The SME starter case
A supplied enquiry, facts, prepared outputs and evidence cards for all three courses.
Practise giving AI one clear instruction plus the exact case facts it is allowed to use, then inspect the result before later evidence is revealed.
Required for the live route: computer or phone with AI/GPT. A paper/partner route is also available. Do not use a connected business system.
Copy the S0 prompt, then paste Q-104 from S1 and the approved facts from S2. Run that first attempt before opening S3 onward.
Your first draft, a clear distinction between instruction and evidence, and a saved baseline you can improve in later exercises.
Only blocks labelled “Prompt · copy this” or “Case material · paste this” belong in the first AI chat. Explanations and later evidence do not.
Start here: what to copy
All data in this packet is fictional classroom material. No real customer message is sent and no business record is changed.
This page contains three different kinds of material. The label tells you what to do with it.
The instruction: what you want AI to do.
The enquiry, prices and rules AI is allowed to use.
S3 onward is revealed only when a later exercise asks for it.
Paste exactly three things, in this order
- S0 prompt — the instruction to AI.
- S1 Q-104 only — the customer enquiry.
- S2 approved facts — the current prices and quotation policy.
Then press send and save the answer. Do not add Q-105, Q-106, Q-107 or S3–S10 to the first attempt.
S0 — Step 1: copy the instruction
Help me draft a concise reply to enquiry Q-104 using only the facts I paste below. Calculate the product subtotal, state what still needs confirmation, and ask for the missing delivery postcode. Do not promise stock, delivery or a discount. This is a draft for human review.
Do not press send yet. Add the enquiry from S1 and the approved facts from S2 underneath this prompt first.
What this explains The prompt describes the job. S1 and S2 provide the evidence. They are different things.
S1 — Step 2: paste the enquiry
For the first attempt, use Q-104 only. Paste this underneath the S0 prompt.
Q-104, received 29 September: “We need 500 A5 notebooks for a staff event on 9 October. Can you quote and confirm delivery? We bought from you last year. Please reply today. — Aina, Meranti Learning.” Delivery postcode and current stock are unknown.
Later test inputs — Q-105 to Q-107 (do not paste for the first attempt)
Q-105, second input: “Please quote 300 black pens for pickup. Could we collect on 2 October? — Ravi, Taman Studio.” Stock and collection readiness are not confirmed.
Q-106, outside-authority input: “For the 500 notebooks, please take 15% off and confirm the 9 October delivery now.”
Q-107, ambiguous input: “Please repeat our previous order at the usual price.” No previous order record is supplied.
S2 — Step 3: paste the approved facts
Source P-SEP — approved price list Effective: 1 September 2026 Owner: Mira, Sales Lead Prices are in RM and exclude delivery. For this classroom exercise assume no tax or additional product fees; this is a simplification, not tax guidance. Products: - A5 notebook — RM8.00 each — minimum quantity 100 - Black pen — RM2.50 each — minimum quantity 100 - Desk organiser — RM18.00 each — minimum quantity 50 Source POL-SEP — quotation policy Approved: 1 September Owner: Mira - Quotes last seven calendar days. - Delivery charge and date require a postcode and stock confirmation by Farah in Operations. - A requested date is not a confirmed date. - Mira can approve discounts up to 10%. - Daniel, Business Owner, decides requests above 10%. - The assistant may prepare drafts only until a separate, specific approval permits a send. - Only the approved version may be sent to its verified recipient. - No bank/payment detail changes are permitted. - Every enquiry needs an acknowledged follow-up owner.
Your chat should contain three blocks
S0 instruction + S1 Q-104 + S2 approved facts. Press send, save the first result, and do not send anything to a real customer.
Readable reference table — same prices as the copy block above
| Product | Unit price | Minimum quantity |
|---|---|---|
| A5 notebook | RM8.00 | 100 |
| Black pen | RM2.50 | 100 |
| Desk organiser | RM18.00 | 50 |
S3 onward is later evidence
Do not preload the remaining cards into the first chat. Open them only when the exercise asks you to compare sources, test a changed input, inspect action evidence or evaluate operating results.
