Proposal · Confidential

Automation and growth systems for a company that is still small on purpose.

We build and operate our own AI products — customer messaging, social publishing, and SEO — and build custom systems on the same infrastructure. This document maps what already exists against what a newly established company in Malaysia actually needs in its first eighteen months.

Prepared forJason · Malaysia
Prepared byZhenheAI
DateAugust 2026
ScopeMarketing, customer service, internal ops

01 — The situation

A new company has the workload of a large one and the headcount of a small one.

The gap is not strategy. It is throughput. The three places it shows up first are the same everywhere:

Nobody answers the messages fast enough.
Enquiries arrive on Facebook, Instagram, LINE and the website at once. Reply time slips past the hour where intent is highest, and the lead cools.
Marketing stops the week you get busy.
Content, posting, SEO and ads all depend on one person having a quiet afternoon. That afternoon never arrives twice in a row.
Admin eats the founders.
Quotes, order notifications, reports, follow-ups, filing. Individually trivial, collectively a full-time role you have not hired yet.

None of these need a bigger team first. They need systems that already work, installed and operated by someone who has run them before.

02 — Who we are

We are not a reseller. Every system in this document is one we built and run.

ZhenheAI operates a portfolio of production AI products serving businesses in Taiwan — multi-channel customer service, social publishing, and SEO tooling. The same stack is what we deploy for clients. That means the answer to "can it do X" is verifiable, and the person configuring it is the person who wrote it.

5Production systems
4Native chat channels
3Social platforms published
100%Built in-house

Infrastructure runs on Cloudflare and Supabase with error monitoring, transactional email, and payment handling as shared internal services — so a client system is not a prototype held together for a demo.

03 — Capability map

Three layers. Start at whichever one is currently bleeding.

Each system below is independently deployable. Nothing here requires buying the whole set.

Layer A — Customer conversations

Replybot

AI customer service across every inbox

One dashboard for Facebook Messenger, Instagram DM, LINE, and an embeddable web chat widget. The AI answers from a knowledge base built by crawling your own website and documents, hands off to a human when it should, and keeps the full conversation history in one place.

Best forAny business receiving enquiries on more than one platform IncludesKnowledge base management, AI reply configuration, team seats, conversation history

Hermes

Your own hosted AI agent workspace

A private AI workspace on your own subdomain, isolated from other tenants, connected to LINE and your internal knowledge. Where Replybot handles inbound customers, Hermes is the assistant your team talks to — briefed on your business, available to everyone, without any data sitting in a shared consumer account.

Best forTeams that want AI internally without leaking company data into public tools IncludesDedicated tenant workspace, LINE integration, knowledge proposals, managed hosting

Layer B — Marketing, content and reach

Autopost

Scheduled social publishing with real analytics

Plan, schedule and publish to Facebook Pages, Instagram and Threads from one calendar, then read post performance and follower movement back in the same place. Built for the case where marketing is one person's third job — the queue keeps running on the weeks they disappear.

Best forConsistent presence without a daily posting habit IncludesMulti-account publishing, scheduling, post analytics, follower metrics

SEOHelp

SEO that reports in findings, not dashboards

A site audit engine that checks the things that actually move rankings — metadata, canonical and indexing directives, structured data, social preview cards, sitemap and robots health, hreflang for multi-language sites, and Core Web Vitals — and returns a prioritised fix list. Available as a dashboard, a command-line tool, and an interface AI agents can query directly.

Best forNew sites, or existing sites that have never been audited properly IncludesAudit engine, dashboard, CLI, edge rule injection for fixes you cannot ship quickly

ThreadScout

Find the conversations worth joining

Searches Threads for posts relevant to your keywords and drafts reply options for each one. By design you choose, edit, and post every reply yourself — the system finds and drafts, a human always decides. That constraint is deliberate and it is what keeps the account safe.

Best forFounder-led brands building presence through participation rather than broadcast IncludesKeyword monitoring, relevance ranking, drafted reply options, human-in-the-loop by design

Layer C — Internal operations

Custom builds

When nothing off the shelf fits

Multi-language websites, internal tools, customer portals, referral and commission tracking, payment and subscription flows, transactional email, and error monitoring. Delivered on the same infrastructure we run our own products on, and handed over with documentation rather than dependency.

Best forRequirements specific enough that a generic tool would be a compromise IncludesScoping, build, deployment, monitoring, handover documentation

Process automation

The manual weekly jobs, removed

Quotations, order and payment notifications, lead routing, follow-up sequences, recurring reports. We map the two or three processes costing the most hours, then build them to run on their own and alert a person only when something needs a decision.

Best forFounders still doing work that has no judgement in it IncludesProcess mapping, build, failure alerting, documentation

04 — Recommended starting points

What we would actually deploy first, by where the company is.

Our recommendation is to start with one layer, prove it for a quarter, then extend. Companies that install everything at once tend to operate none of it.

Stage The bottleneck Start with Add next
Just launched No inbound at all, and no presence to generate it SEOHelp audit on a properly built site, plus Autopost to establish a publishing rhythm ThreadScout, to join conversations instead of waiting to be found
Getting enquiries Messages arrive faster than anyone can answer them Replybot across every channel you are already on, with a knowledge base from your own site Process automation for quotations, order notifications and follow-up
Growing team Admin and repetition consuming people you hired to do other work Process automation across the two or three heaviest manual jobs Hermes as the internal assistant, plus reporting automation
Scaling Systems exist but nothing connects to anything else Custom build to join the pieces, with monitoring so failures surface before customers find them Referral, subscription and commission infrastructure as the model requires

05 — How we work

Build, operate, hand over.

STEP 1

Discovery, one session

We map where enquiries arrive, what happens by hand every week, and which of those is costing the most. The output is a written scope, not a slide.

STEP 2

Pilot on the single worst bottleneck

One system, one process, deployed to production and measured against how it worked before. Short enough that if we are wrong, it is cheap to have been wrong.

STEP 3

Extend across the layer

Once the pilot holds, the rest of that layer goes in. Channels, workflows, or content systems — whichever the pilot proved.

STEP 4

Operate, then hand over

We run and tune it while your team learns it, then hand over with documentation. Ongoing operation is an option, not a lock-in.

06 — Scope, honestly

Things worth knowing before the second meeting.

WhatsApp runs through SalesSmartly. Replybot covers Facebook, Instagram, LINE and web chat natively; WhatsApp is served by integrating SalesSmartly alongside it. Given how central WhatsApp is in Malaysia, we would scope that integration explicitly in the first conversation rather than leave it assumed.

Payments are region-specific. Our payment infrastructure includes a Taiwan-specific gateway and e-invoicing path that does not apply here; for Malaysia the route is Stripe, which we already run in production. Worth confirming early if payments are in scope.

ThreadScout will not auto-post. It searches and drafts; a person must approve and publish each reply. This is a product decision for platform-safety reasons and it is not configurable.

Our content systems were built Chinese-first. They handle English and multi-language sites, but for Malay-language content we would want a review pass on the first cycle rather than claiming it works untested.

07 — Next step

One discovery session, and a written scope you can act on either way.

Tell us which of the three bottlenecks in section 01 is the one hurting now. We will come back with a specific first deployment, a timeline, and commercial terms for that scope.