DEV Community: Sayed Rashid

Half of Singapore's Employers Won't Add Headcount in 2027. Here Is How You Grow Anyway.

Sayed Rashid — Thu, 10 Sep 2026 06:44:00 +0000

More than half of Singapore's employers do not plan to add headcount in 2027. Just over half plan to freeze or moderate wages. That is the Singapore National Employers Federation's poll, released on 4 September, and if you run a small firm the honest reaction is probably: same here.

So the question for the next twelve months is not "how do we hire?" It is "how do we take on more work with the people we already have?" Four other stories from the same week in early September, none of them about hiring, answer that better than any HR guide I have read. A consumer payments study, an air cargo standard, a hospital group's advice to medtech founders, and a newspaper feature on AI and jobs. Read together, they are one instruction. I will take them in turn.

The poll, and the line in it that matters for small firms

SNEF surveyed over 300 companies employing close to 160,000 workers across 20 industries between June and August 2026. The headline numbers:

54% do not plan to increase headcount in 2027. That is down from 58% a year ago, so marginally less frozen, not more.
40% plan to hire, up from 33%. 6% plan to cut, down from 8%.
51% plan wage moderation or a wage freeze, up from 48%.
83% name rising manpower cost as their top challenge, up from 79%.
30% worry about the cost of upskilling and reskilling their people, up from 23%.
48% are prioritising AI exploration and adoption. 46% are prioritising upskilling.

The business outlook is actually a touch better than last year. Fewer employers expect uncertainty in 2027 (63%, down from 72%). But SNEF's own summary is the part to underline: "Small and medium-sized employers were more likely to report uncertain business prospects and poorer business performance, and were also more likely to implement wage moderation or wage freezes in 2027." Retail and F&B are named as the sectors under the most pressure. SNEF calls the economy K-shaped. The firms with the least slack are the ones being asked to hold the line.

Now put three of those numbers side by side. 83% say the cost of people is the problem. 48% say AI is part of the plan. 30% are nervous about what it costs to train people for that plan. That is a business saying: we need more output per person, we know software is part of the answer, and we are not sure what it takes to get there. The rest of this piece is about what it takes.

What "AI stealing jobs" looks like at twenty people

The Business Times ran a long feature on 4 September under the headline "Is AI stealing your job?" Its own answer, in the standfirst: white-collar workers risk displacement, but mass layoffs aren't on the table. It traces the debate back to May, when labour chief Ng Chee Meng tabled a motion calling for no "jobless growth", an economy that expands while job creation lags, and to Manpower Minister Tan See Leng's worry that AI could erode Singaporeans' skills and experience.

Set that next to the SNEF poll and you can see what "no jobless growth" looks like from the employer's chair. Nobody is being fired. Nobody is being hired. The order book still has to grow. The gap between those two lines is what your systems have to absorb.

I wrote on 7 September about why jobs are the wrong unit inside a small business. Nobody in a 20-person company has one job. Your operations person spends perhaps 40% of her week retyping, checking and chasing, and 60% on judgment. The exposed part is the 40%, not her. I will not repeat the argument. I will add the arithmetic. A distributor doing 90 orders a day that grows 15% next year is doing 104 orders a day with the same team. If every order is still retyped from email into the system, that is 14 more retypes a day, every day, landing on someone who was already full. Either the retyping goes, or the growth does.

Your customers just told you how they want automation to behave

The most useful study of that week was not about employers at all. Ant International and S&P Global surveyed 6,000 consumers across nine markets in Asia, Europe and the United States about how they pay when they travel. Singaporeans, alongside consumers in China, Thailand and Malaysia, showed the strongest demand for integrated experiences inside a single app, well ahead of Japan, Germany and Western markets.

Then the two numbers that matter. 81% of consumers use AI tools for travel discovery and planning. Only 26% are interested in using AI for payments and autonomous decision-making. 43% cite privacy concerns, two in five prefer to plan themselves, and a third want human advice. Even on ordinary mobile payments, which 63% now use for most of their spending abroad, 62% worry about transaction security and one in four still carry cash as a fallback.

