What You Actually Build in Claude AI for Business Level 2: A One-Day Intermediate Training Summary
You do not leave Level 2 with notes. You leave with a written job description for an AI worker, a house style it has to follow, and a rule about when it must come back to you.
Level 2 is the middle day of the Claude AI for Business pathway, registered with HRD Corp as Claude AI for Business: Sales Enablement, Documents, Data & Excel at MQF Level 5. Level 1 taught you to work with Claude one conversation at a time. Level 2 is where you stop doing the work yourself and start giving it to somebody.
This post is the module-by-module record of that day: what each module covers as registered, what actually happens in the room, what you carry out at the end, and every prompt card handed out, in full.
Best Answer
Claude AI for Business Level 2 is a one-day, four-module programme carrying seven registered learning objectives, each with a named assessment. Module 1 produces a research brief, a proposal outline and a three-touch follow-up sequence. Module 2 produces a restructured document, a second-audience version and a six-slide pitch outline with speaker notes. Module 3 produces a cleaned data set, a board-level management narrative and three reusable reporting templates. Module 4 produces three explained Excel formulas, a working tracker, a tested Skill and a team-installable Plugin. Participants leave holding six deliverables and 14 prompt cards, all built on their own business process.
Most teams do not have an AI problem. They have a repetition problem that AI is now able to see. The same proposal gets rebuilt from scratch every time a lead comes in. Nobody knows who owes money until somebody sits down and builds the report. The same document gets rewritten three ways for staff, client and board. And the best people spend Monday morning making a report instead of selling.
“We have tried AI. We got impressive answers. Nothing in the business changed.”
That sentence is the reason this day exists, and the obvious answer — better prompts — is not enough on its own. A better prompt improves one answer. What changes a business is writing the instruction down so it runs the same way twice, handing it to somebody else, and deciding in advance where a human has to stop it. That is the whole of Level 2, and it is why the day ends with a Plugin and a supervision rule rather than a longer list of tips.
Who This Is For
The registered target audience is plain about it: Best for sales teams, operations managers, HR staff, admin teams and business analysts who manage client documents, proposals, data summaries and Excel spreadsheets.
In practice, the room fills with people who produce the paperwork the business runs on. If two or more of these are true, this is the day for you.
- You rebuild the same proposal or quotation from scratch every time a lead comes in.
- Nobody knows who owes you money until someone sits down and builds the report.
- Your best people spend Monday morning making a report instead of selling.
- The same document gets rewritten three ways for staff, client and board.
- You have tried AI, got impressive answers, and nothing in the business changed.
It is not a tour of AI features, and it is not a coding day. Nothing is written in code, and agents, connectors built on your own systems and anything inside Claude Code are deliberately out of scope — those belong to Level 3 and Level 4. It is also not a day where you watch. Every assessed artefact is produced by you, on a process you named that morning.
Before the Training Day
Required before the day — these are prerequisites, not recommendations:
- A paid Claude plan, Pro or higher. Claude in Chrome and Claude Cowork are both paid-only. A participant on the free tier cannot complete Module 1 or Module 3.
- The Claude in Chrome extension installed and signed in. Signed in, not merely installed.
- A laptop that can run Claude Cowork. On Mac it works without setup. On Windows it needs the Virtual Machine Platform feature enabled, which requires administrator rights, and not every Windows edition and device is supported.
- Your IT department warned, if your laptop is managed. Tell us at booking — enabling virtualisation usually needs IT involvement, and we send your IT team the instructions in advance.
- Your chosen track confirmed, plus a real or sample prospect brief, a business document, and a dataset, report or spreadsheet to work on. Participant-provided wherever possible.
- Laptop, internet access and the Chrome browser.
A setup and plan check is issued 48 hours before the programme, covering the paid plan, Claude in Chrome, and a confirmed test that Cowork launches on your machine. A short readiness questionnaire goes out three to five days before, so the pacing on the day matches the room.
The Day at a Glance
Two end-to-end pipelines rather than a feature tour. The morning is Proposal-to-Presentation: research, proposal, report, presentation. The afternoon is Data-to-Boardroom: raw data, summary, narrative, toolkit.
Before you write a word
And the follow-up
Six slides, with notes
Data to a decision
A sheet a worker can read
Installed by the team
| Module | Registered title | Duration | What you leave the module holding |
|---|---|---|---|
| 1 | AI for Sales Enablement: Proposals, Pitches & Client Communications | 1.5 hours (9:00 – 10:30 AM) |
A prospect or benchmark research brief, a proposal or hiring-case outline, and a three-touch follow-up sequence. |
| 2 | AI for Documents & Presentations: Reports, Briefings & Slide Outlines | 1.5 hours (11:00 AM – 12:30 PM) |
A restructured business document, the same facts written for a second audience, and a six-slide presentation outline with speaker notes. |
| 3 | AI for Data & Reporting: Summaries, KPI Narratives & Management Reports | 1.5 hours (1:30 – 3:00 PM) |
A cleaned data set, a written data summary, a board-level management narrative, and three reporting templates you reuse every month. |
| 4 | AI for Excel & Programme Close: Formulas, Spreadsheet Planning & Week 1 Commitment | 1.5 hours (3:30 – 5:00 PM) |
Three explained Excel formulas, a spreadsheet structure plan, your own working tracker, a tested Skill, a packaged Plugin and a written Week 1 commitment. |
Four modules of 1.5 hours across a 9:00 AM – 5:00 PM day, with breaks and lunch between them. The delivery ratio is participatory at 80 percent or more, lecture at 20 percent or less. Class size up to 20; above 15 a second HRD Corp certified trainer co-trains.
