AGT-01
The Front Desk
Answers customers, students, or partners from your actual documents (policies, prices, procedures) in your tone, around the clock.
- Trained on
- Your docs, policies & price lists
- Hands off
- Anything unusual → a human, instantly
Practice 01 · Systems · Sheet S3.1
Artificial?
Applied
Intelligence.
Custom agents, knowledge systems, and document processing. Trained on your records, your rules, and your terminology. Not a generic chatbot with your logo on it.
Title block
- Sheet
- S3.1: Applied AI
- Practice
- 01 · Systems
- Scope
- Agents, knowledge & document AI
- First agent live
- 4–8 weeks, phased
- Training data
- Yours, and it stays yours
- Status
- Taking on projects
Fig. S3.2The symptoms
You’ve seen the demos. Now what?
Every tool you own suddenly says “AI-powered.” Meanwhile, back in the actual operation:
SYM.01
Your team answers the same questions, in the same words, every single day.
SYM.02
Documents arrive all day long, and humans retype what’s already written on them.
SYM.03
The knowledge that runs the company lives in two senior heads and a thousand unsearchable files.
SYM.04
You tried ChatGPT. Impressive, and useless, because it knows nothing about your business.
SYM.05
Every vendor pitch says “AI.” None of them can say what it would do for you, specifically.
The gap between AI hype and AI value is one word: context. Yours. That’s the part we build.
Fig. S3.3The roster
Meet your first digital staff.
We don’t sell “an AI.” We build agents with job descriptions: narrow, named roles that do real work and know exactly when to hand off to a person.
AGT-02
The Paperwork Clerk
Reads what arrives (invoices, forms, applications, claims), extracts what matters, and files it into your systems as structured data.
- Trained on
- Your document formats & fields
- Hands off
- Low-confidence reads → human review
AGT-03
The Archivist
Your internal brain. Staff ask questions in plain language; it answers from SOPs, handbooks, and history, with the source attached.
- Trained on
- Your SOPs, wikis & records
- Hands off
- No source found → says “I don’t know”
AGT-04
The Analyst
Watches the numbers across your systems, drafts the recurring reports, and flags what doesn’t look right, before the month ends.
- Trained on
- Your metrics & reporting formats
- Hands off
- Interpretation & decisions → you
AGT-05
The Back Office
Triage, routing, categorising, chasing: the process work that follows rules but eats hours. Done quietly, logged completely.
- Trained on
- Your workflows & rules
- Hands off
- Exceptions → escalated with context
AGT-0X
The Role You Actually Need
Agents are written to a job description, not picked off a menu. Tell us the work that eats your team’s week: we’ll draft the role.
- Trained on
- Your operation
- Hands off
- Never hired blind, scoped first
Fig. S3.4The division of labour
A Tuesday, redistributed.
The honest picture of applied AI isn’t a robot running your company. It’s a workday where the repetitive work quietly changes hands.
Worksheet: one Tuesday, observedObserved
08:52agent
41 overnight emails triaged: orders filed, invoices extracted, two complaints flagged
The complaints sit on top of the queue, history attached.
09:00agent
Yesterday’s invoices entered into accounting
Already done. This used to be somebody’s first two hours.
09:30agent
“What’s the refund policy on bulk orders?” (ninth time this week)
Answered instantly, source document linked.
10:15human
Negotiating terms with the new supplier
Leverage, relationships, judgment. Not agent work.
11:00agent
Monthly performance report across five systems
Drafted by the Analyst. A human reads it, then sends it.
14:00human
Deciding whether to expand the product line
The agent brings the numbers. The call is yours.
16:30human
A complaint that fits no pattern anyone has seen
Escalated by the Front Desk, context already assembled.
RepetitionGoes to the roster
JudgmentStays with people
That split is the whole philosophy. AI takes the work that follows rules, so your people can do the work that doesn’t.
Fig. S3.5A straight answer
The hype, filtered.