That is a design brief, and it applies inside your company as much as it applies to your customers.

First, everything in one place. If Singaporeans want the whole trip in one app, your customers want one place to order, pay, track and ask a question. Your staff want exactly the same thing internally. Every extra system is a tab someone has to remember to check, and the retyping between tabs is where the 40% lives.

Second, automate the research and keep a human at the money. 81 versus 26 is the rule. People are happy for software to do the looking: which stock is low, which customer is late, which quote has sat unanswered for four days. They want a person on the button when money moves. Most of the failed automation projects I have seen got this backwards. They automated the payment or the posting and left the research manual, so the team spent its week gathering information for a machine to act on. Flip it. Let the system prepare the reorder, the reminder, the draft invoice. A person approves. I laid out the approval thresholds that make this safe in The 3AM Purchase Order, and why every payment method is also an accounting object in the Stripe reconciliation piece.

The air cargo industry is retiring paper. Copy the principle, not the technology.

Singapore's air cargo community is working with hubs in the Netherlands, Qatar and the United States to adopt IATA's ONE Record standard. The technical detail is that it replaces the old batch messaging formats, Cargo-IMP and Cargo-XML, and the paper that travelled with them, with a single live digital record per shipment that everyone in the chain reads through web APIs. Customs officers see shipment data on screen during inspection instead of reconciling a stack of documents. Singapore's particular angle is pharmaceuticals, where live temperature and humidity readings from sensors attach directly to the shipment record, so cold-chain compliance is proven before the pallet touches the tarmac.

You do not need any of that technology. You need the principle, which is the one-record rule. One shipment, one record, and the forwarder, the handler, the airline and the regulator all read from it rather than re-keying it into their own systems.

Now count how many times one customer order gets re-keyed in your firm. Email to quote. Quote to sales order. Sales order to picking list. Picking list to delivery note. Delivery note to invoice. Invoice to the collections spreadsheet. Five or six retypes, each one an hour of somebody's week and an error waiting to happen. Allied Container found that the first 25% of its efficiency gain was just filing. If global freight, with its regulators and its dangerous-goods declarations, can agree on one record per shipment, a 20-person trading firm can agree on one record per order.

"Build with, not for": the medtech rule that applies to your back office

SingHealth is the Healthcare Innovation Partner at Medical Fair Asia, which runs at Marina Bay Sands from 9 to 11 September alongside Medical Manufacturing Asia and the first RehaCare Asia. Its message to the medtech companies exhibiting there is blunt. Associate Professor Goh Su Yen, SingHealth's group director for innovation and transformation, puts it this way: "There is a significant gap between a promising idea and a solution that can be adopted in healthcare." And: "Discussions about healthcare innovation often centre on the technology itself. We want to shift that conversation towards what it actually takes for an innovation to work in practice."

SingHealth's answer, through its Alice Lee Innovation Centre of Excellence, is to bring clinicians into product development early, rather than presenting them with a finished product and hoping it fits. The example it gives is an AI triage tool for chest-pain patients, co-developed with clinician-researchers at Singapore General Hospital, built around a problem the doctors on the ground identified first.

Swap "hospital" for "warehouse" and "clinician" for "your operations lead" and you have the reason most small-business system projects fail. The owner chooses the software. A vendor configures it. The team is shown the result on go-live day. Within a week the person who does the actual work has found the three things that do not fit how orders really arrive, and the spreadsheet is back. Goh's gap between a promising idea and an adopted one is exactly that week.

The fix costs about two hours a week of your ops lead's time for eight weeks. She sits in every design session. She owns the layout of the screens she will use every day. She decides what the exception looks like when a customer wants a partial delivery. You will lose some of her output during those eight weeks. You will get a system people use instead of a system people route around.

A 90-day plan for growing on flat headcount

Here is how I would run the next quarter if I had just read the SNEF poll and recognised my own firm in it.