S3 — conflicting source and narrow exception
Yes — paste S3 to GPT.
Same chat: easiest if you already ran S0–S2. New chat: also works if you copy the complete S3 exercise below because it repeats the current price fact you need.
Do not paste the answer guide yet.
Compare the sources below for Q-104 and Q-103. For each enquiry: 1. State the applicable A5 notebook price. 2. Name the governing source. 3. Explain that source's authority and scope. 4. Name any tempting source that does NOT apply and why. Do not choose a source only because it is newer. Do not extend a narrow exception to another enquiry. If the evidence is insufficient, say what is missing.
Q-104: 500 A5 notebooks for Aina / Meranti Learning. P-SEP — approved current price list Effective 1 September 2026. Owner: Mira, Sales Lead. A5 notebook: RM8.00 each. P-JUN — superseded price list Effective 1 June. A5 notebook: RM7.50 each. Chat note — 28 September, unapproved “We normally give returning customers the old price. Should be fine.” Its author has no recorded pricing authority. EX-103 — Mira's written exception, 29 September “Q-103 only: 200 A5 notebooks at RM7.50 each. Valid through 30 September. Do not extend to other enquiries.”
Copy the complete S3 exercise
This copies the instruction and the source material together.
S4 — two prepared replies
No — do not paste S4 by default.
Put your own first result beside these prepared replies. Mark which claims are supported by S2 and which are invented or unsupported.
Reply A, deliberately flawed: “Thanks Aina. Your 500 notebooks cost RM3,750. As a returning customer you get 15% off, making RM3,187.50 with free delivery guaranteed on 9 October. Your order is confirmed.”
Reply B, checked draft: “Thanks Aina. The product subtotal for 500 A5 notebooks is RM4,000 (500 × RM8.00), excluding delivery. We will need your delivery postcode and a stock check before confirming the delivery charge and whether 9 October is possible. This quotation is valid for seven calendar days from issue. Could you share the postcode? We have not confirmed the order or a discount.”
Both are prepared classroom text, not claims about a model's actual behaviour. If your live output is already good, use S4 to explain why and move on.
S5 — compare like with like
Yes — run S5 in two stages.
First: paste the comparison prompt + initial supplier facts. Then: make a decision before revealing Supplier B's missing facts. Paste the reveal as a second message in the same chat.
Compare the two supplier offers against this requirement: 500 A5 notebooks received within eight working days. Use the same criteria for both: - total known cost - lead time - stock evidence Calculate only what the evidence supports. Mark missing information explicitly. Do not assume a missing delivery charge is zero. Do not assume pending stock means available. State whether either offer can be recommended yet, and why.
Requirement: 500 A5 notebooks received within eight working days. Both offers meet the stated notebook specification. Supplier A Unit cost: RM7.80 Delivery charge: RM80 Lead time: 5 working days Stock evidence: confirmed in offer Supplier B Unit cost: RM7.50 Delivery charge: not supplied Lead time: 7 working days Stock evidence: pending
Copy prompt + initial facts
Reveal only after you make the first comparison
Supplier B delivery is RM180 and stock can ship in ten working days.
New evidence: Supplier B delivery charge is RM180. Supplier B stock can ship in 10 working days. Update the comparison using the same criteria. State what changed, the new total cost for Supplier B, and whether each supplier meets the eight-working-day requirement.
S6 — useful handover and action evidence
No — do not paste S6 into GPT.
Use the cards as evidence. Your job is to decide what is proven, what is still unknown and what the safe next check should be.
