
Save Money
with Claude
Most businesses don't have a revenue problem — they have a leak problem. Four worked examples that find the money you're already losing: subscriptions, supplier prices, outsourced spend and manual hours.
Led by practitioners,
not theorists
Chan Wei Khjan
Audit Partner, YYC · MIA Board
Chartered Accountant (ACCA · C.A.(M) · FCA Singapore) and MIA board member. Featured in Business Insider for pioneering AI inside the accounting profession.
Marcus Chia
AI Product Manager & Harness Engineering Expert
Builds and ships AI workflows in the wild — Claude projects, MCP servers, multi-agent pipelines and data dashboards. Turns messy business problems into working AI solutions.
Warren Leow
Founder, AITraining2U PLT
Drives AITraining2U's mission to equip 100,000 Malaysians with practical AI skills — hands-on with AI agents, automation and applied analytics for enterprise teams.
Today you'll build alongside us — every example paired with something you can run on a real invoice, quote or statement the moment you're back at your desk.
Trusted by leading Malaysian organisations























Banking, telco, retail, healthcare, energy & government-linked companies — teams across Malaysia upskill with AITraining2U.
Margin leaks in four places
None of them look like a crisis on their own. Together they are the cheapest money you will make this year — because you have already earned it.
Most owners aren't short
on ways to make money.
They're short on visibility over what leaves — the renewals nobody approved, the supplier who quietly raised 18%, the invoice you pay every month for work your team could now do in an afternoon.
A real finance function has a cost controller, a procurement lead and an ops analyst asking those questions every month. Run Claude as all three, and the review that never gets done takes a prompt.
Three moves that pay for
everything else
None of these is a prompt. They're the conditions that decide whether the next four examples actually land money in your account — or just make a nice slide.
The cheapest hire you'll make this year
| Line | Per month | Per year |
|---|---|---|
| 5 seats on the primary engine | RM 475 | RM 5,700 |
| 2 seats on the other two, for cross-checks | RM 190 | RM 2,280 |
| Total tool spend | RM 665 | RM 7,980 |
| Cash found by the four examples | — | RM 84,040 |
| Payback | About 5 weeks — then roughly 10× for the rest of the year | |
Seat prices are indicative at roughly RM95 per seat per month for an entry paid tier — check current rates, and ask about annual and team pricing before you buy.
Hire for judgment. Let the tool supply the experience.
| The work | Used to need | Now | Senior still does |
|---|---|---|---|
| Monthly ops report | Outsourced analyst, 8 days | Junior finance exec + AI, 2 hrs | Reads it, signs it |
| Social content, 24 posts | Agency retainer | Marketing exec + house brief | Approves the month |
| Supplier & market research | 2 days of a manager | Junior + AI, 2 hrs | Decides what to do with it |
| Board pack | Consultant formatting | Junior drafts from our numbers | Writes the commentary |
This is about the shape of your next hire and where senior time goes — not a case for cutting the people who already know why the numbers matter.
“Save where we can” saves nothing
| Line | RM |
|---|---|
| Total outgoings | 3,492,549 |
| less payroll & bonus | (1,484,387) |
| less outlet rent | (1,152,000) |
| Controllable base | 856,162 |
| The goal — 10% | 85,616 |
| Found across the four examples | 84,040 |
| Still to find | 1,576 |
Plus 26 hours a week returned — that's capacity, not cash. Count it on a separate line or you'll report a saving you can't find in the bank.
Four ways to save money
with Claude
Every example ships with a set of copy-ready prompts — tap Copy, paste into Claude, edit the bracketed bits, run.
You can't cut what you can't see
Spend audit prompt set
Act as our cost controller. I'm attaching [12 months of bank / card transactions, or our P&L export]. 1. Categorise every line: software & subscriptions, suppliers, outsourced services, marketing, payroll, other. 2. Rank categories by annual spend and show each as a % of total. 3. Flag anything questionable — duplicate charges, two tools doing one job, anything that rose more than 10% vs last year, and any charge we can't identify. Output: a summary table by category, plus a "questionable spend" table with amount per year and why it's flagged. Ask me before assuming what an unclear line item is.
Now take the software & subscriptions list and review our whole stack. For each tool: what it does, cost per month and per year, seats paid vs seats actually used [I'll paste usage], which other tool in our stack overlaps with it, and the cheaper route — annual plan, lower tier, free tier or a competitor. Then give me three lists: - CANCEL NOW - DOWNGRADE - CONSOLIDATE INTO [tool] Each line: amount saved per year, and what we actually lose by doing it. Sort by savings, total it at the bottom.
Automate this. Using Cowork, Schedule it for the 1st of every month: 1. Read our latest statement export and compare it to last month. 2. Flag NEW charges, any amount that went up, and every renewal falling due in the next 45 days. 3. Re-check the cancel list — did those charges actually stop? Email me a one-page spend digest via Gmail, with a "cancel before this date" section at the top. Draft it — don't send.