Weeks 1 and 2: the hours audit. Everyone lists their recurring tasks and the hours each takes per week. Mark each one as retype, check, chase or decide. The first three are candidates for a system. The fourth is where you want your people spending more of their time. If you want to put a dollar figure on the first three, the free calculators do the arithmetic.
Weeks 3 and 4: pick one flow and apply the one-record rule. Order to cash, or purchase to pay, not both. Draw the current path of one order and count the retypes. The target is one entry that flows through quote, order, delivery and invoice without being keyed again.
Weeks 5 to 10: build with the doer. Weekly sessions, the ops lead in the room, screens designed around how the work actually arrives. Test with last month's real orders, including the ugly ones.
Set the money rule before go-live. Decide the thresholds. Below a value, the system acts and logs it. Above it, a named person approves. Your customers told you this is what they trust. Your team will trust it too.
Weeks 11 to 13: measure and show the team. Orders handled per person per week. Days from delivery to invoice. Hours in the retype, check and chase columns compared with week one. Then show people where the recovered hours went, because that is the answer to the 30% of employers worried about the cost of upskilling. The cheapest reskilling programme is a system that removes the low-value tasks so people learn the higher-value ones on the job.

Why so many Singapore SMEs stall at the AI pilot stage is mostly step two being skipped. There is no single record for the AI to read, so the pilot runs on exports and dies when the person doing the exports goes on leave.

The one thing to take from the week

Five headlines, one instruction. Employers are planning for fewer people per dollar of revenue. Consumers want everything in one place and a human at the money. Freight is moving to one record per shipment. Hospitals are telling builders to design with the people who do the work. And the jobs debate, once you shrink it to twenty people, is a debate about tasks.

If you do only one thing this quarter, find the flow with the most retyping and give it one record. Everything else in the list gets easier once that exists.

Which order in your business gets typed the most times before the customer pays for it?

Sources

Vulcan Post, 4 Sep 2026: More than half of employers in S'pore plan to freeze hiring & moderate wages: SNEF poll
The Business Times, 4 Sep 2026: Is AI stealing your job?
Asian Banking & Finance, 7 Sep 2026: Singaporeans lead demand for integrated travel apps but distrust AI for payments (Ant International and S&P Global study)
Travel And Tour World, 6 Sep 2026: Singapore teams up with other countries and regions for ONE Record cargo standards
The Business Times, 4 Sep 2026: How SingHealth wants medtech firms to build with hospitals, not just for them

I'm Sayed. I work closely with business owners to automate their operations and build workflows that actually make sense. I write The Gantry about ERP and automation for SMEs. If this was useful, the archive has the rest.

Originally published at The Gantry on 9 September 2026.

AI Will Not Hit Your Team Evenly. Here Is How to Plan For It.

Sayed Rashid — Thu, 10 Sep 2026 06:43:26 +0000

There is a serious argument being made in Singapore right now: AI will not affect workers evenly, so the policy response cannot be a blanket one. It has to be targeted, and it has to be fair.

Kok Ping Soon, chief executive of the Singapore Business Federation, has made the argument carefully, and he brings numbers. Graduates in full-time permanent employment within six months fell from 87.5% in 2022 to 74.4% in 2025. Graduates receiving no job offers at all more than doubled, from 4.1% in 2023 to 8.5% in 2025.

He names who is exposed: junior and mid-level professionals in finance, law, professional services and software development — the entry-level research, basic coding and analysis work. His phrase for the risk is the one worth remembering: "the career ladder will be weakened at the bottom."

And a counterweight most coverage skips. A Ministry of Trade and Industry study found that early AI adoption correlated with higher firm revenue and higher total employment — not immediate job losses. The firms that moved early grew. That is worth holding onto before anyone concludes AI means shrinking.

All of that is true at the national level. Almost none of it helps you on Monday morning, because you are not running a country. You are running a company where one person does the invoicing, the customer chasing and the stock count, and you cannot restructure the labour market — you can only decide what happens to her role.

