"Johnny Yang, we want to automate every process in the company."
I hear this several times a year. It usually comes about twenty minutes into the first meeting. The owner has finished talking about revenue, about not having enough people, about the system some competitor just put in. Then he leans forward and says it. He expects me to nod. I usually don't.
I'm Johnny Yang, Brightstream Technology, Taiwan. I've led implementation teams for seven years, across 252+ projects in 32+ industries. Put plainly, I sell shovels. In theory, the more an owner wants to dig, the happier I should be. But I've seen this on site too many times: a task that someone could have handled in twenty minutes gets turned into a mechanism that takes three months to build and needs someone to look after it every month. Six months later that person is more tired than before, and nobody dares touch the thing.
So here is the conclusion first: whether to automate something has nothing to do with how hard it is to build. It depends on four things — frequency, how stable the rules are, the cost of getting it wrong, and whether the thing is really a "process" at all. If it runs too rarely, don't build it, because you can't afford the maintenance. If the rules are still changing, don't build it, because you'll weld a wrong rule into the system. If a mistake is extremely expensive and hard to undo, don't make it fully automatic — make it "the system calculates, the system alerts, a person clicks confirm." And if it's really about a relationship, not a process, automation will slowly lose you customers. Before you put in any system, measure everything against these four tests. The money and time you save will be more than what you'd save by picking the right vendor.
Your instinct to automate is usually right
Let me take your side for a moment. Owners who want automation usually haven't been reading trend reports. They're in real pain. They see the sales support staff start reconciling accounts at 3:30 every afternoon. They see a shipping order copied by hand three times. They see the same data moved back and forth between Excel, a LINE group and a handwritten notebook. They know this isn't right. That instinct is correct, and more accurate than a lot of consultants' analysis.
Where it goes wrong is that "this annoys me" and "this should be solved with a system" are not the same thing. It might be annoying because the process itself is designed wrong. It might be annoying because this person shouldn't be doing it at all. Or it might be annoying only because it happens twice a month, but always on the busiest day. These three kinds of annoyance have completely different fixes. Only the third one is worth automating, and even then it depends on the cost.
Most of the projects I see go like this: the owner translates "pain" directly into "we need a system," then starts comparing vendors, reading feature lists, negotiating specs. What gets skipped is the most important step — deciding whether this thing should go into a system at all. Skip that step, and what you usually buy isn't a solution. It's a more expensive, harder-to-change version of the same problem.
Type one: things that happen only a few times a year
The real cost of automation isn't building it. It's maintaining it. This is the number owners get wrong most often. What you pay is a one-time build. What you actually carry is this: every time a rule changes, someone has to update it. Every time staff change, permissions have to be reset. Every time the system upgrades, it has to be retested. Every time it breaks, someone has to know where to look. None of these costs show up on the quotation, but they happen every month.
So the test is simple: how many times a year does this happen? If it happens every day and takes two hours of someone's time every day, automation almost always pays. If it happens three times a year, half a day each time, then something that took three months to build might need ten years to pay back — and by then nobody will remember how to use it. Parts of the annual stocktake, special reports for a handful of particular clients, the shareholder meeting materials you put together once a year: for things like these, a good template, clearly written steps and a named owner beat any automation.
There's a common pushback here: "But it's painful every single time." Pain and frequency are two different things. For low-frequency, high-pain tasks, the fix is to "prepare" them, not "automate" them — keep the data in a fixed place, write down how it was done last time, list who to ask. Next time it'll hurt half as much, and you won't have spent a cent.
Type two: things whose rules are still changing
The whole point of a system is to fix rules in place. That's its strength, and in this situation it's exactly what kills you. If you've changed the rules for something twice in the last three months, it shouldn't go into a system now.
The most typical examples I've seen are a new bonus scheme, new pricing logic, a newly opened sales channel. When these first launch, the owner is still experimenting: is this tier right, should that discount stay, how do we handle returns. Force them into a system at this stage and two things happen. First, development cost doubles, because every rule change becomes another change request. Second, and worse, once a rule is written into the system it becomes hard to change, so you start bending to the system — you think the tier is wrong, but a change takes two weeks, so you leave it as it is. Your business judgment has been taken hostage by the system. That's the most expensive cost of all.
"Stable rules" isn't a mysterious idea: nobody has asked to change it for three to six months in a row, the people running it can explain it without looking at a document, and you can count the exceptions, with no more than three. Once it reaches that point, put it into a system. Development will be fast, acceptance will go smoothly, and nobody will be complaining every day after launch. Until then, leave it in Excel or on a whiteboard. That's not falling behind. That's deliberately keeping flexibility.
Type three: things where a mistake is expensive and can't be undone
Some processes, when they go wrong, you fix the next day. Others, when they go wrong, the money has already been wired, the goods are already on the truck, the injection has already gone in. You can't judge these two kinds by the same standard.