We’ve told clients not to buy AI from us. That’s exactly why the ones who do, trust it. Here’s the honest sieve we run every AI idea through:
Skip the AI project when
- You want AI because the board keeps saying the word
- The records it would learn from don’t exist digitally yet
- A human does the task better and it only takes minutes a week
- The process underneath is broken: automating a mess makes a faster mess
AI earns its keep when
- The same questions get answered by your team, daily, from documents
- Paper and PDFs arrive at volume and someone retypes them
- Critical knowledge is trapped in a few heads and a thousand files
- Decisions queue up waiting for reports somebody has to compile
Sometimes the honest first step is fixing the data underneath, not adding AI on top.Start with an AI-readiness conversation
Fig. S3.6Guardrails
Boring by design.
The exciting part of AI is what it can do. The valuable part is what it refuses to do. Every agent we ship is built inside these four walls:
Grounded, with receipts
Answers come from your documents and cite them. “I don’t know” is an allowed answer; invention is not.
Knows when to hand off
Every agent has explicit limits. Outside them, it escalates to a person, with the full context attached, not a cold start.
Your data stays yours
Your records never train public models. Access-controlled, isolated, and put in writing, before anything is built.
Everything on the record
Every answer and action is logged and reviewable. When someone asks “why did it do that?”, there’s an actual answer.
Fig. S3.7The build
From demo to duty.
No moonshots. One narrow, measurable job at a time: each agent proves itself before the roster grows.
01
Find the leverage
1–2 weeksWe audit your workflows for repetition, volume, and trapped knowledge, then pick the one use case that pays for itself first. Not the flashiest. The surest.
An AI opportunity map you own
02
Ground it
1–3 weeksThe agent is only as good as what it stands on. We connect and structure the documents, records, and systems that become its source of truth.
Your knowledge, connected and queryable
03
First agent on duty
2–4 weeksNarrow scope, real work, measured daily. It runs beside your team (answering, extracting, drafting) while accuracy is verified against humans.
One job, done by software, verifiably
04
Prove, then expand
per roleAccuracy and hours-returned are measured, not assumed. Only when a role earns its place does the roster grow to the next one.
A roster that grows on evidence
05
Keep it honest
ongoingYour business changes; an unmaintained agent drifts. Monitoring, evaluation, and retraining keep the answers as current as the operation.
AI that stays accurate as you evolve
Your data never stops being yours.
- Training data
- Prompts & configs
- Evaluation logs
- The off switch
Fig. S3.8Asked often
Questions, answered straight.
Will it make things up?
Not on our watch. Agents answer from your documents and cite their sources; when the source doesn’t exist, they say “I don’t know” and escalate to a person. Hallucination is what happens when AI is deployed without grounding; grounding is most of what we build.
Is our data used to train public AI models?
No. Your records stay isolated and access-controlled, they never train models outside your walls, and we put that in writing before anything is built.
Which AI models do you use?
The right one for each job, and that answer changes as models improve, which is exactly why your system is built model-agnostic. You own the architecture; the engine underneath can be upgraded without rebuilding.
Do we need to be “AI-ready” first?
Readiness mostly means your knowledge is digital and reachable, and connecting it is a stage of every build. Sometimes the audit shows the honest first step is fixing your data foundation instead; we’ll tell you, because that’s a better project to do first than a chatbot with amnesia.
Will this replace our staff?
It replaces tasks, not people. The pattern we see: the hours agents return get spent on the work that was always waiting. Customers, quality, decisions. The roster works for your people, not instead of them.
What does it cost?
Phased, like everything we do: the opportunity map is a small fixed fee, and every agent after that is quoted before it’s built, and measured against the hours it returns. If a role doesn’t pay for itself, we don’t recommend hiring it.
Cross-references
Read alongside this.
The pages that usually get opened next, and why.
Fig. S3.9Start here
What would AI actually do here?
That question has a specific answer. For your operation, your data, your volume. Bring one repetitive, painful job to the first conversation and we’ll map it. If AI isn’t your answer, we’ll tell you what is.
Malaysia · Working worldwide