In a small business, jobs are not the right unit

This is where the national framing breaks down.

In a large organisation there is a data entry team. AI arrives, the team shrinks. That is the story everyone tells.

In a 20-person company nobody is a data entry clerk. Your operations person spends maybe 40% of her week on things a system could do without her — retyping orders from email into the system, checking one screen against another, chasing the same five customers, rebuilding the same report. The other 60% is judgment: deciding what to do when a shipment is late, knowing which customer will accept a partial delivery, spotting that an order looks wrong before it ships.

The exposed thing is not her. It is those tasks. And the difference matters enormously, because it changes what you are deciding. You are not deciding whether to keep her. You are deciding what she does with 40% of her week back.

That is a much better decision, and most owners never frame it that way because the public conversation trained them to think in headcount.

"Just and targeted" applies inside your company too

The argument for a fair national transition is that the burden falls unevenly, so support should be aimed at the people carrying it.

Same is true in a business of thirty people, and it is entirely within your control.

Some of your team will be fine. They already work in ways where the tools help them go faster. Others — often the longest-serving, often the ones whose value is precisely that they remember everything — are the most exposed, because their advantage is the thing a system replicates best.

Those are the people who will quietly resist the new system, and they will be right to, because from where they stand it looks like it is aimed at them. If you do not deal with that directly, the project fails. Not through sabotage. Through a hundred small reasons why the old way was easier this week.

Deal with it directly and something else happens: the person who held all the knowledge becomes the person who configures the rules, checks the exceptions and trains everyone else. Their status goes up, not down. I have watched that conversion work many times, and I have watched projects die because nobody had the conversation.

The distinction worth stealing: cognitive vs embodied

Kok makes a split that almost nobody else does, and it is the most useful idea in the piece for anyone running a physical business.

Cognitive AI is the software kind — it reads, drafts, summarises, analyses. It substitutes for desk work, which is why it hits junior professional roles hardest.

Embodied AI is robotics — it moves things. And his argument is that Singapore should accelerate it, deliberately, in sectors that cannot find workers. He even proposes differentiated foreign worker levies to push it along.

For a warehouse, a workshop or a kitchen, that is the opposite of a threat. The labour shortage is the binding constraint on your growth, and this is the category of automation that eases it. Same word, "AI", pointing in two completely different directions depending on whether your bottleneck is desks or hands.

So when you read that AI is coming for jobs, ask which kind. If your constraint is that you cannot hire enough people to move product, embodied automation is on your side.

Our take: automate the transactions first, and be honest about why

There is a sequencing point here that gets lost in the AI excitement.

The repetitive work in a small business — retyping, matching, chasing, reporting — mostly does not need AI. It needs a system that captures data once and reuses it. That is ordinary ERP and workflow automation, it is well understood, it is predictable, and it is where nearly all the hours are.

AI is genuinely useful on top of that: drafting, summarising, forecasting from clean history, spotting anomalies. But it is the second step, not the first, and pointing it at a broken process just gets you wrong answers faster.

So the honest plan is: automate the transaction work, give people the hours back, and put those hours into the work that needs a human — customers, suppliers, quality, the exceptions. Then tell your team exactly that, in those words, before you start. Not as reassurance. As the actual plan, with what each person will be doing afterwards.

The businesses that handle this badly announce efficiency and let everyone draw their own conclusions. The ones that handle it well decide first what people will do with the freed time, and say so on day one.

Pick your most experienced operations person. What could she do with one extra day a week — and have you ever asked her?

Read Kok Ping Soon’s original opinion piece.

Put a number on the hours

The Cost of Manual Work Calculator takes three inputs and tells you what the re-typing, checking and chasing costs you every year — and how many full-time salaries those hours add up to.

Originally published at The Gantry on 7 September 2026.

