's gross profit determine whether advertising works.

6. Success criteria

All numbers below are proposed acceptance targets, not observed CyberG7 or client results. Agree final thresholds after a baseline.

For IMDA pre-approval, the published vendor criteria also include company age, financial stability, support capacity and customer productivity references. The onboarding process typically spans four to five months once ready. The GenAI checklist requires more than a working model: demonstrate knowledge ingestion, relevant engagement capabilities, dashboard/reporting, exports, privacy requirements and third-party security assessment. The admin shell and absent client proof are meaningful gaps. Vendor criteria, onboarding, chatbot checklist

7. Sequence and responsibilities

If a client chooses a grant-supported project, avoid beginning the proposed funded scope before submission. A demonstration using fictional data is different from commencing their implementation. A paid pilot must not be retrospectively rolled into a later grant; separate any genuinely new phase with its own scope and assessment. Discuss the boundary before charging or deploying.

CyberG7 can also be a grant applicant for a genuinely new qualifying internal project using an independent vendor. It cannot manufacture eligible expenditure by invoicing itself, and existing Jarvis/WhatsApp work cannot be made retrospective. Its own external qualifying AI purchases may merit tax review, with LLP treatment assessed correctly.

8. Evidence status and remaining work

Known: official rules checked on 1 October; accessible August estate; current narrow offer and website price differences; local code contains a model composer; health counters show limited successful activity; local CRM documents explicitly distinguish undeployed legacy systems.

Inferred: salons provide the most accessible initial proof opportunity; Kim Pong has the clearest proposed sales–operations–finance story; GetSGPR benefits from secure casework integration. These are fit hypotheses, not diagnosed client problems.

Assumed: every proposal budget, workload, labour rate, time saving, revenue projection and acceptance threshold. None is a customer quotation or result.

Missing: each applicant's exact legal entity, local ownership, group size, tenancy, financial readiness, prior funding, relationship declarations, systems inventory, transaction baseline, access rights, approved package comparison and signed quote. CyberG7/Jarvis pre-approval was not verified; unsuccessful public searches are not proof of absence.

The first Claude artifact required sign-in; the user subsequently supplied its complete playbook text, which is incorporated here and audited in the revised advertisement playbook. The second artifact was accessible. No ad was published, customer contacted, infrastructure changed or grant submitted. The deliverables are reviewable business/application drafts; they do not certify eligibility or approval.

Download this editable Markdown draft ↗

Four sample proposals for three real Singapore businesses

Prepared 1 October 2026. These are illustrative, unsent proposals. Public business descriptions are sourced; internal problems, prices, volumes and outcomes are hypotheses. The businesses' SME status and grant eligibility have not been verified. Nothing here is an existing CyberG7 customer result, approved grant, binding quotation or authority to deploy.

Amounts are in SGD, excluding GST and unpriced third-party usage unless stated. Replace every illustrative commercial line with a real CyberG7/vendor quotation before submission. Do not structure a project to fill a grant ceiling.

EDGE uses one final claim after completion and full payment. Claim submissions open from 1 November 2026 for single-function/ERP activities and 1 December for other activities; these are opening dates, not promised reimbursement dates. A self-funded project started before submission cannot later be claimed as the same EDGE project. EDGE FAQ

Proposal A — Lynn Beauty: Jarvis Receptionist

Objective: reduce repetitive WhatsApp handling while maintaining correct prices, resource availability and prompt human takeover.

Known business context. Lynn Beauty's website offers nail, lash, facial and related services, online appointments and WhatsApp contact. Location references are inconsistent across the page; verify outlets, legal entities and numbers during discovery. Lynn Beauty

Discovery hypothesis. Staff may answer repeat price, duration, opening-hour and booking questions between appointments. Confirm two weeks of enquiry counts, active handling time, handover reasons, appointments and no-shows. Do not presume missed sales or after-hours losses.

Scope, phase one. One outlet, one WhatsApp Business number, approved English/Mandarin knowledge, basic service intent and preferred-time intake, human takeover, CRM lead record and weekly review. Appointments are requests until the calendar confirms actual availability. The pilot excludes autonomous discounts, payments, refunds, medical advice, marketing blasts and other outlets.

Deliverables. Signed tenant answer set; labelled answer sources; approved conversation boundaries; one operator queue with named cover; consent/opt-out handling; agreed CRM fields; customer-specific reporting; UAT evidence; operating and rollback instructions. A dashboard and required exports must be included if this is to become the pre-approved product; a mockup does not meet that objective.