The way to decide is to ask two questions: what's the maximum loss if this goes wrong? Once you find the mistake, how long does recovery take, and how many people? If the answer is "the loss could eat a whole month's profit" and "it takes three people two days to claw it back," then this should not be fully automated. Note that I'm not saying it shouldn't go into a system. The system should calculate the data, flag anomalies in red, and call up the people who need to look. But the final "submit" stays with a person. Payments, external quotations, physical inventory movements, and anything touching a customer's body or safety — I recommend this approach for all of them.
Owners often say at this point: so that's not really automation, is it? It is. The difference is that the person who used to spend forty minutes checking before daring to click now spends three minutes looking at the three anomalies the system flagged, and then clicks. What you've automated is the "checking," not the "deciding." Where to draw that line is what I spend the most time discussing with clients on site, and it's the main thing that decides whether anyone trusts the system after launch. After a few years of this, my view is that you should be conservative here. Trust in a system only has to collapse once, and from then on everyone goes back to opening Excel and quietly redoing the numbers. Then the whole thing was for nothing.
Type four: that's not a process, that's a relationship
This is the most overlooked of the four, and the one that hurts most. Some things look a lot like processes — they have steps, they have timing, you can write an SOP for them. But their value isn't in the steps at all. It's in who does it, and how the other person feels they've been treated.
The most common one is chasing payment. Automated emails cost close to nothing. Sounds perfect. But a long-time customer who is thirty days late reacts very differently to a canned system email than to a phone call from the salesperson saying, "Brother Wang, I took a look at that invoice for you — is something off with the statement?" Same goes for: the first response to a complaint, the last concession on a quotation, checking in on long-standing customers, the conversation with a senior employee before they leave. Automate these, and in the short term efficiency looks better. In the long term you'll find your regulars disappearing one by one, and you won't know why, because the system report has no column for "he feels you've gone cold."
Telling the difference isn't hard: if the other person knew a machine did this, would they be unhappy? If yes, don't automate it. You can automate "remind the salesperson to make this call today." You can automate "pull this customer's history together and put it in front of him." But the call itself has to be made by a person. This is the best split between automation and people — the system remembers, organizes and makes sure nothing is missed. The person provides those three minutes of warmth.
What to do once you've decided
Run your "things to automate" list through these four tests, and it usually shrinks by half. Only the remaining half is worth discussing further. Now the question becomes order — which one first. I go into more detail in which process to automate first. In short: find the one that's high-frequency, has stable rules, has a manageable cost of error, and that the most people will actually feel. Finish it first, get it live first, and get everyone to believe this is going to work.
The ones you crossed off aren't just abandoned. The table below is how I usually handle them on site:
| Situation | Don't | Do |
|---|---|---|
| Happens only a few times a year | Build a dedicated feature | Make a good template, write down the steps and who owns it |
| Rules still changing | Write it into the system's logic layer | Run it in Excel or on a whiteboard for three to six months and narrow down the exceptions |
| Mistakes are expensive and hard to undo | Fully automatic execution | System calculates and flags anomalies, a person makes the final click |
| It's really a relationship | Canned messages, automated mass sends | Automate "reminders and preparation," leave the contact to a person |
One more thing I need to say up front: holding back is much harder than adding one more item. When owners see a vendor's feature list, it's hard not to want all of it — it only looks like a bit more money. But every extra module is one more thing to maintain, one more thing to teach your staff, one more reason the go-live date slips. I covered this in less is more. Do implementation long enough and you find that companies that do less and finish it almost always beat companies that try to do a lot and only get 70% done.
One thing you can do tomorrow
Tomorrow morning, take a sheet of paper and write down everything you've thought "this should be automatable" about in the past six months. Don't organize it. Write whatever comes to mind. Then, next to each item, mark four numbers or symbols: how many times a year it happens, how many times the rules changed in the last three months, roughly the worst-case loss if it goes wrong, and whether the other person would mind that a machine did it.
Once you've marked them, you have your answer. Cross off anything with single-digit frequency. Set aside anything whose rules changed two or more times. Anything whose worst-case loss you can't afford becomes "human confirmation," not full automation. Anything the other person would mind, cross off entirely. Rank what's left, start the conversation with number one, and don't touch the rest yet. You can do this yourself in one afternoon. No consultant, not a cent spent. But it may be the highest-return step in the entire implementation.
If you're currently stuck with cash flow tight and nobody to spare, then I'd suggest putting even this list aside. Keep the company standing first. Getting the order right matters more than anything else — automation is for amplifying a business that already works, not for rescuing one that doesn't. If you do want someone to look at how to rank that list with you, Brightstream offers a free 30-minute diagnostic. Just bring the sheet of paper you wrote.




