This post is for the CEO, COO, or CTO who keeps hearing "digital transformation" in every vendor pitch and board meeting but hasn't been given a straight answer about what it actually takes. After 20+ years of running DX projects — from a 200-outlet coffee chain's data warehouse to AI-powered rice farming — here's my honest framework. No jargon. No magic quadrant. Just the lessons I've paid for with weekends and production outages.
Start with the Uncomfortable Truth: Your Data is a Mess
Let's skip the AI hype for a moment and talk about what's actually sitting in your systems right now.
Most businesses I've worked with have accumulated technology the same way families accumulate kitchen gadgets — one at a time, each solving a specific problem, none talking to each other. A CRM for sales. An ERP for finance. A marketing automation tool someone's cousin recommended. An eCommerce platform. A Zalo group for "urgent" requests.
The result? Every department has its own silo, and "data cooking" has become a full-time job. Someone in operations is spending 5-10 hours a week manually copying numbers from one system into a spreadsheet for a report that's outdated before it reaches management.
Sound familiar? You're in the 99%.
The 70/30 Rule Nobody Tells You
Here's the insight that saves my clients from expensive mistakes: digital transformation is 30% technology and 70% process reengineering, change management, and knowledge capture.
The tech part — setting up a data warehouse, automating a pipeline — is relatively straightforward. The hard part is getting your team to stop using WhatsApp as a project management tool, to document decisions in a central platform instead of someone's memory, and to restructure workflows so the new tools actually get used.
I've seen companies spend $200,000 on a shiny ERP implementation and then watch their staff bypass it because the old spreadsheet was "easier." The technology wasn't the failure. The change management was.
The Sequence That Actually Works
After enough projects, I've settled on a sequence. It's not glamorous, but it works:
Step 1: Fix Your Data Infrastructure
Before you touch AI, before you buy another SaaS tool, answer this question: do you have a single place where all your transactional data lives, is clean, and is queryable?
If the answer is no, that's your first project.
For one of our clients — a major Vietnamese coffee chain with over 200 outlets — this meant consolidating 5 different POS systems and multiple accounting databases into a single BigQuery data warehouse. The result was a 70% reduction in reporting time and, for the first time ever, real-time visibility across all locations.
The practical steps:
- Audit your data sources. List every system that generates transactional data. You'll be surprised how many there are.
- Build or buy a data warehouse. Cloud options like BigQuery or a self-hosted solution on your own infrastructure — we've done both. The choice depends on your security requirements and budget.
- Hire or designate a data person. Not a committee. One person who owns data quality and pipeline maintenance.
Why this matters long-term: in 3-4 years, your tools will change. Maybe you'll swap your POS system. Maybe you'll get a new CEO. But your core data — financial reports, operational metrics, customer records — persists. Invest in the layer that outlives any individual tool.
Fun fact: I learnt SQL from programming books in 1998, and I've been using it actively in production work ever since. The tools change. The fundamentals don't.
Step 2: Get Your Knowledge Out of People's Heads
The second most common bottleneck I see isn't technical at all — it's tribal knowledge. Critical processes live in the heads of three people. Company policies exist in a Zalo group from 2019 that nobody can find. Onboarding a new employee means shadowing someone for two weeks because nothing is written down.
We ran a workshop for Shopee's operational leaders where the pre-workshop survey confirmed exactly this: most participants relied on "pinging colleagues in chat groups" to find internal policies. The fix wasn't AI — it was building a centralized knowledge base using tools like NotebookLM and teaching teams to maintain it.
This doesn't require a six-figure investment. It requires discipline.
Step 3: Now You Can Talk About AI
Once your data is clean and your processes are documented, AI becomes genuinely powerful — because it finally has something accurate to work with.
I'll give you a number that makes the point better than any pitch deck: until late 2024, my own ERP development teams were 20-25 people — project managers, product owners, business analysts, UI/UX designers, tech leads, backend devs, frontend devs, mobile devs, DevOps, QA. That was the "lean" version. In early 2025, I had to let go of my entire team due to cost pressures. Today, to build the same calibre of ERP system, I need three people: a project manager, one fullstack developer, and a QA engineer. AI fills every other seat. That's not a projection — that's my actual org chart right now.
But here's the critical point: that only works because our data infrastructure and processes were already solid. The AI agents pulling from messy, undocumented systems would produce confident garbage.
Another example: we work with hybrid retail businesses — physical stores plus eCommerce — where a handful of AI agents now handle 80%+ of customer service interactions, at a monthly cost comparable to a single senior employee's salary. The agents pull from clean product data, structured FAQs, and documented escalation procedures. Without Steps 1 and 2, these agents would hallucinate confidently and damage the brand.
For the CTOs in the room — managing AI's non-deterministic behaviour is an engineering problem, not a magic problem. Our approach is divide and conquer:
- Break workflows into deterministic logic flows. If/else at the core — the AI handles the grey areas within bounded contexts.
- Use visual workflow tools like Node-RED to orchestrate API calls and feed context to AI tasks.
- Maintain a structured knowledge base that the AI references. We use Notion internally — but the tool matters less than the habit of keeping it updated.
- When working with transactional databases, the AI reads from clean aggregated reports via RAG, augmented with templated SQL queries for real-time management questions. The AI never writes to production databases.
With this approach, we run multiple AI agents across project management and client-facing tasks with 99.9% control over outputs.
Where We've Taken This Further
The same framework — clean data → structured processes → targeted AI — applies beyond office work:
- Digital Agriculture: We use Node-RED to control sensors, monitor crops, and digitise rice farming operations in the Mekong Delta. The data infrastructure is identical to what we'd build for a retail chain — InfluxDB for time-series data, Grafana for dashboards, MQTT for sensor communication.
- Cold Energy Storage (I.C.E. Battery): Our IoT monitoring system processes over 10,000 data points daily from thermal storage installations, using the same distributed architecture we deploy for client projects.
The technology transfers. The pattern is always the same.
The Resistance You're Going to Face
I'll be direct: your transformation will face resistance from every level. Employees will push back because change is uncomfortable. Middle managers will resist because process transparency threatens the information asymmetry that some rely on. Even leadership will wobble when the first quarter doesn't show dramatic ROI.
This is normal. This is why the 70% is about people, not technology.
When I spoke to Home Credit's tech team at their offsite retreat earlier this year, the question that came up most wasn't "which AI tools should we use?" It was "how do we get the rest of the organisation to actually adopt this?" The answer is always the same: leadership has to commit publicly, accept the 6-12 month timeline, and not retreat to the old spreadsheet the moment things get difficult. Frankly, the world does not care whether you are ready or not — businesses still go on, competitors still adopt, and the gap widens every quarter you wait.
One Universal Truth
After 20+ years of driving these projects, I've learned one thing that applies to every single engagement:
Technology alone doesn't transform businesses — mindset does.
That's the lesson from every engagement we've run. The technology is the easy part. The rest is leadership, patience, and showing up every day to maintain the discipline until it becomes habit.
We've covered the technical side of this framework in more detail: Data-First Principle Thinking, AI workflow automation for operations teams, and how a 200-outlet coffee chain consolidated 5 POS systems into one data warehouse.