The Registered Learning Objectives
Seven objectives, reproduced exactly as registered and approved with HRD Corp. Each one is delivered in a named module and assessed by a named artefact — there is no objective on this list that is only talked about.
| # | Objective Registered | Module | Assessed by |
|---|---|---|---|
| 1 | Apply Claude in Chrome to research a prospect or competitor before drafting a proposal. | 1 | Research brief |
| 2 | Produce business proposals, quotations, pitch decks and client presentation outlines using Claude. | 1, 2 | Proposal or pitch outline against the sales brief template |
| 3 | Draft follow-up sequences, client briefing notes and sales communications tailored to different audiences. | 1 | Follow-up sequence exercise |
| 4 | Adapt a single business document for internal, client-facing and board-level audiences. | 2 | Audience-adaptation checklist |
| 5 | Convert raw data and report text into written management summaries and KPI narratives. | 3 | Data summary and management narrative |
| 6 | Construct and explain Excel formulas from plain-language descriptions, and plan spreadsheet structures for trackers and dashboards. | 4 | Three formulas, the structure plan and the prompt card |
| 7 | Package a proven reporting workflow as a Claude Skill and as a team-installable Plugin. | 4 | Packaged Plugin |
AI for Sales Enablement: Proposals, Pitches & Client Communications
The one idea: the quality of any AI output is decided before you ask for it, not during.
What this module covers Registered
| Topics covered |
|
| Learning activities |
|
| Assessment method | Review of prospect/competitor research brief and proposal/pitch outline against a sales brief template |
What actually happens in the room
- You name your process before you touch a tool. One track, one process you actually want to fix, written in a sentence and said out loud to your neighbour. Everything you build for the rest of the day runs on that sentence, so “improve efficiency” is sent back.
- Everyone types a deliberately bad prompt first. “Write me a proposal.” Two of the answers are read out. The contrast with what comes next is the lesson, which is why the bad version is never skipped.
- Then Claude interviews you before it answers. The framework is built live, one part at a time, and the room adds a line inviting Claude to ask up to three questions first. The only question taken from the floor afterwards is “what did it ask you that you had not thought of?” — and the answer is usually the thing that changes the document.
- A real prospect, researched live, screens down. Hands off keyboards. The trainer narrates the thinking rather than the clicking — why ask what changed recently, why that produces an opening line that could not be sent to anyone else — and lands a one-page brief.
- The same three steps, on a completely different subject. The demonstration is a sales prospect. The exercise the room does together is a salary benchmark. Identical three steps, only the subject changed — which is how the day transfers to the people whose function is neither.
- The lazy follow-up, typed badly on purpose. “Just following up, any update?” The room is asked what the customer could possibly reply to. Nothing. It is then rewritten to carry one new fact, and all three touches are drafted in one go so the sequence has a shape.
- Cowork, demonstrated: a whole proposal packet from one instruction. A folder in, one instruction, and a proposal, covering email and follow-up sequence come back. That is a worker doing in one instruction what the room did by hand an hour earlier.
- P — Purpose. What business outcome are you trying to achieve?
- R — Role. What expert role should Claude adopt?
- I — Instruction + Interview. What should Claude do — and what should it ask you back?
- S — Structure. How should the output be organised?
- M+ — Measure. What defines a successful result?
The plus is not a sixth letter. It rides on Measure, and underneath it sit six enhancers — context, audience, tone and style, constraints, grounding and source, quality check. Level 1 stops at the five letters. The enhancers are the Level 2 layer. And the “I” is an invitation, not just an instruction:
Before you produce anything, ask me up to three questions about anything you need and I have not told you. Wait for my answers.
Add that line whenever the job matters and you are not sure you have said enough — which, for a first proposal, a salary band or a board narrative, is most of the time.
Every workflow you build today has a gate. If you cannot name where a human stops it, you have not finished designing it. Research needs none — nothing leaves the building. A proposal needs one. A follow-up needs one per message.
AI for Documents & Presentations: Reports, Briefings & Slide Outlines
The one idea: you do not write three documents — you write the facts once, then decide who is reading.
What this module covers Registered
| Topics covered |
|
| Learning activities |
|
| Assessment method | Review of business document draft and presentation outline against an audience-adaptation checklist |
What actually happens in the room
- A genuinely bad document, read aloud until the point arrives. It does not arrive. Several hundred words of relationship-building go by before the letter says anything, and the prices that matter are buried on the second page.
- Ask for the structure. Approve it. Then ask for the prose. Two sends, never one. The structure is edited on screen before a sentence exists — one heading moved, one deleted. It costs fifteen seconds and it is the only place you still have control.
- One policy change, read out twice. The staff version and the board version, back to back, out loud. You cannot hear that a board version is only the staff version in a suit by looking at a screen. You can hear it in a room.
- Six slides. Not seven. Headlines of eight words or fewer, bullets of ten, speaker notes of forty to sixty words that do not repeat the bullets. Somebody always produces eleven slides and has to cut. The constraint is what forces the thinking.
- Never open with your company profile. The buyer is deciding whether you understand them, and that is settled by slide two — the cost of leaving the problem alone, which is usually the one thing the proposal never actually stated.
- Your own document, from the last month, the messier the better. Then the same facts written for a second reader. The most common gap is a “board version” that is the staff version with longer words, and it gets pushed back on by name.
- The room does the arithmetic on the letter’s own claim. The restructured version repeats every original error in better English, on a page that looks more trustworthy. Nobody is told the answer — they find it, which is the difference between a fact and a habit.
