A 2026 progress report credits 300,000 firms and NT$11.27bn of output; the 87% under 30 staff and a missing statute are the harder story
By: Robert Sterling – AsiaPresswire – On 21 October the National Development Council met to review the Small and Micro Enterprise Diversification Programme, the four-year, NT$11.6 billion package that the Executive Yuan launched in January 2025 to help Taiwan’s smallest companies digitise, decarbonise and find buyers abroad. The progress report was, by the standards of these documents, impressive: more than 300,000 enterprises assisted, NT$11.274 billion of output value created, NT$257.05 billion of loans approved.
Three numbers, three different kinds of claim, presented as if they were the same kind. That is the first thing worth noticing, and it is not a criticism of the programme. It is a criticism of how the programme is being reported back to the public that pays for it.
Taiwan has 1.715 million small and medium enterprises, 98% of all registered companies and roughly 79% of employment. Any intervention at that scale is a rounding exercise, and the honest question is not whether 300,000 is a lot. It is what 300,000 firms received, how much of it was money, and what would have happened without it. The public version of the answer does not say.
Three hundred thousand is not a denominator
Start with arithmetic. Three hundred thousand assisted firms against 1.715 million is 17.5%. Against the roughly 1.5 million firms with fewer than 30 employees, which is the programme’s actual target group, it is about 20%. Against the programme’s own stated goal of 600,000 firms over four years, it is half the target at the 21-month mark, which is roughly on schedule if the pace holds and roughly meaningless if “assisted” includes a firm that attended one seminar.
That is the crux. The term used in the report, rendered as coaching or guidance, covers everything from a three-year consultancy engagement to a single advisory phone call, a subsidy application the firm abandoned, or a listing in a matchmaking database. There is no published breakdown by intensity. Without it, 300,000 cannot be compared with anything, including next year’s number.
A firmer figure does exist inside the same report, and it is much smaller: more than 50,000 firms that actually adopted digital tools. That is 3% of the SME population and about a third of the headline number. It is also the only number in the release that corresponds to a change in how a company operates.
Other jurisdictions have run into the same reporting problem and solved it in different ways. Korea’s smart factory programme, which had a comparable ambition in the 2010s, eventually published a breakdown by level of intervention, from basic data collection to full automation, because the aggregate count had become useless for evaluation. Japan’s IT introduction subsidy publishes disbursement by firm size and by software category, which makes it possible to see whether the smallest firms are actually being reached. Taiwan’s release has the raw material for both tables. It has chosen not to publish them.
An output figure lands suspiciously close to the budget
The programme’s budget is NT$11.6 billion. The output value it reports having created is NT$11.274 billion. The two numbers sit within 3% of each other, and that coincidence deserves a question rather than admiration.
Output value in Taiwanese programme reporting, is usually a modelled estimate: the assisted firm’s incremental revenue, sometimes self-declared in an application, sometimes extrapolated from a sample, rarely audited against tax returns. It is not the same as value added, and it is certainly not the same as tax revenue or wages. A NT$11.27 billion output figure at a 20% corporate tax rate on a 6% net margin would be about NT$135 million of tax, or roughly 1.2% of what the programme cost. Nobody in the release claims otherwise. But a reader skimming the headline will infer a return, and the release does nothing to discourage that.
The comparison that would actually settle the question is not output against budget. It is the audited change in revenue, payroll and tax paid for assisted firms against a matched control group of similar firms that did not apply. Taiwan has the administrative data to build that comparison: the Ministry of Finance holds business tax filings, the Ministry of Labor holds insured payroll records, and the two can be linked at the firm level. Until that comparison is published, the output number is a claim, not a finding.
The absence of a control group is not an oversight peculiar to this programme. It is the default in industrial policy everywhere, because the counterfactual is unflattering to whoever commissioned the evaluation. A firm that applies for a digitisation subsidy is, almost by definition, a firm that was already inclined to digitise. Some share of the reported output would have happened without any subsidy at all, and the honest question is what that share is.
Loans are not spending
The NT$257.05 billion of approved loans is the largest number in the release and the least comparable to the budget. It is credit, extended by banks, mostly guaranteed by the SME Credit Guarantee Fund rather than paid out of the NT$11.6 billion. The fiscal cost is the guarantee fund’s expected loss, which is a small fraction of the loan book.
This is a real and useful policy instrument, and it is being described in a way that flatters the programme. A guaranteed loan helps a firm that has a bankable project and lacks collateral. It does nothing for a firm with no project. Presenting NT$257 billion alongside NT$11.6 billion invites the reader to add them, and adding them is meaningless.
