Build, buy, or wait: making the software call in a growing Malaysian business

Over the years we’ve been pulled into a lot of software decisions — sometimes as the builder, often just as the second opinion. A pattern emerged early and has never once broken: companies rarely get hurt by choosing the wrong product. They get hurt by one of three timing mistakes.

The three failure modes

Buying too big. The mid-sized company that licenses an enterprise suite because the salesperson had good gula-melaka biscuits and a slide about “digital transformation.” Two years later they use 10% of it, the renewal invoice arrives anyway, and operations still run on the spreadsheets the suite was meant to kill. The software wasn’t bad. It was answering questions the business wasn’t asking yet.

Building too early. The startup that commissions a custom platform before its process has stabilised. Every pivot becomes a change request. The build burns cash chasing a moving target, and eventually gets abandoned for the off-the-shelf tool they should have started with. Custom software freezes your process in code — a terrible idea when your process is still finding itself.

Waiting forever. The most common and least discussed. The business that knows the spreadsheet is groaning — the Monday-morning report takes half a day, two systems disagree about inventory, one keyed-in typo caused that incident nobody jokes about — and yet keeps deferring the decision, because the current pain is familiar and any change is not. Waiting feels free. It’s the most expensive option on this list; you’re paying in staff hours and errors, permanently, in instalments small enough to ignore.

The staged path

The framework we actually use with clients is boring, sequential, and works:

Stage 1 — Spreadsheets, proudly. Early on, spreadsheets are correct. They cost nothing and change as fast as you do. The signal to move on isn’t embarrassment — it’s the first time two people need to edit the same numbers at the same time, or one manual error costs real money.

Stage 2 — Buy the boring layer. Standard accounting (SQL Account, AutoCount), standard payroll, a standard POS. Common processes deserve common software — we’ve written an honest guide to the ERP version of this call, including when off-the-shelf flat-out wins.

Stage 3 — Integrate before you replace. This is the most-skipped stage and the highest-leverage one. When your systems stop agreeing with each other — POS vs inventory vs accounts — the instinct is to replace everything with One Big System. Usually what you actually need is a thin platform that makes the existing systems agree: pull from each, reconcile, produce the one daily picture management wants. Smaller build, faster payback, and nobody has to be retrained.

Stage 4 — Go custom where you’re genuinely different. Every business has one loop where it wins or loses — and it’s different every time: for a delivery brand it was consolidating six sales channels into one order book; for a clinic group it was group-wide analytics that respect patient privacy; for a property agency it was turning a city’s office market into structured data. Build there, and only there. Everything else stays bought.

Stage 5 — Put agents inside the platform. Once a platform runs your operation, the repetitive work flowing through it becomes automatable in a way it never was inside spreadsheets — here’s what that looks like in practice. This is where the compounding starts: the platform gives the agents somewhere to stand.

The stages are a ratchet, not a race. Skipping ahead is how you land in failure modes one and two. Refusing to advance is failure mode three.

How AI moved the line (a builder’s confession)

Something real has changed in the economics, and it’s worth being precise rather than breathless about it.

AI-assisted engineering has cut the cost of custom software substantially — we rebuilt our own studio around it before we ever advised a client to. Senior teams ship in months what used to take a year. That moves the Stage 4 threshold: the size of business for which a bespoke platform is rational is smaller than it was three years ago, and the “integrate, don’t replace” builds of Stage 3 have become genuinely fast.

What AI has not changed: the judgment. Knowing which loop is your Stage 4, what to leave alone, and what sequence won’t disrupt operations — that’s still experience, and it’s still where projects live or die. Cheaper construction makes good judgment more valuable, not less, because you can now afford to act on it.

(It’s also created a new failure mode we’re starting to see: building too much, because building got cheap. The discipline of “only where you’re genuinely different” matters more now, not less.)

Deciding this quarter

If you’re staring at this decision right now, three questions cut through most of the fog:

1. Which stage are you actually at? Not aspirationally — actually. Most companies that think they need Stage 4 are at Stage 3.
2. What is the current mess costing, in hours? Count the Monday reports, the double entry, the error corrections. Multiply by salaries. That’s the budget waiting waste is already spending.
3. What would you automate if the data were clean? If the answer excites you, that’s your Stage 5 — and it tells you what the platform underneath needs to look like.

We have this conversation with founders regularly — sometimes it ends in a build, sometimes in “buy the off-the-shelf thing, call us in two years.” Both are good outcomes. The only bad outcome is another year of Monday mornings that start with copy-paste.