Funding route. The future standard package could fit EDGE's GenAI Customer Engagement Chatbot activity. The current activity supports up to 50% for SMEs, with a S$15,000 activity ceiling and S$7,500 grant ceiling, through an approved package. CyberG7/Jarvis approval is unverified. A direct CyberG7 sale must therefore be treated as self-funded unless the exact package/vendor route is validated. A partner arrangement must retain the actual approved principal, allowed scope, delivery obligations and matching invoice; it cannot simply borrow a logo. EDGE chatbot activity

Economic screen. Assume 600 routine enquiries/month, with active handling reduced from four minutes to one. Capacity recovered = 600 × 3 ÷ 60 = 30 hours/month. At an assumed S$25/hour, annual capacity value is S$9,000. That alone does not cover the illustrated recurring subscription and maintenance of S$13,200. This is a deliberate screen: renegotiate scope/price or measure additional contribution before selling this budget as an attractive investment.

For example, 15 genuinely additional completed appointments/month at S$40 contribution each would add S$7,200/year. That is an unvalidated scenario, not forecast revenue. Measure incremental completed appointments against comparable periods; do not count all AI-assisted bookings as incremental. Avoid counting the same staff time twice.

Acceptance targets. At least 95% factual accuracy across 100 agreed cases; zero critical price/safety errors; 100% of agreed handovers delivered; 95% of eligible routine replies within 15 seconds; at least 20% reduction in measured active handling time. Review a two-week baseline and four-week pilot. Continue only if the observed business benefit and operator workload justify the price.

Owner hook. “Still answering prices and opening hours after your last appointment? Try a receptionist that replies from your approved answers and hands the rest to you.”

Client responsibilities. Approve factual answers and processing arrangements; provide authorised number/calendar/CRM access; name the operator and response coverage; confirm outlets and grant eligibility; approve the exact rollout. CyberG7 owns configuration, verification, reporting and support within the agreed scope. Rollback means switching automated replies off while preserving records and human access.

Proposal B — Kim Pong Engineering: Jarvis-assisted integrated ERP

Objective: connect the enquiry, quotation, service job and invoice so staff do not repeatedly reconstruct the same customer request.

Known business context. Kim Pong Engineering's public site describes commercial ACMV design/build, supply/install, A&A, preventive maintenance and testing/commissioning, with WhatsApp contact. Kim Pong Engineering

Discovery hypothesis. The business may have separate quote follow-up, technician scheduling and invoice preparation. Verify this with the office manager, service lead and finance owner. A competent existing system may already cover these needs.

Scope. One Singapore applicant entity and one ERP platform, with three implemented functions: customer engagement/sales; core service operations; finance. Jarvis prepares and routes actions. The ERP retains authoritative customer, order, asset, job and invoice records. A human approves commercial quotes and issued invoices. An accepted order creates the job; signed job completion enables the related invoice draft. WhatsApp provides intake and customer status within the approved policy.

Deliverables. Configured modules; a tested shared-ID data flow; migration reconciliation; WhatsApp intake adapter; action approvals and audit history; duplicate/retry controls; role-specific training; owner dashboard; deployment identifier; UAT and handover pack. Select the ERP foundation after requirements review. Do not assume Twenty/GHL plus a calendar is automatically a qualifying ERP.

Funding route. Candidate EDGE Integrated ERP, using CyberG7 as a non-pre-approved vendor only if a suitable approved package cannot provide the required functions. Document the comparison. Published assessment is ten weeks for that route. The activity's SME ceiling is 50% of assessed eligible costs, up to S$30,000, on a cost ceiling of S$60,000. Integrated ERP

Sector alternative to check. BCA's Built Environment PSG remains open and is scheduled to transition in December 2026. Confirm Kim Pong's eligibility and whether an appropriate pre-approved solution meets the needs before selecting EDGE. This exception does not allow a bespoke unapproved Jarvis package under BCA PSG. Do not seek both grants for the same costs. BCA PSG notice

The software access lines are proposed CyberG7 bundle pricing, not ERPNext/Odoo list prices. If the selected foundation is free software, do not fabricate a third-party licence cost. Distinguish genuine proprietary software access, hosting and implementation; recalculate the eligible basis from the real quote. The example grant becomes zero if no eligible line is accepted.

One-year benefits scenario. Assume 400 quote/follow-up events/month saving four active minutes each, 160 scheduling/completion events saving twelve minutes each, and 160 invoice events saving ten minutes each. This gives 85.33 hours/month. At S$28/hour, annual capacity value is S$28,672. It is not cash saved while payroll remains unchanged. The office must identify productive work replacing the recovered time, or independently evidence avoided overtime/outsourced costs. The companion application shows the calculation and conservative sensitivity.

