Your Competitor Can Buy the Same AI. They Can’t Download What You Know About Your Business. 

AI is making it easier for SMEs to automate work, build software, and introduce capabilities that once required significant technical resources. But there is a catch: technology can only work with what a business actually knows about itself.

For Charles Lau, Founder of Digital Perpetual, that lesson comes from working directly inside SME operations. After 17 years in MNC digital marketing, six years in business, and four years studying IT, he now helps SMEs replace manual work with practical digital systems across sales, operations, HR, accounting, CRM, inventory, and delivery.

What he often finds is not a lack of technology. SMEs often struggle with fragmented customer records, undocumented processes, inconsistent definitions, and crucial operational knowledge that exists only in someone’s head.

In this conversation with NewInAsia, Charles explains why the rush toward AI can expose those gaps, why businesses should keep certain processes manual, and why understanding exactly how your business works could become a bigger competitive advantage as technology gets cheaper.

Your SME Has AI. But Does It Have a Customer List?

Q: Singapore is pushing businesses toward deeper AI adoption, yet you still encounter SMEs relying heavily on spreadsheets, paperwork, and disconnected systems. Are we asking SMEs to become AI-ready before many are digitally ready?

A: Yes — and the way it usually shows up is not that a company has no technology. It’s that they have AI before they have a customer list.

One client, a training and coaching business, already had an automated WhatsApp assistant answering enquiries when I came in — by any headline measure, an AI-adopting SME. But their customer data lived in three unconnected places: two lead-generation workshops in spreadsheets, purchases inside their course-sales platform, and website sign-ups. Nobody could answer, “How many people do we actually know?” without opening three things and eyeballing the overlap.

We rebuilt it around one database, and the first honest result was uncomfortable: around 5,500 contacts once merged, and a large number of them uncategorisable.

So we did what sounds like an admission of defeat but was the most useful design decision in the project — we created a stage called “Unknown”, a triage bucket for anyone whose relationship to the business nobody could state. Not a prospect, not a customer — unknown. Only once that bucket existed could the daily syncs start filling it in properly.

That’s the state of a lot of SMEs right now. The AI layer is bolted on and visible; the layer underneath — one list, one definition of a customer, one place a name lands — is missing.

AI doesn’t fix that. It reads whatever you have and answers confidently, so an inconsistent foundation just produces confident inconsistency, faster.

So when policy says “become AI-ready,” I translate it as: get your information into a state where a machine could read it without asking a clarifying question.

Most SMEs are one honest audit away from finding out they’re not there. The encouraging part is that the audit is cheap, and the fix is usually just a few weeks of deciding what things are called.

When “100 PCS” Doesn’t Actually Mean 100 Pieces

Q: AI is often presented as a shortcut to productivity. From what you’ve seen on the ground, what problems do SME owners commonly try to solve with AI that are actually process problems in disguise?

A: The clearest case came from something as mundane as a supplier invoice at a business supplying hardware and equipment to ships.

Their overseas supplier sends costing sheets stating quantities like “100 PCS/PKT.” A reasonable person — or a language model — converts that literally. But we found the identical wording maps to a multiplier of five for one product, twenty for another, and one for a third.

The factor was never in the text. It lives in the head of the person who opens the carton.

Then it got worse: one product had been booked at a factor of one in January and twenty in June. The company’s own history disagreed with itself. No model resolves that from the document, because the answer isn’t in the document.

What we built wasn’t clever: a table of product code, from-unit, factor, and to-unit, seeded with 146 conversions mined from real shipments.

Where the history conflicted, the system stops and flags the line for a human rather than guessing. We only set a rule after checking five years of shipments to establish the true basis. Every conversion preserves value, so the invoice total reconciles regardless of which factor applies. Even when the system is wrong about the unit, it’s never wrong about the money.

That’s process work — tedious, unglamorous, and exactly what made the automation trustworthy.

Ask an AI to “just read the invoices,” and it would have produced beautifully formatted numbers that were quietly wrong, uncaught until a stock count months later.

So the honest summary is this: when an SME owner says, “I want AI to tell me what to order,” what they need first is agreement on what a product is and what a unit means.

AI is an amplifier — amplifying a process that disagrees with itself just produces louder disagreement.

The Best SME System Might Be a WhatsApp Message

Q: You’ve worked inside large MNCs and now directly with SMEs. What assumptions about digital transformation work well in an enterprise but break down completely when applied to a small business?

A: Let me answer through an engineering firm I’ve been digitising.

The first enterprise assumption is that there’s an old system to migrate from.

There wasn’t one — the company’s entire commercial history lived in a shared cloud folder of Excel and PDF files. Migration in an MNC means mapping fields between two databases; here it meant writing a parser for how a human had laid out a spreadsheet, and loading several hundred purchase orders and payment vouchers, and thousands of line items, out of that folder.

There was no schema. The schema was a habit.

The second assumption is that master data exists.

Consolidating their company records out of four separate modules produced 275 companies — roughly nineteen were the same customer spelled differently. An enterprise has a data governance function whose job is preventing that. An SME has whoever typed fastest that afternoon.

The third — and the one I’d most want other consultants to hear — is that people need tools that actually get used.

That firm needed staff attendance; the enterprise answer is a biometric terminal and an HR module. What we built instead was a WhatsApp punch-card — staff send a message with a photo, and it records against the project.