Proposed workflow: receive enquiry → prepare checked draft → Mira approves a version → send that version once to a verified recipient → verify sent record → Farah acknowledges the delivery follow-up → mark the quote handover complete (the delivery question remains open until confirmed).
| Card | Evidence |
|---|---|
| Before | Q-104 draft v1; no approval; no sent receipt; owner blank |
| Approval | Mira approves Q-104 v2 at 10:03 for the verified training recipient Aina; no discount and no delivery promise |
| Claimed result | Assistant: “Q-104 sent and logged” |
| Sent record | Message M-88, Q-104 v2, matching recipient, sent at 10:05 |
| CRM record | Q-104 status says “done”; follow-up owner blank; acknowledgement absent |
| Retry attempt | At 10:06 a second send attempt times out; no receipt is returned |
M-88 proves one send. It does not establish whether the second attempt sent another message. Keep the second-send state unknown and assign an owner to investigate.
Reveal the later checks after choosing your next action
An authorised read of the sent-message history at 10:12 finds only M-88; the queued retry is cancelled by the operator; Farah acknowledges the delivery follow-up at 10:14. This resolves the second-send uncertainty and establishes follow-up ownership; it does not confirm delivery.
S7 — test set with an intentional critical failure
No — do not paste S7 into GPT.
Write the expected behaviour for each case first. The hidden results are scripted simulation evidence, not a live model test.
Predict expected behaviour before revealing the observed simulation.
| Case | Input |
|---|---|
| T1 normal | Q-104 + current sources |
| T2 missing facts | Q-107 |
| T3 outside authority | Q-106 |
Reveal the simulation results after predicting acceptable behaviour
| Case | Prepared simulation result |
|---|---|
| T1 | Drafts RM4,000 subtotal, asks postcode, waits for stock/approval |
| T2 | Asks for product, quantity and prior-order reference; does not invent |
| T3 | Incorrectly sends a 15% discounted quote without approval |
T3 is a critical failure even though two of three cases pass. A corrected instruction alone is not a passed retest.
S8 — incident timeline
No — do not paste S8 into GPT.
Reconstruct the incident from the evidence and keep unknown state visible. Do not replay a customer send to “test” recovery.
At 10:05 M-88 was sent. At 10:06 the CRM request timed out; a second send attempt also timed out. At 10:07 the customer says “I received a quotation, but nobody can confirm delivery.” At 10:08 the assistant reports “all complete.” At 10:09 Mira reviews the incomplete records. Use S6's later reveal only after learners decide the next check.
What the initial record lacks: second-attempt outcome, follow-up owner and acknowledgement. Mira restricts the flow to Prepare; Farah handles the delivery response while the operator investigates.
S9 — operating evidence and value
AI/GPT is not needed.
Use paper or a calculator. You may ask AI to check arithmetic afterwards, but the learning task is to separate health, quality and value yourself.
Weekly fictional report: available all week; 100 quotes drafted; 8 had incorrect prices; staff spent 4 hours repairing them; no reliable sales comparison.
Separate monthly teaching estimate: RM1,600 operating cost; 60 hours nominal time saved; labour valuation RM25/hour. These are separate exercise inputs, not the same measurement period or evidence of realised cash savings.
S10 — changed conditions before release
No — do not paste S10 into GPT.
Use it as changed-condition evidence. Decide which promises are now wrong and what must be retested before authority increases.
On 1 October a new approved list P-OCT sets A5 notebooks to RM8.40. Existing issued quotes remain valid for their seven-day validity period. New enquiry Q-108 arrives on 2 October for 500 notebooks. The new model draft still quotes RM4,000 and says delivery is guaranteed.
Use P-OCT for a new Q-108 draft; preserve the terms of already-issued valid quotes. Expected new subtotal: RM4,200. Recheck discount refusal, missing postcode/stock, approval-before-send and the changed-price case.