Acceptance targets. Twenty representative jobs complete the entire quote–job–invoice chain; no duplicate financial record under retry; invoice references reconcile; at least 80% of eligible pilot jobs use the new process; at least 20% reduction in measured handling time. Finance signs off invoicing behaviour and migration totals. No promise of faster payment unless measured separately.

Sequence. Complete discovery and approved-package comparison; the SME submits before project commencement/payment; allow ten weeks for assessment; then an illustrative eight-week implementation and stabilisation period within the awarded dates. Starting after submission but before approval is possible only under an explicit client risk decision, not assumed here.

Owner hook. “From WhatsApp enquiry to service job to invoice, with one customer record and approvals at the right steps.”

Proposal C — GetSGPR: secure AI casework transformation

Objective: give consultants a reliable case checklist, organised follow-up and evidence-linked document review so qualified staff spend less time searching and chasing.

Known business context. GetSGPR publishes immigration consultancy services, a consultation journey and document-handling policies. Its website says sensitive identity documents should be handled through a controlled workspace and not sent over WhatsApp. GetSGPR

Discovery hypothesis. Staff may repeat document requests, reconcile versions and assemble timelines across inboxes, calendars and files. Establish current software, intake fields, permissions, document types, case volume and active handling time before confirming any problem.

Scope. Redesign intake → consultation → secure upload → case checklist → consultant review → client update. WhatsApp handles general service enquiries and carefully scoped status notifications. Sensitive documents stay in an authorised workspace. AI drafts a missing-document checklist, source-linked discrepancy notes and a consultant briefing. The consultant validates every substantive finding.

Deliverables. Documented current/target workflows; CRM–calendar–workspace integrations; role controls; source/version references; scoped AI extraction; protected audit trail; human approval for substantive advice; client training; baseline/pilot results; business-impact report and claim pack if approved. No autonomous eligibility determination, probability of ICA approval, legal advice, ICA submission or use of client material for general model training.

Funding route. Candidate EDGE Large-scale Digital Transformation only if this is a meaningful cross-system operating change. A small FAQ bot or isolated document summariser does not establish that scope. This activity asks for project/vendor proposals, last-year financial statements, three-year benefit projections and a one-year projection of specified new jobs. Do not invent hiring: zero is a valid forecast when truthful. Claims involve a panel auditor. The page gives SME support up to 50%; it does not publish a ten-week promise for this activity. Digital transformation

If all S$60,000 were accepted at 50%, support would be S$30,000. That is a conditional scenario. Each line, consultant qualification where relevant, cap treatment, audit charges and recurring costs require assessment and actual quotes. No maintenance, hardware, advertising or GST is embedded in this example. Costs of audit, continuing software and usage are still missing from the all-in budget.

Economic screen. Assume 80 cases/month with 45 active minutes saved per case. That is 60 hours/month; at S$35/hour, S$25,200/year of capacity. This fails to recover S$60,000 in one year on time savings alone. Even with the illustrated grant, it takes about 14.3 months to recover S$30,000 if that capacity is worth cash and before recurring/audit costs. Do not sell this forecast as ROI. Lower the scope or measure real contribution from additional completed cases, avoided outsourcing and reduced rework.

For application planning, an illustrative constant-volume projection is S$25,200 capacity in each of years 1–3, zero attributed revenue growth and zero newly created jobs above S$5,000/month until discovery supports another forecast. Revenue and overseas-sales projections must be completed honestly; a document automation tool does not justify export-growth claims.

Acceptance targets. Approved access boundaries pass all role tests; every extracted critical field links to its source; no unreviewed AI finding reaches a client; at least 90% completeness on an agreed labelled set; at least 20% reduction in active case administration; retained records remain recoverable. Use non-sensitive or properly authorised pilot data. A failed integration reverts to the existing human workflow.

Owner hook. “Your consultants review the case. Jarvis organises the checklist, the evidence and the next follow-up.”

Proposal D — GetSGPR: smaller self-funded AI adoption alternative

This is an alternative to Proposal C, not an extra claim on the same project.

Objective. Introduce a consultant-facing AI briefing tool and a business-specific enquiry assistant without rebuilding the whole operation.

Scope. Approved service knowledge; guarded drafts; source-linked internal briefing; human-only advice; controlled workspace; operator training and usage reporting. Existing software remains authoritative. Validate permissions and processing terms before supplying client documents to any model.