Tip: use your accounting export or a redacted statement — strip account numbers and staff names first. Date, description and amount is all Claude needs.
What comes back: the leaks, in ringgit
Our running example — “Kopi Kita”, a fictional 14-outlet KL specialty-coffee chain — runs the spend audit. Claude reads 12 months of transactions and hands back the leak list:
| Line item | RM / year | Flag | Action | Saved |
|---|---|---|---|---|
| Design tool × 2 licences | 8,400 | Duplicate | Consolidate to 1 | 4,200 |
| Analytics SaaS — 9 seats | 12,960 | 6 seats unused | Downgrade to 3 | 8,640 |
| Delivery-app fee tier | 45,780 | Rose 18% | Renegotiate | 7,000 |
| Legacy POS add-on | 3,600 | Nobody uses it | Cancel | 3,600 |
The 18% fee rise renewed automatically — nobody approved it, nobody noticed it. One phone call is worth more than the other three lines combined.
- • Cancel now RM3,600 — this week, zero risk
- • Downgrade seats RM8,640 — next renewal
- • Consolidate tools RM4,200 — one migration
- • Renegotiate fee tier RM7,000 — target
- Total ≈ RM23,440 / year
Illustrative figures for teaching — not real client data. A live run works off your own export and shows its working, line by line.
Nobody pays list price — except the people who never ask
Supplier & quote prompt set
Here are [2-4] quotes for the same job: [paste or attach each one]. Normalise them to a like-for-like comparison: 1. Restate each scope in the same words. Flag anything one includes that the others don't. 2. Build a table: supplier | headline price | setup / one-off | minimum commitment | payment terms | contract length | exit cost. 3. Work out the 1-year and 3-year total cost of ownership for each, at our real volume of [X units / month]. Tell me which is genuinely cheapest, and which one looks cheap but isn't. End with every question I should ask before I sign.
Using the live web, benchmark what [service / product / category] actually costs in [Malaysia / Klang Valley] for a business our size — [X] outlets, [Y] staff, [volume per month]. Give me: - the fair market range (low / typical / high), with sources - where our current price of [RM amount] sits in that range - the 3 levers that move price most here (volume, contract length, payment terms, bundling, off-peak...) - 3 concrete asks I can make, and the target price I should walk in with Be blunt about whether we're overpaying.
Draft the renewal email. Firm, polite, no threats: what we pay now, what the market says, our 3 asks, our target price, and a clear deadline. Give me two versions — one for a supplier we want to keep, one for a supplier we'd happily replace. Then, using Cowork + Schedule, build our renewal calendar. For every contract I list with its end date, remind me 60 days before via Gmail with the supplier, what we pay, and the last benchmark. Draft it — don't send.
Tip: run prompt 2 before the renewal call, not after the invoice. Sixty days of notice is the whole negotiation.
Three quotes, one honest comparison
Kopi Kita is retendering its cup & packaging supply — 120,000 units a year. Three quotes go in; Claude normalises them and benchmarks the market:
| Supplier | Headline | Setup + fees | 3-yr TCO | Verdict |
|---|---|---|---|---|
| Supplier A | RM0.42 / unit | None | RM151,200 | Cheapest overall |
| Supplier B | RM0.38 / unit | RM9,600 setup + RM4,000/yr platform | RM158,400 | Looks cheapest, isn't |
| Supplier C | RM0.45 / unit | RM3,000 setup | RM165,000 | Out |
B's unit price is 10% lower — and it still costs RM7,200 more over three years once setup and the annual platform fee land. Headline price is not price.
- • Market range, KL, 100-150k units: RM0.36-0.48
- • We sit at RM0.42 — mid-range, not a bargain
- • Ask: 24-month term + quarterly payment
- • Target: RM0.375/unit
- ≈ RM5,400 / year saved
Illustrative figures for teaching — not real supplier data. A live run cites every source it used for the benchmark.
Keep the specialists. Stop paying agency rates for first drafts.
What moves in-house is volume, not expertise — first drafts, resizes, monthly reports, product copy. Specialists keep the work where their judgment is the product.
Insourcing prompt set
Here's everything we currently outsource: [for each line — supplier, what they deliver, how often, what we pay]. Classify every line into: - KEEP WITH SPECIALIST — real expertise, brand or legal risk, or genuinely good value for what we get - CO-PRODUCE — we draft, they review and sign off (pay for judgment, not typing) - BRING IN-HOUSE — volume, templated, low risk For each: cost per year now, likely cost after, hours it would take us, and the risk of getting it wrong (low/med/high). Be honest — tell me which lines I should NOT touch, and exactly why.
Build our house-style brief so every in-house draft comes out consistent. From [paste 3-5 of our best pieces, our brand guide, or our website], extract: - who we're talking to and what they care about - our tone in 5 rules, each with a do and a don't - words, claims and formats we never use - the structure of each deliverable: [social post / product description / monthly report / SOP] Output it as a reusable brief I paste at the top of every future prompt. Then use it to produce [this month's deliverable] end to end.