- Their situation, in their words. Not your company profile.
- The cost of leaving it alone, quantified if you can.
- What you propose, in three moves.
- Why you: one proof point, not a client logo wall.
- Investment, and what is included.
- The next step, with a date on it.
The same shape works for an internal approval deck. Slide 1 has to fail the swap test — you could not paste another company’s name into it.
AI made that document more persuasive without making it more true. Claude restructured your facts. It did not verify them, and your name is still on it.
AI for Data & Reporting: Summaries, KPI Narratives & Management Reports
The one idea: a table tells you what happened; a narrative tells you what to do about it.
What this module covers Registered
| Topics covered |
|
| Learning activities |
|
| Assessment method | Review of data summary, management narrative and 3 reporting prompt templates |
What actually happens in the room
- The sample export is dirty on purpose, and nobody cleans it for you. One source spelled four ways, dates stored as text, duplicate records, blank statuses. If the sample were clean you would learn analysis on data that does not resemble yours, and blame Claude the first time you tried your own export.
- Claude proposes the cleaning. You approve it. It lists the duplicates and how it spotted them, proposes one standard name per group, flags the dates that are not dates, and says what a blank probably means. It changes nothing until you say so. That distinction is the card.
- The business question, not “analyse this”. Which source produced revenue, not which produced the most leads. The finding is buried in the data twice over, and it is a participant who finds it, not the trainer.
- The question nobody in the business has ever answered. What does a blank clock-out mean? Forgot, worked past midnight, or the system dropped it. All three are plausible and nobody has ever decided — which is the module’s whole point, made concrete.
- Three templates written, not discussed. Summary, exception and narrative. Each names its source file, its date range and its owner. The trainer walks the room and physically checks all three exist, because this is the beat most likely to be faked.
- The exception report gets calibrated, then re-run. Exceptions should land at roughly 5 to 15 percent of records. Above a quarter the threshold is too tight and nobody acts; near zero it is too loose. Changing the rule and saying what you changed is the competency — not the first output.
- An honest slide about what Claude cannot reach. Most business connectors let Claude read, summarise, draft and flag — not post or update. Nothing fires automatically when something happens in another system. It is said plainly, because a participant who leaves believing otherwise will be angry in a week.
- The number that changed, and by how much.
- One sentence on the cause — the cause, not the correlation.
- What it means for cash, capacity or risk.
- The decision you are asking for, or the decision you already took.
A board narrative is not a shorter report. It is a different document with a different job: it asks for a decision. A narrative with no decision in it is a summary, and summaries do not get read twice.
Analysis is 20% of the work. The other 80% is agreeing what a row means. Do that first, out loud, before you ask a single question.
AI for Excel & Programme Close: Formulas, Spreadsheet Planning & Week 1 Commitment
The one idea: a formula in a badly shaped sheet is a faster wrong answer.
What this module covers Registered
| Topics covered |
|
| Learning activities |
|
| Assessment method |
|
What actually happens in the room
- The module opens on your own screens, not a slide. Three people show what they built after lunch, sixty seconds each. Peer proof beats trainer proof, and at half past three the room needs restarting on its own work.
- One row per what? The grain question, asked before a single column is named. One row per lead, never per conversation. One row per staff per day, never per person. Get it wrong and every formula after it is wrong politely.
- Three formulas — and you have to explain them out loud. The formula and the explanation are asked for together, every time. A participant who cannot explain their formula has not passed the card, and that is said before anyone starts. Test on five rows where you already know the answer, before trusting five hundred.
- Two sheets, the same records, only the shape differs. Nothing is dirty in either — that was fixed before lunch. One of them cannot answer a simple question without being retyped; the other answers it with one formula. The hand-typed total at the bottom of the bad one is wrong, and only slightly wrong, which is the kind nobody spots.
- Colour is not a status. Somebody is asked what the green row means. They know. Then they are asked to show where the sheet says so. It does not — it says it in somebody’s head, and an automation cannot read that.
- A marketplace Skill that does not fit here. A well-made, highly-rated Skill written by someone who has never run a business in Malaysia. The room finds what breaks. The cheap catch is the currency; the expensive one is a rule about weekends, in a country where the weekend is not the same in every state. Adapting is the skill. Installing is not.
- Package it, then the colleague test. Run it twice on different inputs first — if the output differs in shape, the instruction is not finished. Then it is packaged, and the measure is simple: a colleague installs it, runs it once, and does not message you.
- One row per event — decide the grain first, everything follows.
- Headers in row 1, data from row 2.
- No merged cells, no blank spacer rows.
- One fact per column.
- Dates as dates.
- Status from a fixed list, never colour.
- A raw tab and a summary tab — never a total inside the data.
Every rule here exists because a real workflow broke on it. A merged cell is the single most common reason an automation silently returns nothing.
The supervision rule — written down before anyone leaves the room.
- Runs unwatched: reading, summarising, drafting, flagging — nothing leaves the building.
- Comes back to you: anything that reaches a customer, a candidate, or payroll.
- Never unsupervised: price, salary, disciplinary wording, contract terms, personal data.
- Who may install it: named people, and what the Plugin may never do.
Generic answers do not count. “Anything important” is not a line — you have to name the thing.
The moment it becomes a Plugin, you stop being the only person who can do this job.
What Every Participant Carries Out
Six deliverables, every one of them built during the session on the participant’s own process. Not one of them is a demonstration file.
1. Client proposal packet & research brief
A one-page brief on a real prospect or market, and the proposal or hiring case that comes out of it — assembled, in the demonstration, from one delegated instruction.