Not the sophisticated answer, but the one with the highest chance of being used daily, because it asks nobody to learn or carry anything new.

There’s also the plain arithmetic of attention. In an MNC, transformation has a sponsor, a project manager, and a training budget. In an SME, the “project team” is the operations manager who’s also quoting jobs and covering the front desk.

Any plan requiring forty hours of that person’s attention won’t happen. So we ship in small pieces, each useful the day it lands.

The one enterprise discipline that transfers completely is a single source of truth. Large companies learned that lesson painfully. Small ones haven’t yet, and the cost of ignoring it is proportionally worse when you’re small.

One caution runs the other way: MNCs buy a platform and bend the business to fit it, because standardising across forty offices beats local optimisation.

For an SME, the process often is the competitive advantage — the reason a customer calls them, not the bigger competitor. Bending that to fit off-the-shelf software is how a small company makes itself average.

Sometimes the Smartest Digital Transformation Is Deleting the Feature

Q: There is enormous pressure for SMEs to “transform,” but every new CRM, automation, or AI tool also introduces cost and complexity. When is staying manual actually the smarter business decision?

A: I’ll answer with two things I built and then deleted for a construction project company.

The first was a 3D model viewer inside their project portal — upload a model, view it in-browser, tie it to project stages. Technically satisfying.

Then we watched who used it: nobody.

So we stripped it all out.

The second was worse, because it was the “correct” enterprise design — custom team roles, per-workflow permissions, a folder assignment matrix.

I removed all of it and replaced it with shared access plus a single gate on the one folder that genuinely needed protection. In a firm where the team sits in one office and covers for each other daily, a permissions matrix protects nothing — it just generates a phone call when someone can’t open a drawing at 7am on site.

Both deletions made the product better, because every screen you don’t need is a screen someone has to learn, scroll past, and eventually ask you about.

My rule of thumb has three parts.

Automate what’s high-volume, stable, and boring — order entry at a hundred a week qualifies.

Stay manual where it’s low-volume or high-judgement — do the arithmetic honestly; under an hour a month will almost never pay back.

And respect the maintenance tail: every automation can break at an inconvenient hour, usually because someone upstream changed a format.

That last one is underrated. At one client, we restricted a user account to a specific office IP range for security—within days, our team had to edit the rule three times as the real-world network reality became clear.

Nothing about that is a failure; it’s just what owning an automation feels like.

A large company absorbs it with an IT team. An SME absorbs it with the owner’s evening. So the real question is never, “Can this be automated?” but “Is this worth owning forever?”

A fourth case catches people out: don’t automate a process that’s still changing.

If your quoting workflow is different this month from last, automating it means rebuilding it every month. Let it settle — sometimes “we’re still figuring this out” is the correct answer, and a spreadsheet is genuinely the right tool for a process that hasn’t found its shape.

The Competitive Advantage Your Rival Can’t Download

Q: As AI makes it easier to build software and automate workflows, could the real competitive advantage for SMEs become something much less glamorous: simply understanding their own operations better than their competitors do?

A: I think that’s exactly right, and it’s already happening.

Building software is no longer the bottleneck. Across our client systems, close to five thousand logged changes span roughly a hundred systems—and the constraint has never been how fast developers write code. It’s how fast we find out what actually happens on the floor.

A small example has stayed with me.

We built a live kitchen dashboard for a dessert brand — orders arrive, and the kitchen sees a running summary by flavour, sugar level, and portion count.

It worked, except bundled products appeared as their own separate flavours, so the kitchen was preparing the wrong mix.

The fix wasn’t technical: a bundle’s dessert component is prepared exactly like one standard flavour and should be counted with it.

That knowledge existed nowhere in any system. It lived in the kitchen, in a person, and until someone thought to ask, no software was going to get it right.

That pattern repeats at every client.

The hard part is not the build. The hard part is the question — and the question can only be answered by someone who genuinely understands the operation.

That’s encouraging for small businesses. If code is becoming cheap, the durable advantage shifts to whoever can describe their own business most precisely — what the steps really are, where the exceptions live, and which decisions actually require judgement.

That’s not a technology budget; that’s attention.

A twenty-person firm can out-think a two-thousand-person one at it, because the owner is close enough to the work to know the truth.

So my advice to SME owners is deliberately unfashionable: before you buy anything, write down how you actually work — not how you think you work, and not how the vendor’s diagram says you work.

Sit with the person doing the job and watch.

That document is now a real business asset, because it’s precisely the input AI needs to be useful to you. And unlike the tools, which get cheaper every quarter, that understanding is the one thing your competitor cannot download.

Final Thoughts: Before You Ask AI, Ask the Person Doing the Job

As AI lowers the barrier to building software and automating workflows, the differentiator for SMEs may shift away from simply having access to technology.

The examples Charles encounters on the ground point to something less glamorous but more consequential: knowing what a customer is, what a unit means, why an employee performs a task a particular way, where exceptions occur, and which processes are stable enough to automate in the first place.

For SMEs, that proximity to the work can be an advantage. The owner may be closer to the people who know why the spreadsheet looks the way it does, why the kitchen handles a bundle differently, or why the supposedly sophisticated system is making someone’s job harder.

AI can make building cheaper and execution faster. But it cannot automatically recover operational knowledge a business has never captured.

Before asking what AI can transform, the more valuable question may be much simpler:

Do you understand your own business well enough to tell it what to do?

Read the Chinese article here.