Proof of work, not brochure claims. Everything on this page is real and checkable: systems that are live right now, programmes you can look up, ten workflow rebuilds participants took back into their own jobs, and outcomes reported in verified reviews. Where a figure is a prototype target rather than an audited result, it says so.
Marketing automation
A multi-site AI marketing system Ivan built and operates — it pulls SEO and funnel signals daily, diagnoses gaps, and drafts the blog and social content for maplecommerce.com and client sites. Governed autonomy, not a black box: every publish passes a human approval gate, and every autonomous action is recorded in an append-only audit ledger with a reversal recipe.
The content marketing on this site is its output.
Free assessment tool
A self-assessment that maps where you are on the AI capability ladder and generates a personalised 30/60/90-day learning roadmap with curated, link-checked resources for each step.
Live and free to use. learn.bestcourses.sg/ai-capability-planner
Agent engineering
WhatsApp and Telegram agent bots, multi-platform agent frameworks, desktop apps and CLI tools — built in TypeScript on the Anthropic Claude Agent SDK and run in production, not demos.
The same patterns taught in the agentic AI courses below.
Learning & development
The full L&D lifecycle, deployed — not just slides: evidence-based course research using the FRAME methodology (Find evidence → Roles & Rules → Assemble → Measure & Modify → Expand), curriculum development mapped to funded programmes, live cohort delivery, and rubric-driven AI-assisted capstone assessment with integrity checks and a full grade log per cohort.
Running across the live programmes listed below — every stage human-reviewed.
Named institutions, programmes and credentials — verifiable, not anonymised.
Ten workflow rebuilds across ten sectors, designed and prototyped by working professionals in the AI training and coaching programmes. Each starts from a workflow the person actually owns, measures what it costs today, and rebuilds it with a human approval step kept in the loop. Full write-ups at bestcourses.sg/case-studies.
Property services
In production7 working days 1 monthly management reporting
A 24-room in-house dormitory for ~200 workers, run across WhatsApp threads, shared sheets and manual email — about 918 staff-hours a year going into hand-built reports. Requests now arrive through a form, AI checks them against allocation rules and live bed capacity, and the operations coordinator makes the final call.
Request processing 5 calendar days → under 3.
Multi-service SME
In production10 minutes under 4 to draft and send a quote
A three-person operator running minibus charter, hair and beauty, reflexology and career coaching with no booking page at all — every enquiry tracked from memory across scattered chat threads. Now a single direct-booking page plus an AI admin assistant that drafts quotes and tracks open tasks by business unit.
A direct booking channel that did not previously exist, paying no aggregator commission.
HR advisory
30 minutes under 5 per straightforward enquiry
An HR team fielding 30–50 enquiries a day against a policy library scattered across intranet, file registries and SharePoint. An AI project pre-loaded with the cleared current policies drafts a grounded reply; the HR business partner strips identifiers, reviews and sends every one.
Roughly 15–17 hours of HR capacity returned per week — about S$2,080 a month of redirected time.
Consumer goods finance
25 hours 10 per quarter
A finance analyst building a quarterly deck on three listed rivals reporting on staggered dates, so the deck is reworked every time a release lands. The extraction schema and deck structure now live as reusable project files — the analyst still reads the filings and verifies every figure.
About 60 hours a year. Final release to leadership deck: 5–6 business days → 3.
Recruitment
45 minutes 15 per application
Consultants manually reviewing candidate documents and assembling each MOM submission, where missing or inconsistent documentation delays filing. AI extracts the key fields, flags what is missing and drafts the summary; a consultant approves before anything reaches MOM.
First-time submission success targeted from 70% to 95%.
B2B services
over 2 hours under 30 min per client enquiry
A competitive-intelligence firm receiving 20+ enquiry forms a month, each needing a fit judgement, a mapping against a 30-deliverable taxonomy and a tailored proposal — over 40 hours of senior business-development time monthly spent on intake rather than closing.
Around 30 hours of senior BD time recovered monthly — roughly S$2,400 at a fully-loaded S$80/hour, with no added headcount.
Logistics
4 hours under 30 min per rollout module
A postal and logistics operator translating dense vendor configuration documents into plain-language, floor-ready checklists during a multi-party system rollout — the bottleneck stalling training schedules before launch week.
Over 85% less translation time, about 35 hours across an estimated 10 modules. Vendor jargon reaching the floor targeted to zero.
Home services
4 hours under 90 min lead response time
An air-conditioning service business taking around 30 unstructured email leads a day, each read, categorised, checked against service history and matched to a technician slot by hand — with leads lost to faster competitors. Nothing is sent until a human reviews it.
About 7 minutes saved per enquiry — roughly 3.5 hours of admin time a day at 30 leads.
Finance operations
7.5 minutes under 2 per invoice
A process owner hand-keying 10–30 vendor invoices a month from PDFs and scans into Excel before uploading to SharePoint, where a mistyped amount has to be chased and corrected after it is already in the system.
Monthly entry time from ~2.5 hours to under 40 minutes — roughly 20–35 hours a year off low-value admin.
Regional marketing
6 hours under 2 per monthly report
A communications manager reconciling a shared tracker maintained by four PR agencies across Malaysia, Thailand, Singapore and Indonesia — hundreds of rows, four formatting conventions, and a slide deck built by hand every month.
About 48 hours — six working days — back each year, with a first draft ready 15 minutes after the automation runs.
Ten sectors, ten different tool stacks, and the same four moves underneath every one of them. This is the part that transfers — the tools will have changed by next year.
01
Every project names what the workflow costs today in minutes, hours or days. Without that number there is nothing to approve and nothing to prove — and no way to tell afterwards whether it worked.
02
In all ten, AI drafts and a person approves. Nothing reaches a customer, an employee, a regulator or a board pack unreviewed. This is the same governance model as the marketing stack running this site.
03
Invoice retyping, report assembly, policy lookup, jargon translation. Unglamorous, repetitive, and exactly where the hours actually are. The interesting-sounding use case is rarely the expensive one.
04
Several projects compared a no-code platform or a paid vendor suite and chose a simpler build instead — because it kept data inside an already-approved environment and needed no new vendor agreement.
In their own words, from reviews collected and verified via Endorsal. Read all reviews →
"After this course, I can produce management reports in minutes rather than hours. I built a repeatable reporting skill that I can use in my day-to-day work."
SK Tan
HR Practitioner
"Me, not a technical savvy person, with almost zero knowledge about AI — 3/4th in completing the course I comfortably know how to build workflows."
Tiam Fook
Advanced Certificate in Generative AI, SMU Academy
"Ivan's mentorship in SEO strategies was exceptional. His clear guidance and practical insights significantly enhanced our project's success."
Joyce Tay
Amili Pte Ltd
Trainer
Start with a free 30-minute discovery call. You explain your situation — Ivan will tell you honestly whether there is a fit and what a realistic outcome looks like.