Module 1 · assessed2. Six-slide pitch deck outline
Six slides in a fixed order with speaker notes of forty to sixty words each, cut from a document you already had rather than written from nothing.
Module 2 · assessed3. Executive management data narrative
One page off your own numbers: what changed, why, what it means for cash, capacity or risk, and the decision being asked for.
Module 3 · assessed4. Custom Excel formula toolkit
Three formulas on your own tracker, each with a plain-English explanation you can give without opening Claude, plus the completed prompt card.
Module 4 · assessed5. Installable team Plugin
The instruction you reused all day, written down as a Skill, tested twice, and packaged so a colleague can install it and run it without asking you how it works.
Module 4 · assessed6. Week 1 commitment
One process, one worker, a first run date and the one person you will show it to. The 30 and 90-day follow-up is measured against this exact page.
Module 4 · collectedAlongside the assessed six you keep the working material: a three-touch follow-up sequence, a cleaned data set, the same document written for a second readership, three reporting templates that run every month unchanged, a spreadsheet structure plan, and your own four written supervision lines. There is also a one-page map of your process — trigger, workflow step, gate, output — built up across the four modules and photographed before you leave, which is the thing most participants show their manager first.
Choose Your Track
Everyone picks one process before the day and produces that process’s real documents, reports and spreadsheets throughout. The six tracks are registered as part of the programme.
| Track | The process Registered | What you will produce |
|---|---|---|
| T1 · Sales & Marketing | Lead capture to quotation follow-up | Proposal, quotation, follow-up sequence, pipeline tracker |
| T2 · Order Fulfilment & Inventory | Order to delivery, stock visibility | Order confirmation pack, stock report, reorder tracker |
| T3 · Accounting & Invoicing | Invoicing, collections, aged receivables | Collection letters, aged-receivables summary, cash tracker |
| T4 · HR, Payroll & Attendance | Leave, claims, attendance, payroll prep | Staff briefing, attendance summary, claims tracker |
| T5 · Project Management & Timesheets | Job tracking, time capture, billing | Project status report, timesheet summary, job-cost tracker |
| T6 · Service & Workshop (3S) | Job cards, parts, service history | Job-card pack, service throughput report, parts tracker |
The trainer demonstrates in Track 1 and the room works together in Track 4 — deliberately, and then reversed after the break. That pairing is the transfer mechanism: identical steps, a sales document and an HR document, so the day lands for the participant whose function is neither. If your track is one of the other four, you sit with whichever is closer and fill your own column throughout. In-house cohorts can also run vertical tracks — FMCG field sales, clinic management, fleet management and GPS, or manufacturing with bill of materials and a master production schedule.
The Sample Businesses and the Data Rule
Three Malaysian SMEs run through the whole card pack, so participants watch one business’s documents develop across the day rather than meeting four disconnected examples. None of them is a real client. They are written to be real-shaped — the right size, the right mess, the right arguments — and every fault planted in their files is deliberate.
- Sinar Teguh Engineering — industrial pipe fabrication and on-site maintenance for petrochemical plants, in Pasir Gudang, Johor. Enquiries are up; quotation conversion is down. A large shutdown-maintenance quotation has gone quiet after “noted, will discuss internally”.
- Kopi Wangsa Group — four kopitiam outlets across KL and Selangor, shift-based, with high staff turnover. Paper claim forms are moving to a monthly deadline, the owner and the clerk have written that deadline down differently, and the attendance export has more staff IDs in it than there are staff.
- Anugerah Parts Trading — automotive spare parts in Butterworth, Penang, selling walk-in and over WhatsApp. Their supplier price-increase letter spends two-thirds of a page being polite before it reaches the point, and its own summary of the increase does not match the prices listed underneath it.
Participants bring their own real document, prospect and export for the assessed work, because the point of the day is that the artefacts are usable on Monday. The rule stated in the room is consent first: no customer or staff detail enters a shared file before that check, and staff-facing documents name roles rather than individuals unless an individual has to be named.
Claude drafts. You decide whether it sends. That distinction is the whole of Level 2.
The Complete Prompt Card Set
These are the cards participants are handed on the day — all 14 of them, across the four modules plus the rescue card that stays on the table from nine o’clock. Every one is here in full, so you can read what the day actually asks people to type and copy any prompt straight into Claude.
- Replace everything in {CURLY BRACKETS} with your own detail. This pack marks its slots with braces rather than the angle brackets you may have seen on other card sets. Leave a slot unfilled and Claude will invent something, and it will be wrong in a way that is hard to spot.
- Fictional or sandbox data only. Several cards read a whole file. In the room that file is your own, with the consent check already done. Copying a card from this page and pointing it at a live customer list or a real staff claim creates exactly the problem the training was designed to avoid.
Click any card to open it. The cue on each card says who is typing — watch the trainer, everyone together, or on your own — rather than a clock time.
Module 1 · AI for Sales Enablement 3 cards
Research before you write. You leave the module with a one-page brief on a real prospect or market, the proposal or hiring case built from it, and a three-touch follow-up sequence that gives the customer something to reply to.
P1.1The PRISM+ skeletonCLAUDE DESKTOPTHE WHOLE ROOM
P — PURPOSE
I need to {WHAT_YOU_ARE_TRYING_TO_ACHIEVE}, so that
{THE_DECISION_OR_ACTION_IT_LEADS_TO}.
R — ROLE
You are {WHO_CLAUDE_SHOULD_BE} with experience in
{THE_RELEVANT_DOMAIN} in Malaysia.