Illustrative price allocation. AI system subscription S$12,000; AI-specific workflow development S$8,000; AI-related training S$2,000; unrelated general website work S$2,000. Total S$24,000. Candidate AI portion S$22,000, subject to the accountant's classification and actual invoices. No grant is assumed.

At 400%, S$22,000 would generate a S$88,000 deduction. With a fully usable 17% marginal rate, the simplified total tax effect is S$14,960, of which S$11,220 exceeds ordinary 100% deductibility. Actual benefit depends on taxable income, exemptions, rebates, basis period and expense treatment; there is no supplier discount or cash conversion. IRAS EIS, company tax basics

Decision. Use this route when the smaller scope has its own economic justification, the buyer can use the tax deduction, and timing matters more than grant assessment. A loss-making client must not be sold an immediate saving it cannot use. CyberG7 supplies functional descriptions and itemised invoices; the client's accountant owns the tax position.

Questions to resolve before any proposal becomes a quote

1. What exact legal entity operates the business, pays the invoice and owns the deployment?
2. Which problem is verified by workload and handling-time evidence?
3. What systems, number ownership, data-processing terms and access already exist?
4. Which actions are pre-authorised by policy and which require transaction approval?
5. Can the client fund the full cost and continuing operation without the subsidy?
6. For ERP, what approved package fails which necessary requirement, and where is that documented?
7. What is the real eligible software/service allocation, including all continuing costs?
8. Who approves scope, prices, UAT, privacy arrangements and rollback on both sides?

CyberG7's case-study permission must be obtained separately. Buying a pilot is not consent to publish customer names, conversations or performance data.

Download this editable Markdown draft ↗

Worked application: Kim Pong Engineering, Jarvis-assisted integrated ERP

Draft for discovery and review, 1 October 2026. Not submitted or approved.

This packet demonstrates a substantive application structure for EDGE Integrated ERP using a non-pre-approved vendor. Kim Pong's business description is public. Every operating assumption and budget is illustrative. The packet cannot be submitted as factual until the applicant replaces assumptions, confirms the approved-package gap, obtains real quotations and completes its eligibility documents.

Official activity: Integrated Enterprise Resource Planning. Published assessment time: ten weeks for a non-pre-approved vendor. Publication of that time does not guarantee this application will be approved within it.

Before choosing this route, check Kim Pong's eligibility for the still-open BCA Built Environment PSG and suitable approved packages. BCA's transition to EDGE is scheduled for December 2026. This packet demonstrates the EDGE route; it does not establish why it is preferable to a currently available sector route. BCA notice

A. Applicant and project identity

Project title: Integrated enquiry, service operations and finance platform for ACMV service delivery.

B. Applicant-authored business narrative

Kim Pong Engineering provides ACMV design, installation, maintenance and related commercial services. We propose to integrate customer engagement and sales, core service operations, and finance on one ERP platform. Jarvis will assist staff with intake, draft preparation and follow-up under defined permissions; the ERP will retain the authoritative transaction records.

Our current workflow and handling-time measurements are attached in the baseline schedule. The project will link an enquiry to an approved quote, accepted service order, scheduled job, completion record and invoice draft. The same customer and job identifiers will follow the transaction, reducing repeated entry and allowing the owner to see incomplete work and billing exceptions.

We seek assessment of the selected non-pre-approved solution because the attached comparison establishes that suitable pre-approved packages cannot meet the required functions. Our cost-saving projection identifies its assumptions and supporting records. We will validate the deployed workflow through representative transactions, staff training and documented acceptance before submitting a claim.

The final paragraph is conditional drafting. Remove it or change vendor route if discovery does not establish the package gap. Do not assert the baseline is attached until it exists.

C. Three modules and shared data

The platform foundation and licences are to be selected after discovery. Supply its product/version, deployment plan, module configuration and data model in the real vendor proposal. The design must genuinely run these three functions on a single ERP platform. A diagram over disconnected applications is insufficient evidence.

WhatsApp is an intake/status channel. Jarvis is the workflow assistant. Neither is counted as a fourth core module to inflate the application. Existing GHL may remain the external lead entry layer if required, with tested reconciliation; only one authoritative ERP customer/order record is used.

Integration specification:

Customer enquiry
 → verified intake and approved knowledge
 → customer + opportunity
 → human-approved quotation
 → accepted order
 → scheduled service job
 → authorised completion evidence
 → invoice draft
 → finance approval and issue
 → payment/exception status for authorised staff

D. Required-function gap: complete before choosing this route

Use the current SMEs Go Digital ERP listing. Obtain written vendor answers wherever a feature is uncertain.