The pricing inside those loans deserves a closer look, because it tells you who the programme is really for. The multi-purpose loan carries a rate cap of 2.22%, a 1.5 percentage point discount for the first six months on amounts under NT$10 million, and a guarantee ratio of 100% for loans below NT$1 million, falling to at least 90% above that. A 100% guarantee removes the bank’s credit risk almost entirely. That is good for the borrower and it is exactly the design you want when the goal is speed, but it also means the bank’s incentive is to process the file, not to assess the project.
At 2.22% against a typical small-business lending rate in the 3% to 4% range, the interest saving on a NT$3.5 million loan is roughly NT$35,000 to NT$60,000 a year. Useful, and smaller than the accounting cost of the application for a firm that has to hire someone to prepare it. The firms that benefit most are the ones that already have a bookkeeper and an existing credit line, which is to say the ones that would have borrowed anyway.
The same conflation runs through the tariff response. As of 15 October, the Ministry of Economic Affairs reported four measures under a NT$46 billion special budget, opened for applications on 7 August 2025.
| Measure | Approved | Amount | What it actually is |
| Export loan guarantees | 192 cases | NT$1.284bn guaranteed financing | Credit guarantee, not a grant |
| SME multi-purpose loans | 2,203 cases | NT$20.49bn disbursed | Subsidised lending at a 2.22% cap |
| R&D and equipment grants | 1,276 cases, 1,418 firms | NT$6.019bn granted | Cash grant, 50% cost-share required |
| Overseas order development | 168 cases, 185 firms | NT$620m granted | Cash grant, 50% cost-share required |
Read the right-hand column and the pattern is clear. The measures that move the biggest numbers are credit. The measures that are actual spending are small, require the firm to put up at least half the cost, and have reached about 1,400 firms in fourteen months. Fourteen hundred firms is 0.08% of the SME population, and the recipients cluster in metal products and machinery, which is exactly where the tariff shock landed.
A NT$100,000 voucher against a NT$29,500 minimum wage
The programme’s flagship digital measure gives a firm with 30 or fewer employees up to NT$100,000 to train staff in digital skills. The design logic is explicit in the Executive Yuan’s own briefing: the subsidy is meant to offset the cost of the minimum wage increase, which for 2026 is NT$29,500 a month and NT$196 an hour.
Set the two against each other. NT$100,000 is 3.4 months of one minimum-wage worker’s salary, before the employer’s labour insurance and health insurance contributions, which add roughly another 20%. It is a meaningful sum for a three-person shop and a rounding error for a thirty-person factory. It is also a one-off, while the wage increase is permanent.
None of that makes the voucher a bad idea. It makes it a small idea being sold as a structural one. The structural problem the programme names, in its own documents, is that small firms lack the knowledge to specify what to buy. A NT$100,000 training credit does not create that knowledge; it pays for a course, and the course is chosen by the same owner who did not know what to ask for.
Do the unit arithmetic on the voucher and it shrinks further. Twenty staff on a one-day course at NT$5,000 a head is NT$100,000, minus the trainer’s travel and the materials. What a bakery with eight employees can do with the same money is send two people to a two-day spreadsheet and inventory course and have nothing left for the software. The voucher pays for learning. It does not pay for the thing learned.
Employers also carry roughly 20% on top of wages in labour insurance, health insurance and pension contributions, which is why the minimum wage increase lands harder on small firms than the headline percentage suggests. A 3.2% rise in the monthly minimum becomes a 3.2% rise in a cost that is already 1.2 times the wage. For a firm with fifteen minimum-wage staff, that is around NT$500,000 a year of additional cost, five times the one-off voucher.
Net zero is the thinnest line, and the most honest
Buried in the same progress report are two figures that deserve more attention than they got: more than 90,000 firms advised on net-zero transition, 228,100 tonnes of CO2 equivalent reduced, and 418 million kilowatt-hours of electricity saved.
Divide the electricity saving by the firm count and you get about 4,640 kWh per firm per year. For a small machine shop with a 30-kilowatt connected load running 2,000 hours, annual consumption is around 60,000 kWh, so the reported saving is roughly 8% of one firm’s bill. That is a plausible number for a lighting, compressed-air and motor-replacement programme, and it is small enough to be believable.
Now cross-check the carbon figure against the electricity figure, which is where programme reporting usually falls apart. Taiwan’s grid emission factor is around 0.47 kilograms of CO2 per kilowatt-hour. Four hundred and eighteen million kilowatt-hours at that factor is about 197,000 tonnes, which is within 15% of the 228,100 tonnes claimed, with the difference explained by non-electric measures such as fuel switching and refrigerant replacement.
This is worth saying plainly, because it is the one place in the release where two independently reported numbers reconcile. The output value figure does not reconcile with anything. The net-zero figures do. A programme that can produce consistent numbers in one workstream and not in another is not a programme with a data problem. It is a programme with a reporting priority problem, and the priority is set by what makes a good headline.