I — INSTRUCTION + INTERVIEW
{THE_EXACT_JOB — one sentence, one verb}.
Before you produce anything, ask me up to three questions about
anything you need and I have not told you. Wait for my answers.
S — STRUCTURE
Give it to me as {FORMAT: e.g. a one-page brief with 4 headings /
a table with these columns / 6 slides with speaker notes}.
M+ — MEASURE
It is good enough when {THE_TEST — e.g. I could hand it to my
manager without editing / every figure traces to the file I gave you /
it fits on one page}.
and underneath M, the enhancers you need — not all six:
Context: {WHAT_YOU_MUST_KNOW_THAT_ISN'T_OBVIOUS}
Audience: {WHO_READS_IT}
Tone: {HOUSE_VOICE}
Constraints: {LENGTH / LANGUAGE / WHAT_TO_LEAVE_OUT}
Grounding: work only from {THE_FILE_OR_PAGE_I_GAVE_YOU}
Quality check: before showing me, check {THE_ONE_THING_THAT_IS_USUALLY_WRONG}
P1.2The research briefCLAUDE IN CHROMEYOUR OWN PROCESST1 AND T4 VERSIONSASSESSED
The Track 1 version — research a prospect before you write a word.
P — PURPOSE
I am preparing a first proposal for this company and I need to
understand them before I write a word, so my opening line proves I
did the work.
R — ROLE
You are a B2B research analyst who prepares call briefs for a
Malaysian {YOUR_INDUSTRY} company.
I — INSTRUCTION + INTERVIEW
Read the page I am on and tell me: what they sell, what changed for
them in the last 12 months, who is likely to sign, and what they
would care about that I can actually help with.
S — STRUCTURE
A one-page brief, four headings, maximum 6 bullets each. Add a final
line: "Opening sentence I would use:".
M+ — MEASURE
Good when a colleague who has never heard of this company could walk
into the meeting with it.
+ Grounding: use only what is on the pages I show you. If you are not
sure of something, write "not found" — do not guess.
+ Constraints: no marketing adjectives. Facts and one inference.
+ Audience: me, before a first call.The Track 4 version — change these three lines, and nothing else.
P — I am setting a salary band for {ROLE_TITLE} and need to defend
the number to my boss.
I — Read these market pages and tell me: the typical range for this
role in {STATE_OR_CITY}, what benefits are standard, and what a
good candidate will ask for that we do not currently offer.
M — Good when I could quote a range in a salary conversation and
explain where it came from.P1.3The three-touch follow-upCLAUDE DESKTOPYOUR OWN PROCESS
P — PURPOSE
{PROSPECT_OR_PERSON} has not replied to {WHAT_I_SENT} sent on
{DATE}. I need a follow-up sequence that gets a decision without
making me sound desperate.
R — ROLE
You are a Malaysian B2B salesperson who follows up firmly and
politely, and who never wastes a customer's time.
I — INSTRUCTION + INTERVIEW
Draft three follow-up messages: day 3, day 10 and day 21 after the
original. Each message must ADD something new — an answer, a
relevant case, a deadline or a decision. None may say "just
following up".
S — STRUCTURE
Three messages. Subject line + body. Under 90 words each. Label
them Day 3, Day 10, Day 21.
M+ — MEASURE
Good when each message could stand alone as a reason to write, and
the third one makes it easy for them to say no.
+ Context: {WHAT_THEY_ASKED_ABOUT / WHAT_STALLED_IT}
+ Tone: warm, direct, no chasing language.
+ Constraints: English. Day 3 by {CHANNEL: email/WhatsApp}, Day 10 by
email, Day 21 by {CHANNEL}.
+ Quality check: before showing me, confirm none of the three
contains the phrase "following up" or "checking in".
Module 2 · AI for Documents & Presentations 3 cards
One source, many audiences. You leave the module with a restructured document whose decision is visible by the second heading, the same facts written for a second readership, and a six-slide outline with speaker notes.
P2.1Proposal to six-slide deckCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
P — PURPOSE
I have to present {THE_PROPOSAL} to {WHO} on {DATE}. I need a deck
outline that survives a real meeting, not a document read aloud.
R — ROLE
You are a presentation editor who cuts decks down, not one who pads
them out.
I — INSTRUCTION + INTERVIEW
Turn the attached proposal into exactly SIX slides, in this order:
1 their situation in their own words · 2 the cost of leaving it alone ·
3 what I propose, in three moves · 4 why us — ONE proof point ·
5 investment and what is included · 6 the next step with a date.
Invent nothing that is not in the proposal.
S — STRUCTURE
For each slide: a headline of 8 words or fewer, at most 3 bullets of
at most 10 words, and SPEAKER NOTES of 40–60 words saying what I say
out loud — which must not repeat the bullets.
M+ — MEASURE
Good when someone reading only the speaker notes would understand
the whole argument, and slide 1 could not be about any other company.
+ Grounding: use only the attached proposal. If something is missing,
write "[MISSING: ...]" rather than filling the gap.
+ Audience: {DECISION_MAKER_ROLE}.
+ Constraints: exactly 6 slides. No company-profile slide.
+ Quality check: confirm no bullet exceeds 10 words before showing me.
P2.2Structure first, then proseCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
Send 1 — the structure only. No prose.
SEND 1 — STRUCTURE ONLY
P — PURPOSE
This document is not landing with {WHO_READS_IT}. I need it
restructured so the decision is obvious in the first ten seconds.
R — ROLE
You are an editor who fixes structure before wording.