Month-End Shouldn't Take Three Weeks

Sayed Rashid — Thu, 10 Sep 2026 06:36:12 +0000

Ask an SME owner what last month’s gross margin was, and you’ll usually get one of two answers: a guess, or “the accounts aren’t ready yet.”

Both answers cost money. You just can’t see the line item.

A slow close is an operations problem, not an accounting one

A three-week close isn’t an accounting problem. It’s an operations problem wearing accounting’s clothes. The close is slow because the month’s data was never captured properly while it was happening: delivery orders that never became invoices, supplier bills sitting in someone’s inbox, petty cash in a notebook, stock adjustments done “later.”

Finance doesn’t close the month. Finance reconstructs the month. That’s the three weeks.

How the three-day companies actually do it

The businesses that close in three days aren’t working harder in week one. They’ve simply made the transaction and the record the same event. The delivery order creates the invoice. The goods receipt creates the bill accrual. The payment matches itself against the statement.

When the month ends, there’s nothing left to reconstruct — you’re just reviewing.

Run this diagnostic on your last close

Here’s a diagnostic you can run this week: list the five slowest items in your last close. For each one, ask — when did the business event happen, and when did it get recorded? The gap between those two timestamps is your close time. Shrink the gap, and the close shrinks with it.

You don’t need a bigger finance team. You need fewer things that happen twice.

Where the three weeks actually go

When I sit with a finance person and time their close, the days never go where the owner thinks. Nobody is slow at accounting. They are slow at finding out what happened.

A typical close for a 15-person trading or services firm breaks down like this:

Days 1–4: sales catch-up. Delivery orders and completed jobs that were never invoiced. Someone walks the floor, opens the WhatsApp groups, and works out what actually went out last month.
Days 4–8: supplier bills. Bills sit in three inboxes and a drawer. Half arrive after the month has ended. Each one is typed in, coded, and matched to a purchase order that may only exist as an email.
Days 8–14: bank reconciliation. Payments matched to invoices by hand, one line at a time. If you take card or gateway payments, every payout is a net amount that has to be broken back into invoices, fees and FX before it matches anything (this gets worse with every payment method you add).
Days 14–18: stock and adjustments. The count says one thing, the sheet says another, and the difference has to be explained before cost of sales can be posted.
Days 18–21: the review. The only part that actually needs a finance brain, squeezed into the last three days.

Notice that the first eighteen days are not accounting. They are archaeology. The month is being reconstructed from evidence because it was never recorded while it happened.

How many days is acceptable?

The number I use with owners: if you see final numbers inside five working days, you are fine. Between five and ten, you are paying for the delay without noticing it. Past ten, the numbers arrive too late to change any decision they describe, which is the polite way of saying you are running the business on last month's picture.

A good close week looks unremarkable. Day one, most transactions are already posted because they were created by the events themselves. Day two, someone works a short list of exceptions the system flagged. Day three, the owner reads a margin report and asks questions. That is the whole thing.

This matters beyond convenience. A director's duty to exercise reasonable diligence assumes the numbers you review are current. Six-week-old numbers do not meet that bar, however carefully you read them.

What to fix first, in order

You do not fix a slow close by working on the close. You fix it by moving each record to the moment the event happens. Do them in this order, one at a time:

Invoice from the delivery, not from memory. The delivery order or completed job should become the invoice with one click, the same day. This alone usually removes the first four days.
Capture supplier bills on arrival. Bills come in by email; have them land in the system on the day they arrive, matched to the purchase order, waiting for approval. Nothing lives in an inbox.
Turn on bank feeds and let the matching run. Most invoices match themselves on amount and reference. Your finance person should only see the ones that did not.
Post stock adjustments on the day. A count difference recorded on the 12th is a small correction. Twenty of them found on the 30th are a mystery.
Set tax codes by configuration, not by habit. Once a customer or product carries the right GST treatment, every invoice inherits it. InvoiceNow is about to send that data to IRAS as you issue it, so this one is no longer optional.

Each step is a small project measured in days, not a system overhaul. Together they are the difference between a three-week close and a three-day one.