The statute that does not exist
Here is the detail that best explains the shape of the whole thing. At the same 21 October meeting, the National Development Council asked the Ministry of Economic Affairs to speed up drafting a Small and Micro Enterprise Transformation and Upgrading Act.
A four-year, NT$11.6 billion programme with a stated target of 600,000 firms is, at present, running on administrative orders and annual budgets rather than a statute. That matters for three reasons. It means the tax incentives are scattered across existing laws rather than consolidated. It means the programme’s legal basis can be revisited every budget cycle. And it means the target group, which is defined by employee count, has no statutory definition that survives a change of government.
The tax measures themselves are real: an R&D credit for SMEs, a deferral on intellectual property contributed as capital, incentives tied to hiring and to wage increases. They are also the kind of measure that a firm needs an accountant to use. The Bureau of Industry’s own outreach material walks applicants through it, which tells you something about the complexity.
Where the money went
The distribution inside the tariff package is the most informative thing in either release. Metal products first, machinery and equipment second, plastic and rubber third, wholesale fourth. Those are the sectors exposed to a 20% US tariff and to the customer pressure that follows it: a buyer in Ohio asks a Taiwanese fastener supplier to absorb part of the duty, and the supplier either absorbs it or loses the order.
The firms in that position are not micro-enterprises. They are 30 to 200 person exporters with a finance manager, a bank relationship and the capacity to complete a 20-page application. They are exactly the firms that the loan and grant measures are built for, and they are exactly the firms that do not need the NT$100,000 training voucher.
Meanwhile the 1.4 million firms with fewer than ten employees, the barbers, the repair shops, the two-truck freight operators, the breakfast shops, are the population the programme’s name invokes and the population least able to use it. Their constraint is not a 2.22% interest rate. It is that nobody has time to fill in the form.
The threshold structure quietly decides this. A programme built around firms with fewer than 30 employees sounds inclusive until you notice what it excludes and what it assumes. It assumes a firm with a payroll function, a bank account used for business, three years of filed statements and someone who can spend two days assembling documents. Above 30 employees, a firm almost always has all four. Below five, it almost never does, and it is still eligible on paper.
That gap is where the 1.4 million live, and it is also where the marginal social return on digitisation is probably highest, because a four-person firm that replaces a paper ledger with a shared spreadsheet gains more proportionally than a fifty-person firm that installs an ERP system it half uses. The programme’s instruments are designed for the middle of the distribution and its rhetoric is aimed at the bottom. That mismatch is not unique to Taiwan, and it is not solved by raising the subsidy ceiling.
What would make this auditable
The sceptical position is not that the programme is failing. It is that the public cannot currently tell, and that the reporting format is designed in a way that makes telling harder.
Three changes would fix it, and none of them requires new money.
- Publish the assisted-firm count broken down by intensity: advisory only, diagnostic completed, subsidy disbursed, tool deployed. Four columns would end most of the argument.
- Publish an audited comparison of revenue, payroll and tax for assisted firms against matched unassisted firms, using the tax and labour data the government already holds.
- Report credit guarantees and cash grants in separate tables with separate totals, so that a reader cannot add NT$257 billion of lending to NT$11.6 billion of spending.
The third change is the cheapest and the most revealing, because the material already exists. Credit guarantee records are held by the fund, grant disbursements by the ministry, and both are keyed to the same business registration number. A single two-page appendix would let any journalist, auditor or opposition legislator check the arithmetic without filing a freedom of information request. The absence of that appendix, in a country with Taiwan’s administrative data quality, is a choice rather than a limitation.
There is also a timing question. The tariff package opened for applications on 7 August 2025 and had reached 1,400 firms with cash grants by mid-October 2026, a fourteen-month window in which the average grant was about NT$4.2 million per firm. The digital programme spent the same period handing out vouchers of NT$100,000. Both are defensible on their own terms. Together they describe a government that knows how to move money quickly to firms that can absorb it, and has not yet decided whether the smallest firms are a target or an aspiration.
Ask for the disbursement line
The programme will probably hit its 600,000-firm target, because the definition of assistance is broad enough to make the target reachable. That is not corruption. It is what happens when a policy is measured by the only number that is cheap to collect.
The number that matters is smaller and harder: how many of those firms changed a process and kept the change for two years. If the answer turns out to be 50,000, that is not a failure. It is a real result, at a real cost of about NT$232,000 per firm, and it would be worth defending in public. The current reporting makes that defence impossible, which is a strange way to run a programme that is, on the evidence available, doing something useful.
So the question to put to the next progress report is not how many firms were helped. It is what the disbursement line says, in four columns, against a control group. A government that can produce 300,000 as a headline in twenty-one months can produce that table in a fortnight. Until it does, the most accurate description of the programme is the one its own documents support: a large, well-funded effort whose results are probably real and currently unverifiable, in a country that exports accounting standards to the rest of the region.