I — INSTRUCTION + INTERVIEW
Do NOT rewrite anything yet. Give me a proposed structure only:
the headings, in order, with one line each saying what goes there
and why it is in that position.
S — STRUCTURE
A numbered outline. Maximum 7 headings.
M+ — MEASURE
Good when I can see the decision being asked for by heading 2.
+ Grounding: work only from the attached document. Change no facts
and no figures.
+ Constraints: structure only. No prose.Send 2 — now the prose, and only after you have approved or edited that structure.
SEND 2 — NOW THE PROSE (only after you have approved or edited the structure)
Write the document to that structure. Keep every number exactly as
it appears in my original — if a figure is unclear, flag it as
"[CHECK: ...]" rather than adjusting it. One idea per paragraph.
Target {WORD_COUNT} words.P2.3One source, three audiencesCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
P — PURPOSE
The same facts have to reach three different readers this week, and
right now I write it three times from scratch.
R — ROLE
You are a business communications editor who changes framing without
changing facts.
I — INSTRUCTION + INTERVIEW
From the document above, produce THREE versions:
(1) STAFF — plain language, what changes for you, when it starts,
who to ask. No policy or clause numbers.
(2) MANAGER — how to answer the four questions their team will ask,
and what to escalate.
(3) OWNER / BOARD — cost impact, compliance position, and the risk
if we do nothing. Ends with the decision being asked for.
S — STRUCTURE
Version 1: max 150 words. Version 2: max 250 words plus a Q&A of 4.
Version 3: exactly one page, ending in a single decision line.
M+ — MEASURE
Good when the three versions share every fact and share almost no
sentences.
+ Grounding: no fact may appear in one version that is absent from
the source.
+ Tone: version 1 warm, version 2 practical, version 3 neutral.
+ Constraints: English. {ADD_BM_VERSION_IF_NEEDED}.
+ Quality check: list any figure that appears in one version but not
the others, and tell me whether that is deliberate.
Module 3 · AI for Data & Reporting 4 cards
Data does not become a decision until somebody writes the sentence. Cleaning comes before counting, so these cards run in that order — clean, narrate, flag the exceptions, then turn the whole thing into a standing instruction.
P3.2Clean before you countCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
P — PURPOSE
Before I analyse anything I need to agree what a row in this file
means, because right now the same customer appears three times and
I do not trust any total.
R — ROLE
You are a data analyst who cleans before counting and who tells me
what they changed.
I — INSTRUCTION + INTERVIEW
Read the attached file and, before any analysis:
1. list the duplicate records you can see and how you identified them
2. list every value used for {THE_MESSY_COLUMN — e.g. Source} and
propose ONE standard name for each group
3. tell me which dates are not real dates
4. tell me what a blank in {THE_BLANK_COLUMN} probably means, and
what you would need to confirm it
Then produce a written summary of what the file actually contains.
S — STRUCTURE
Four short sections matching the four points, then a summary of no
more than 200 words. Show counts, not percentages, at this stage.
M+ — MEASURE
Good when I can state the total number of real, unique records with
a straight face.
+ Grounding: work only from the file. Do not fill gaps from general
knowledge.
+ Constraints: change nothing yet — propose, do not apply.
+ Quality check: tell me the one assumption you made that, if wrong,
would change the totals most.
P3.1The board narrativeCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
P — PURPOSE
{WHO_DECIDES} has ten minutes and will not read a table. I need the
one page that tells them what to do about these numbers.
R — ROLE
You are a finance-literate business writer who writes for owners,
not for analysts.
I — INSTRUCTION + INTERVIEW
Write a management narrative from the analysis above, in this order:
the number that changed and by how much · one sentence on the cause,
not the correlation · what it means for cash, capacity or risk · the
decision I am asking for or the decision I already took.
S — STRUCTURE
One page. Four short paragraphs. A single bold decision line at the
end. No tables, no bullet lists.
M+ — MEASURE
Good when the reader could act on it without opening the spreadsheet,
and every figure in it appears in the source data.
+ Grounding: every number must come from the data I gave you. If you
cannot source a figure, leave it out and say so.
+ Audience: {OWNER / GM / BOARD}.
+ Tone: neutral. No good-news framing, no alarm.
+ Constraints: one page. Plain language — no "attrition", no "synergies".
+ Quality check: list every number you used and where it came from.
P3.3The exception reportCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
P — PURPOSE
Nobody reads the full report. I need only the things that are
outside normal, so somebody actually acts on them.
R — ROLE
You are an exception reporter. You do not summarise everything —
you surface only what breaches a rule I set.
I — INSTRUCTION + INTERVIEW
Using the cleaned data above, list every record where:
- {RULE_1 — e.g. no contact for more than 14 days}
- {RULE_2 — e.g. weekly hours above 45}
- {RULE_3 — e.g. status blank for more than 7 days}
For each one, give me the record, which rule it breached, and the
single next action.
S — STRUCTURE
A table: Record · Rule breached · Days/amount over · Next action ·
Owner. Sorted worst first. If nothing breaches a rule, say so
explicitly — do not pad the list.
M+ — MEASURE
Good when the list is short enough that someone will act on all of
it today.
+ Grounding: use only the cleaned data above.
+ Constraints: no commentary, no summary paragraph. The table only.
+ Quality check: confirm the count of exceptions and tell me what
percentage of all records that is.