If you want to know what the current close is costing you in salary, the Month-End Close Cost Calculator will give you the number in a minute. It is usually larger than the software that would fix it.

Originally published at The Gantry on 21 August 2026.

Why Singapore SMEs Get Stuck at the AI Pilot Stage

Sayed Rashid — Thu, 10 Sep 2026 06:35:33 +0000

UOB polled business owners across the region and found something most vendors will not put on a slide: 65% of businesses have adopted AI in some form, but only 15% have reached what the study calls advanced capability.

Read that again. Four out of five companies using AI are stuck somewhere between "we tried it" and "it changed how we work."

The barriers the respondents named are where it gets interesting. Data and system readiness: 47%. Funding: 47%. Talent: 39%.

Funding and talent are the answers people give. Data readiness is the answer that is true.

The pilot always works. That is the problem.

Here is how it goes in real companies. Someone uploads three months of sales data to a chatbot and asks it to find the slow-moving stock. It comes back with a clean answer in eleven seconds. Everyone in the room is impressed. The pilot is a success.

Then you try to run it every week, across every product line, and it falls apart. Not because the AI got worse. Because the three months of data in the pilot were cleaned by hand by the one person who understood the mess. Nobody has time to do that every week for the whole catalogue.

The pilot worked on a sample. Production needs a system.

This is why the same study shows a 28-point gap in digitalisation success rates between large and small enterprises, and why only 69% of small enterprises here are digitalised against 93% of large ones. It is not that big companies have better AI. They have somewhere for the data to live.

What "data readiness" actually means

Strip out the consultant language and it means four unglamorous things.

One record per thing. One customer, one code. Not "ABC Trading", "ABC Trading Pte Ltd" and "ABC" as three separate lines that no tool can add together.

Numbers that update themselves. If your stock figure is a spreadsheet someone types on Fridays, every answer built on it is Friday’s answer.

History that survives people leaving. If the reason a customer gets 12% is in someone’s head, no model will ever find it.

A way to get the data out. If your system cannot export or connect, you will be pasting into a chat window forever — which is a demo, not a process.

None of that is AI. All of it has to exist before AI is worth paying for.

Our take: the order matters more than the tool

Business owners are being sold the roof before the foundation.

The sequence that works is boring and it is always the same. Get your master data clean — customers, products, suppliers, prices. Get your transactions into one system, so a sale, a stock movement and an invoice are the same event rather than three people typing. Then, and only then, put AI on top of it, where it has something real to read.

An ERP is not the exciting part of this. It is the part that makes the exciting part possible. Its actual job is to be the single place where things are true. Once that exists, AI stops being a demo and starts being a function: forecasting from real sales history, flagging margin drift as it happens, drafting the purchase order instead of the person who has done it 4,000 times.

Do it the other way around and you get exactly what the survey found. A pilot that impressed everyone, and no change to the P&L.

There is a second thing hiding in that data. 47% of businesses reported cost reductions from AI and 46% reported productivity gains. Those are real results, and they came from the companies that had somewhere to put it. The tool was available to everyone. The foundation was not.

What to do this quarter

Do not start with AI. Start with one question: if I asked for last month’s gross margin by product line right now, how long would it take and how much would I trust it?

If the answer is "a day, and not much" — that is your project. That is the whole project. Fix that, and AI becomes a small step afterwards rather than a leap you keep failing to make.

If the answer is "ten minutes, and completely" — you are in the 15%, and you should be moving faster than you are.

Rising costs were named the top challenge by one in three businesses in the same study, and improving profitability edged out growing sales as the top priority. That tells you what people actually need this year. Not a clever pilot. A business where the numbers are right and the manual work is smaller.

Read the original coverage of the UOB study.

Check your own readiness first

The ERP Readiness Scorecard asks eight questions on data, process, ownership and time — the things that actually decide whether a system project works. It will happily tell you "not yet", which is the most valuable answer it can give.

Originally published at The Gantry on 27 August 2026.