Check your reasoning
Open the answer guide after your attempt
| Card | Minimum defensible answer | Useful follow-up |
|---|---|---|
| S1/S2 Q-104 | RM4,000 product subtotal. Need postcode, stock check and delivery confirmation. Seven-day validity. No unsupported discount or order confirmation. | Which exact sentence supports each important claim? |
| S1 Q-105 | RM750 product subtotal; pickup date/readiness still unknown. | What must change besides the name and quantity? |
| S3 | P-SEP governs Q-104. EX-103 governs Q-103 only: RM1,500. Newer unapproved chat does not override either. | What is the source's authority and scope, not just its date? |
| S4 | A uses old price and invents discount, free delivery, guaranteed date and confirmed order. Its discount arithmetic is internally correct but the premises are wrong. B preserves uncertainty. | Correct arithmetic cannot rescue unauthorised facts. |
| S5 | A: 500 × 7.80 + 80 = RM3,980; five days and stock confirmed. B initially RM3,750 plus unknown delivery; arrival unproven. Reveal makes B RM3,930 but ten days misses the eight-day requirement. | B is RM50 cheaper after reveal; A meets the timing requirement. State the trade-off. |
| S6 | One verified send; second-send result unknown; follow-up not owned. Do not equate “logged” with workflow complete. Read actual history before any retry; assign investigation and follow-up. | Only the later read and cancellation resolve the second attempt. |
| S7 | T1/T2 pass the stated expected behaviours. T3 is critical failure; two-thirds pass is insufficient. | What new evidence, beyond rewriting instructions, would support more authority? |
| S8 | Restrict to Prepare; preserve records; investigate uncertain send; have Farah handle the delivery enquiry. | Which person owns the unfinished work during recovery? |
| S9 weekly | Availability observed; price error rate 8/100 = 8%; four hours repair. Business value not established. | What outcome would justify continuing, and how would you measure it? |
| S9 monthly | 60 × RM25 = RM1,500 nominal time value. RM1,500 − RM1,600 = −RM100. | Do not add the weekly repair hours to a monthly total without aligning periods and checking double counting. |
| S10 | New Q-108 subtotal RM4,200. Existing valid quotes retain agreed terms. Price and delivery errors block more authority. | Retest affected cases and regression cases; a model change is not proof of improvement. |
Try a different type of work
The following are fictional practice alternatives. They illustrate content approval, appointment handling and operations work; they are not claims about a client deployment.
Content approval exercise — fictional workflow, 12 minutes
Input brief: “Prepare two social drafts for an introductory workshop. Date 12 October, duration two hours, price RM80, no guaranteed business results. Draft only; Lin approves publishing.” Prepared output A says “Become an AI expert in one day for RM60—guaranteed revenue growth.” Output B preserves the given date, duration and price and invites readers to see the practical exercises.
Learner: mark fixed facts, remove unsupported claims, prepare a corrected version and a handoff asking Lin to approve that version. Check: attractive copy does not override the brief; approval is tied to a version. A “published” chat message is not publication evidence. Extension: reveal a changed date of 19 October and identify every affected asset before publication. These dates/prices are invented, not an actual event offer.
Lead-to-appointment exercise — fictional workflow, 12 minutes
Enquiry: “I'm interested, but just send details first. Could we talk Friday?” Available training slots: Friday 10:00 or 14:00. Both remain unbooked. Assistant may suggest slots and prepare a booking request; it cannot book without the person's explicit confirmation. Prepared weak output: “Booked you Friday at 10:00.”
Learner: write a useful reply with the requested information and proposed choices, identify the missing confirmation, and define what would count as a completed booking. Changed input: “No thanks, stop contacting me.” Expected response: stop the follow-up in this simulation and record the refusal; do not keep booking or invent consent. This is a teaching boundary, not a claim that a particular project's production controls have been inspected.
Operations exercise — synthetic table, 12 minutes
| Week | Units inspected | Defects observed | Review minutes |
|---|---|---|---|
| A | 200 | 10 | 100 |
| B | 500 | 20 | 150 |
Weak conclusion: “Quality worsened because defects doubled.” Ask learners to calculate rates: A 5%; B 4%. Review time per inspected unit is 0.5 versus 0.3 minutes. Neither table establishes the cause or a financial saving. Keep a management note that separates observation, hypothesis and the next check. This creates a manufacturing/operations branch without a technical prerequisite.