These are candidate discovery requirements, not established deficiencies in any approved product. Retain the package names, dated listing, quotations, capability demonstrations and correspondence. If a suitable approved vendor satisfies them, select that route. “CyberG7 has more agents” and “we want a bespoke AI OS” do not establish necessity.

E. Itemised quotation model

Assessed software basis assumed: S$18,000. At an assumed awarded rate of 50%, illustrated support = S$9,000. Client's cost after reimbursement = S$19,400, before GST, messaging/model usage and other unpriced costs. Require full upfront capacity; the grant is not a supplier advance.

These figures are a pricing model for discussion, not an offer or platform list price. The vendor quote must explain the actual product entitlement, number of users, usage limits, service term, ownership, hosting allocation and renewal price. Free underlying software must be identified honestly. Do not rename labour, hosting or maintenance to make it appear eligible. Recalculate grant support from the accepted Letter of Offer, and recalculate affordability if support is zero.

F. One-year benefit worksheet

Required official projection: expected cost savings for one year after completion. The following is a provisional, assumed calculation to demonstrate method.

Formula: volume × (before − after) ÷ 60. Use unrounded totals for calculations.

At an assumed loaded S$28/hour: S$28,672/year of capacity value. No payroll reduction is assumed. This must not be submitted as cash cost savings without a real cost reduction or defensible productive redeployment.

Example realisation plan, only if evidence supports it:

The realised-benefit scenario totals S$31,200/year; it is not additive to S$28,672. They are alternative valuations of the same hours. Revenue/contribution is separate from the required cost-saving figure. Do not count overtime avoided if it was never paid, or backlog contribution if demand does not exist.

Software plus maintenance renewal in this example is S$20,400/year before usage. Compare that to evidenced recurring benefit. A 50% haircut to time savings produces only S$14,336 capacity value/year; the example then fails on capacity alone. This sensitivity should change scope or pricing rather than disappear from the proposal.

Required baseline evidence: two representative weeks of timestamps and time studies, sample quote/job/invoice chains, rework log, volume reconciliation, loaded-rate basis and the applicant's endorsed realisation plan. Treat waiting time and active labour separately. Do not extrapolate peak-week volumes uncritically.

G. Implementation after the funding decision

Illustrative eight-week delivery after approval, within the awarded project period:

No project work/payment before submission. This example waits for approval before implementation. The official ERP completion window is eighteen months from approval; supported software is capped at twelve months. Final Letter of Offer dates govern. Scope/vendor changes are not casually interchangeable: this ERP page only permits changes to project-end and claim-due dates. Integrated ERP terms

H. UAT and business acceptance

1. Twenty representative transactions traverse sales, service completion and invoice draft with matching identifiers and amounts.
2. Repeat, delayed and failed webhook tests create no duplicate billable records; errors remain visible and recoverable.
3. Unapproved quotations/invoices remain drafts; the final action is bound to the approved values.
4. Customer status views reveal only authorised records; role tests prevent cross-client and inappropriate finance access.
5. All agreed unknown-price, complaint and safety cases reach a named human; human takeover suppresses further automated replies.
6. At least 80% of eligible pilot jobs use the process in the final two weeks, with at least 20% reduction in active handling time against the matched baseline.
7. Backup restoration, automated-reply shutdown and human fallback are demonstrated without losing transaction history.

The thresholds above are proposed CyberG7/client acceptance criteria, not official guaranteed-award thresholds. Go-live requires sign-off by sales, service operations, finance and the authorised client sponsor. Any critical factual, financial or access-control failure blocks it.

I. Submission and claim evidence index

The ERP claim is a single final claim after all approved deliverables are complete and paid in full. The portal accepts ERP claims from 1 November 2026; this project's own readiness depends on its awarded timeline. Payment follows claim assessment and approval. EDGE FAQ

CyberG7 supplies a defensible delivery proposal; it does not apply, pay or claim on the SME's behalf. The applicant keeps the funding decision, authorised access and declarations. A complete, accurate dossier gives the assessor something concrete to evaluate. It cannot substitute for eligible facts or an approved Letter of Offer.

Download this editable Markdown draft ↗

Jarvis advertisement playbook: sell the workflow, prove the benefit

Prepared 1 October 2026. Revises the user's supplied “PSG to EDGE” playbook around the updated GPT receptionist. All copy is a draft for review; no advertisement or message has been published or sent.