P3.4The standing instructionCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
P — PURPOSE I will run this same report every month. I do not want to rewrite the prompt each time, and I do not want a colleague to get a different answer from the same data. R — ROLE You are a process documenter. You turn one-off instructions into repeatable ones. I — INSTRUCTION + INTERVIEW Take the prompt that produced the output above and turn it into a reusable template. Pull every specific value out into a clearly named slot in CAPITALS. Add a short header stating: what this template is for, which file it runs on, what date range, and who owns the output. S — STRUCTURE A template block I can copy, with the slots listed underneath and one line each explaining what to put in them. M+ — MEASURE Good when a colleague who was not in this room could run it next month and get an output the same shape as mine. + Constraints: nothing specific to this month may remain in the template body. + Quality check: read your own template back and list anything still hard-coded that should be a slot.
Module 4 · AI for Excel & Programme Close 3 cards
Structure before formula. The tracker is built on the data you cleaned before the break, and the instruction you have reused all day finally gets written down and handed to the team.
P4.1The Excel formula cardCLAUDE DESKTOPYOUR OWN PROCESSASSESSED
P — PURPOSE
I need my tracker to tell me {WHAT_YOU_WANT_TO_SEE} without me
reading every row.
R — ROLE
You are a spreadsheet builder who explains formulas to people who
will have to maintain them.
I — INSTRUCTION + INTERVIEW
Write a formula that {DESCRIBE_THE_OUTCOME_IN_PLAIN_ENGLISH — e.g.
"flags any staff member over 45 hours in a week" / "shows days since
last contact"}.
My data is in {SHEET_NAME}. Headers are in row {N}. The columns are:
{COLUMN_LETTER} = {WHAT_IT_HOLDS} (list every column you need)
Give me the formula AND an explanation of what each part does, in
plain English.
S — STRUCTURE
1. The formula, ready to paste
2. Which cell to paste it into
3. A plain-English explanation, one line per part
4. One example: "if row 7 contains X, this returns Y"
M+ — MEASURE
Good when I could explain this formula to the person who inherits
this sheet, without opening Claude.
+ Constraints: {GOOGLE_SHEETS or EXCEL} — say which, they differ.
+ Quality check: tell me what this formula does if the cell is blank,
and what it does if the date is stored as text.
P4.2Adapt a Skill, do not adopt oneCLAUDE SKILLSYOUR OWN PROCESSASSESSED
P — PURPOSE
I have been giving Claude the same instruction all day. I want it
written down as a Skill so it runs the same way every time, and so
it fits how business is actually done here.
R — ROLE
You are an SOP writer. A Skill is a digitised SOP — a procedure
followed the same way twice.
I — INSTRUCTION + INTERVIEW
Turn the instruction below into a Skill definition. Include: a name,
when to use it, when NOT to use it, the file or data it runs on, the
steps, and the output shape.
Then tell me which parts of it assume a non-Malaysian business, and
propose fixes.
[PASTE YOUR BEST PROMPT OF THE DAY HERE]
S — STRUCTURE
Part A: the Skill definition.
Part B: a table — Assumption found · Why it does not fit here ·
My fix.
M+ — MEASURE
Good when a colleague could run it without asking me a question,
and when I can name at least ONE thing I deliberately changed for
Malaysian practice and say why.
+ Context: we are a {INDUSTRY} business in {STATE}, {N} staff.
+ Constraints: plain language. No jargon a new staff member would
not know.
+ Quality check: list anything in the Skill that would break if the
source file's column names changed.
P4.3Skill to PluginCLAUDE PLUGINSYOUR OWN PROCESSASSESSED
P — PURPOSE
Right now I am the only person who can run this. I want my team to
install it and get the same result without asking me how it works.
R — ROLE
You are packaging this for a colleague who was not in the room and
who will not read documentation.
I — INSTRUCTION + INTERVIEW
Take the Skill above and prepare it for packaging as a Plugin.
Give me:
1. a one-line description a colleague would understand
2. what they must have ready before running it
3. the three most likely mistakes and what to do about each
4. what this Plugin must NEVER do without a human approving first
5. who in the team should be allowed to run it
S — STRUCTURE
Five short numbered sections. Section 4 as a bulleted list of hard
rules written in the negative ("must not ...").
M+ — MEASURE
Good when a colleague installs it, runs it once, and does not
message me.
+ Constraints: plain language. Assume no AI experience.
+ Quality check: reread section 4 and tell me whether anything in
sections 1-3 contradicts it.
When you get stuck — keep this one on the table 1 card
Handed out with the workbook at nine o’clock and never collected. Try the first two before you raise a hand — most problems are a missing part of the framework, and diagnosing it yourself is the skill.
RescueThe rescue lifelineCLAUDE DESKTOPTHE WHOLE ROOM
“Explain this card to me.”
Explain this prompt card to me as if I have never used Claude before. What is it actually asking you to do, and what will I get back? Two short paragraphs, plain language.
“My output looks wrong — diagnose it.”
The output you just gave me is not what I need. Do not rewrite it yet. First tell me which part of my instruction was unclear or missing, and what you assumed. Then ask me the ONE question that would fix it.
“I do not know what to put in this placeholder.”
I am filling in a prompt and I am stuck on {SLOT_NAME}. Ask me three
short questions about my business, then suggest what should go in
that slot.“Make this output more useful for me.”
This is technically correct but I cannot use it. My situation is:
{ONE_SENTENCE_ABOUT_YOUR_BUSINESS}. Rework it so I could act on it
today, and tell me what you changed and why.What To Do Next
- Name one process, today.
Not “improve efficiency”. A named process with a start and an end — the one you would fix first if somebody handed you a spare afternoon. - Run the framework card on a job you did last week.