We pay [RM X / month] for [tool] and only use it for [these 2-3 things]. Using Cowork, spec and build the simple internal version: 1. Write the one-page spec — inputs, what it outputs, who uses it, what it must NOT do. 2. Build it as a single self-contained page or sheet I can run today. 3. Then tell me honestly where the paid tool is still worth it — support, compliance, integrations, scale. If it is, say so and stop.
Tip: run prompt 1 with the invoices in front of you. The line you feel defensive about is usually the one worth keeping.
The honest split — what moves, what stays
Kopi Kita lists everything it outsources. Claude splits it three ways and prices each route:
| Line item | RM/yr now | Call | RM/yr after |
|---|---|---|---|
| Brand identity & packaging design | 18,000 | Keep with specialist | 18,000 |
| Social content — 24 posts/month | 28,800 | Bring in-house | 3,600 |
| Product & menu copy | 9,600 | Bring in-house | 0 |
| Monthly ops report build | 14,400 | Co-produce | 4,800 |
| Loyalty dashboard SaaS | 10,800 | Build instead | 0 |
| Total | 81,600 | — | 26,400 |
Brand identity doesn't move. It's the one line where a specialist's judgment is the product — and at RM18k it's already good value. Cutting it would cost more than it saves.
- • 24 posts/month drafted from one house brief
- • Agency retained for brand & final sign-off
- • Report co-produced — they check, we build
- • Dashboard rebuilt internally in a day
- Net ≈ RM55,200 / year
Illustrative figures for teaching — not real client data. Your split will look different; the point is to make it deliberate rather than habitual.
Every manual hour is payroll. Price it, then automate it.
This isn't about a smaller team — it's about your team spending Monday with customers and suppliers instead of copy-paste.
Back-office automation prompt set
Act as our operations analyst. Here are the recurring tasks my team does: [task, who does it, how often, roughly how long it takes]. Our loaded cost per hour is [RM X] — salary + EPF/SOCSO + overhead, divided by hours worked. Build one table: task | hours per week | cost per year | how rule-based it is (1-5) | how much judgment it needs (1-5) | what breaks if it's done wrong. Rank by cost per year. Then name the top 3 to automate first — highest cost, most rule-based, lowest risk — and tell me what the team should do with the hours we get back.
Take the #1 task from that list. Write the full workflow as if you're handing it to a new hire on day one: 1. Trigger — what starts it, and when 2. Inputs — where the data comes from, in what format 3. Steps — numbered, unambiguous, with the exact rules and thresholds 4. Exceptions — every case a human must decide, and who decides it 5. Checkpoints — where a person signs off before anything leaves the company 6. What "done" looks like Then give me the prompt (or agent instructions) that actually performs steps 1-4.
Now run it. Using Cowork, Schedule it [daily at 8am / every Friday 5pm]: 1. Run the workflow above on this period's inputs. 2. Handle the clean cases end to end. 3. Put anything unusual in an EXCEPTIONS list with the reason — never guess. 4. Track: items handled, hours saved, exceptions raised. Email me the run summary via Gmail — exceptions at the top, the month's savings tally at the bottom. Draft it — don't send.
Tip: keep a human checkpoint on anything that touches money leaving the company or a customer's inbox. Automate the preparation, not the approval.
Hours, priced — then the top two gone
Kopi Kita maps its back office at a loaded rate of RM35/hour. The ranking makes the decision obvious:
| Task | Hrs/week | RM / year | Rule-based | Call |
|---|---|---|---|---|
| Supplier invoice matching | 14 | 25,480 | 5/5 | Automate first |
| Outlet daily sales roll-up | 12 | 21,840 | 5/5 | Automate first |
| Quote & order confirmations | 12 | 21,840 | 4/5 | Automate next |
| Staff roster changes | 9 | 16,380 | 2/5 | Keep human |
26 hours a week came back. Kopi Kita didn't cut a role — those hours went to outlet visits and supplier calls, the two things that were always “when we have time”.
- • Automated: invoice matching + sales roll-up
- • 26 hrs/week returned ≈ RM47,320/year
- • Exceptions raised: ~4/week, all human-checked
- • Roster stays human — judgment, not rules
- Payment runs still need your approval
Illustrative figures for teaching — not real client data. Use your own loaded hourly rate; the ranking matters more than the precision.
The cost-saving operating system
Across the worked example, four prompt sets found ≈RM131,000 a year — part cash saved, part hours reclaimed. Do it once, then let it run monthly.
Claude isn't another subscription.
It's the one that pays for the rest.
The owners who win 2026 aren't spending less on growth — they've stopped leaking margin on tools nobody uses, prices nobody questioned, work they no longer need to buy, and hours nobody counted.
Our courses at AITraining2U.com
All courses are HRDC-claimable — available as enterprise & public workshops. AITraining2U.com

Thank you.
Pick one example from today, run it this week on a real statement, quote or invoice — and tell us what you found.
Practical. Proven. Purposeful. — AITraining2U PLT