Campaign proposition: “Customer enquiries keep arriving while you work. Give them an approved answer, a useful next step and a person when needed.”

Use the receptionist demonstration as the destination. Keep its fictional salon label visible. Present booking, ERP integration and measured savings as separately demonstrated capabilities or proposed implementation work. A general Jarvis promise is not evidence that a particular client workflow already runs.

1. What to change in the supplied playbook

Authority: BCA notice, ERP route, chatbot ceiling, EDGE FAQ, IRAS EIS.

2. Ten hooks to test

Use two initial angles: repeated questions and owner control. Grant education supports discovery; it should not carry the core product claim.

3. Six channel drafts

Facebook / Instagram

Primary text: You are serving a customer. Another WhatsApp enquiry arrives. Then another asking the same price. CyberG7's Jarvis Receptionist uses your approved business information to handle routine questions and routes exceptions to your team. Explore the fictional salon demo, then discuss a pilot for your own workflow.

Headline: Still answering the same questions?

Description: Approved answers. Human handover.

CTA: Learn More → receptionist page. Offer a 20-minute discovery call only after the actual booking path and calendar are ready.

LinkedIn

An AI receptionist needs a clear job: answer routine questions from approved information, capture a useful enquiry and bring in a person when needed.

That is the starting point for CyberG7's Jarvis. The next step is a measured pilot: establish today's handling time, agree the boundaries and review the result before expanding into operations.

Considering EDGE? The funding route depends on the project scope and vendor conditions. We can help prepare the workflow and proposal; your authorised team makes the application.

Explore the labelled demo: Jarvis Receptionist.

Google Search

Headlines below are within 30 characters; descriptions are within 90 characters.

Initial search themes: WhatsApp AI receptionist Singapore; salon WhatsApp enquiries; AI customer service for SMEs. Give EDGE searches a dedicated explainer destination with current rules rather than implying Jarvis is already pre-approved.

WhatsApp follow-up draft

For contacts who have agreed to receive this type of message:

Hi [Name], it's [Sender] from CyberG7. We have updated the Jarvis WhatsApp receptionist demo. It shows approved answers and human handover using a fictional salon. If repetitive enquiries are taking your team's time, we can discuss a measured pilot. Would you like the demo link? Reply STOP to stop these updates.

This is draft copy, not permission to broadcast. Keep consent and opt-out records. The source playbook's blanket DNC instruction is incomplete: PDPC distinguishes B2B messages from marketing to individuals. Do not infer a B2B exemption merely because a recipient owns a business. PDPC business guidance, PDPC B2B clarification.

Thirty-second video

Do not show fabricated customer dashboards, bookings, counters or time-saving results. Record the demonstrated flow before producing the creative.

Email

Subject: Which WhatsApp question keeps interrupting your team?

Body: Hi [Name], CyberG7 has updated its Jarvis Receptionist demo. It shows how routine enquiries can use approved business answers, with exceptions handed to a person. If that fits your team's workload, the next step is a scoped pilot with a baseline and clear acceptance criteria. We can also assess the relevant funding route before implementation. Explore the fictional salon demo.

4. The systems that make the campaign work

GHL is the observed CRM direction in the estate; the Twenty/n8n/Chatwoot migration is a separate implementation decision. Choose one working record of opportunities rather than treating every listed hostname as a required sales system.

Judge campaigns by qualified opportunities, paying pilots, delivery contribution and renewal. Calculate acquisition cost using actual spend and won clients; set the affordable ceiling from conservative customer contribution and a chosen payback period. Cheap conversations without suitable clients do not establish success.

5. Readiness and the next commercial decision

Before paid acquisition, reconcile the public product claims, verify the actual booking path, publish the pricing basis, and demonstrate a complete tenant answer/handover flow. Begin with three paid pilots over 90 days; target positive contribution, two renewals or expansions and one permissioned case study. These are proposed management targets.

The grant strategy has two tracks: assess a genuine three-function ERP project where required-function gaps justify the other-vendor route; separately build the same-solution evidence for a receptionist package. IMDA's vendor criteria include at least 18 months of incorporation, five independent SME references with six months' current use, productivity evidence, financial stability and support capability. The normal evaluation process takes four to five months once ready. A broad estate or five unrelated website clients does not satisfy this evidence. IMDA criteria, pre-approval guide.

Use the companion proposals as discovery documents. Their economics deliberately distinguish recovered capacity from cash savings. Reduce scope or price when conservative benefit cannot support recurring cost; a grant should not conceal that decision.