ANASR SERVICES

Translate

Old method vs new AI

Old Method vs New AI: What I Actually Understand (And What I Don't)

By Aris Nasrol Osman, Chartered Accountant

Let me be clear from the start: I am not an AI engineer. I do not write neural networks. I do not deploy large language models alone.

But I am a Chartered Accountant who has built large IT systems (9-module ERP for Yayasan Sarawak) and managed multi-billion ringgit project finances (Pan Borneo Highway). And I have spent enough time studying AI to know what it can do, what it cannot do, and where the hype ends and reality begins.

This post is for business owners who are confused by the AI noise. I will tell you what I know – and what I don't.

What I Actually Understand (From Experience)

1. How data flows through large systems
When I managed the Yayasan Sarawak IT project (9 modules: Financials, HRM, Asset Management, Procurement, Business Intelligence), I had to understand how every department's data connects. Where does an invoice start? Where does it end? Who approves it? Where is the audit trail?
AI does not change this principle. AI sits on top of existing data. If your underlying data is messy, AI will give you messy answers – just faster.

2. Where audit trails break
In every large system I have worked on, the risk is not the technology. The risk is human processes – someone skipping a step, entering data twice, or approving something without checking.
AI can help detect anomalies (e.g., "this payment is 3x the normal amount"). But AI cannot fix your process. Only you and I can do that.

3. Integration is harder than the AI itself
Most businesses run on old systems – maybe spreadsheets, maybe an old ERP, maybe nothing at all. Connecting AI to these systems is called orchestration. This requires integration engineers, API specialists, and careful planning. I do not do this alone. I would collaborate with IT engineers, AI engineers, and integration engineers. My role is to ensure the financial and audit requirements are not lost in the technical work.

What I Am Still Learning (Honestly)

❌ I am not yet an expert in:
- Training custom AI models from scratch
- AI-to-AI agent protocols
- Deploying agentic AI at scale
- Writing production-grade AI code

✅ But I understand enough to:
- Ask an AI engineer the right questions (e.g., "How does this agent keep an audit trail?")
- Estimate whether AI is worth the cost for your specific business
- Design business processes that prepare your data for AI – even if you don't adopt it today

The Hard Truth About AI Costs (Most Consultants Won't Tell You)

AI is expensive. Here is what realistic costs look like (Malaysia context, 2026):

  • Basic automation (e.g., OCR + rules): RM10,000 – RM30,000 setup, plus monthly API costs.
  • Custom AI agent for anomaly detection: RM50,000 – RM150,000+ if you need integration with old systems.
  • Full agentic AI orchestration: RM200,000 – RM500,000+ – enterprise territory.

Is the market ready? For most SMEs in Malaysia – not yet. The ROI is still uncertain. But we are moving towards it. My advice: fix your data processes first. Clean data makes AI possible later. Messy data makes AI useless.

My Role (Realistic, Not Hype)

If you come to me and say: "I want to use AI for my accounting or audit" – I will not pretend to build it alone. Instead, I will:

  1. Assess your current data quality – Is it clean enough for AI? (Mostly, no.)
  2. Design the business process – What should the AI actually do? Detect fraud? Categorise expenses? Predict cash flow?
  3. Introduce you to AI engineers – I have a network of integration and AI specialists. I manage the finance/audit requirements. They handle the technical build.
  4. Be honest if AI is not worth it yet – Sometimes the answer is: "Fix your bookkeeping first. Come back in 12 months."

Quick Comparison: Old Method vs AI-Assisted (Honest)

TaskOld Method (Manual / Rule-Based)AI-Assisted (Current Reality)
Expense categorisationHuman reviews each receiptAI can guess 70-80% correctly – still needs human review
Anomaly detection (fraud flags)Periodic audit samplingAI can flag outliers in real time – but many false positives
Audit trail completenessHuman designed, system enforcedSame principle – AI does not fix broken processes
Cost for SME (setup)RM5k – RM20k (ERP module)RM30k – RM150k+ (AI integration)

Source: My own experience + conversations with AI engineers. Your mileage may vary.

šŸ“§ Email me if you want to discuss AI honestly – no hype, no pressure

Oher blog post

Featured post

“YOUR BUSINESS HEALTH REPORT” V2

Your Business's Health Report: What Financial Statements Actually Tell You (And What They Don't) By Aris Nasrol Osman, Chartered Accountant Let me start with something most accountants won't say: A profit and loss statement alone will not save your business. Neither will a balance sheet. Or a cash flow report. Why? Because financial statements are backward-looking . They tell you what happened last month, last quarter, or last year. They do not tell you what will happen tomorrow. But here is what they can tell you – if you know how to read them honestly. What A Business Health Report Actually Does (No Hype) Think of your financial statements like a medical check-up. A blood test does not cure you. It gives you data. Then you and your doctor decide what to do next. Similarly, a Business Health Report gives you: A clear picture of where money came from (sales, loans, investments) A clear picture of where money went (expenses, salaries, supplies, taxes) An...