Fill all five parts, add the interview line, and see what Claude asks you that you had not thought of. That question is usually the gap between a vague prompt and a usable one. - Clean one export before you analyse anything.
One standard name per group, dates that are dates, and a single count of unique records you would defend out loud. Everything downstream depends on it. - Write your four supervision lines before you automate anything.
What runs unwatched, what comes back to you, what is never unsupervised, and who may run it. If you cannot name where a human stops it, it is not finished. - Bring your process to a cohort, or ask about an in-house one.
Public intakes run against the registered outline. In-house cohorts can run a vertical track on your own paperwork. Message us and say which process you would bring.
Your trainer
Kalmen Chia — Co-Founder, E-Global SCM & DigitalAI Business Club. HRD Corp accredited trainer (Train the Trainer). Delivering ERP and AI automation for Malaysian SMEs since 2008, across Johor, Sabah, KL and Penang.
WhatsApp: +6016 773 7850
One Process. One Worker. Running by Friday.
Claude AI for Business Level 2 — registered as Claude AI for Business: Sales Enablement, Documents, Data & Excel. One day, four modules, six things you can put in front of your management on Monday.
Next intake: REPLACE-INTAKE-DATE · Registered and approved with HRD Corp under SBL-Khas. Claiming is done by the employer through their own HRD Corp application and remains subject to HRD Corp approval and the prevailing circulars.
WhatsApp Kalmen: +6016 773 7850 See the 3-Level PathwayFrequently Asked Questions
Click any question to open it.
What is Claude AI for Business Level 2?
Claude AI for Business Level 2 is the intermediate day of the Claude AI for Business pathway, registered with HRD Corp as Claude AI for Business: Sales Enablement, Documents, Data & Excel. It is a one-day, four-module programme in which participants research a prospect or a market with Claude in Chrome, produce a proposal and a six-slide pitch outline, adapt one document for three readerships, turn a dirty data export into a board-level narrative and three reusable reporting templates, build a working tracker with three explained Excel formulas, and package a proven instruction as a Skill and then as a team-installable Plugin.
Do I need to know how to code to attend Level 2?
No. Nothing on the day is written in code. Participants work in Claude Desktop, Claude in Chrome, a browser and a spreadsheet, and every instruction is given in plain English. The two technical-sounding words on the agenda, Skill and Plugin, are handled without any programming: a Skill is a written procedure, and packaging it as a Plugin is a guided walkthrough. Coding is Level 3.
What do I need before the training day?
A paid Claude plan, Pro or higher. This is a prerequisite and not a recommendation, because Claude in Chrome and Claude Cowork are both paid-only and a free-tier participant cannot complete Module 1 or Module 3. You also need the Claude in Chrome extension installed and signed in, a laptop that can run Claude Cowork, your chosen track confirmed, and a real or sample brief, document and dataset to work on. A setup and plan check is issued 48 hours before the programme. On Windows, Cowork needs the Virtual Machine Platform feature enabled, which requires administrator rights, and not every Windows edition and device is supported. If your laptop is managed by your company IT department, say so at booking so the instructions can be sent to them in advance.
What do I leave the day holding?
Six things, all built on your own process rather than a demonstration file: a client proposal packet and research brief, a six-slide pitch deck outline with speaker notes, an executive management data narrative, a custom Excel formula toolkit of three formulas plus the completed prompt card, an installable team Plugin, and a written Week 1 commitment. Alongside those you keep a three-touch follow-up sequence, a cleaned data set, a second-audience version of your document, three reusable reporting templates, a spreadsheet structure plan, and your own written supervision rules.
How is Level 2 assessed?
Against the registered assessment methods, module by module, using the work you produce in the room. Module 1 reviews your research brief and proposal or pitch outline against a sales brief template. Module 2 reviews your document draft and presentation outline against an audience-adaptation checklist. Module 3 reviews your data summary, management narrative and three reporting prompt templates. Module 4 reviews three Excel formulas, the spreadsheet structure plan, the packaged Plugin and the completed Excel prompt card. There is also facilitator observation during the exercises, a summative review of the reviewed package at the end of the day against a 70 percent pass mark, the HRD Corp evaluation form at close, and a follow-up check-in 30 to 90 days later.
Is real business data used on the day?
The teaching examples are not. The sample businesses, the dirty exports, the bad letter and the marketplace Skill used for demonstration are all written to be realistic rather than real, and every planted fault in them is deliberate. Your own work is a different matter: you bring a real document, a real prospect or role, and a real export, because the point of the day is that the artefacts are usable on Monday. Where a card asks Claude to read customer or staff records, the rule stated in the room is consent first, and no customer detail enters a shared file before that check. If you are copying a card from this page rather than working in the room, use fictional or sandbox data.
How is Level 2 different from Level 1?
Level 1 teaches you to work with Claude, one conversation at a time, and a free plan is viable. Level 2 is where you stop doing the work and start giving it to something: a Skill is the job description, a Plugin hires that worker into the whole team, a reporting template is the standing instruction, and a brand-voice and audience checklist is the house style. Level 2 also requires a paid plan, because Claude in Chrome and Cowork are paid-only. Level 1 introduces Skills at a personal level; Level 2 extends this to Plugins in a non-technical way; Level 3 goes fully technical inside Claude Code.
Is this programme HRD Corp claimable?
The programme is registered and approved with HRD Corp under the SBL-Khas scheme. Attendance and the HRD Corp evaluation form are collected on the day as claim evidence. Claiming itself is done by the employer through their own HRD Corp application and remains subject to HRD Corp approval and the prevailing circulars.