AI Voucher Support Program
AI Voucher Program
What is the AI Voucher Support Program?
- Government vouchers to cover SME AI adoption costs
- Consortium structure matching demand companies with supplier companies
- Success hinges on problem definition, data readiness, and business-unit involvement
What the AI Voucher actually changes is not the budget size but the sequence of decision-making. SMEs usually start from 'let's try AI,' get a quote, and stall at budget approval. The voucher makes them first write down, in a single application, current process figures, the form and volume of the data they hold, and their improvement targets. A vague task in the executives' minds becomes 'what to improve and by how much,' and whether or not the company is selected, that document remains the basis for later internal investment decisions. One more point: the voucher is a designated payment method, not cash. No grant money lands in your bank account. It can only be spent on fees for registered supplier companies, and any remaining balance cannot be redirected to other uses.
Projects are split into tracks by size, and funding limits and self-funding ratios are adjusted in each year’s announcement. In every track, part of the total project cost is self-funded by the demand company in cash and in kind; in-kind counts as labor cost for participating staff, but the cash share really leaves the company’s bank account. Boundaries with adjacent programs should also be sorted out before applying. The Data Voucher is for buying and processing data for training, while the Ministry of SMEs and Startups’ Smart Factory program centers on equipment integration and MES build-out. Where to apply depends on whether your project is 'stuck for lack of data' or 'stuck for lack of a solution'.
The most common misconception is that 'the government builds an AI system for us.' What the voucher pays for is the supplier’s development and consulting fees; the company covers its own data cleanup, labeling review, and the time business-unit staff spend in meetings. A riskier pattern is putting your company name on an application the supplier wrote. You may get selected, but the deliverable is someone else’s template that doesn’t fit your processes, and the moment the project ends and the license expires, you can’t reuse the models and screens built so far. Unless the contract fixes deliverable ownership, post-project maintenance rates, and retraining cycles, a year later you’re left with a dashboard nobody opens.
John KimInstructor John Kim, author of "B2B Marketing Design" and CEO of Nanum Consulting
Seungtae ParkAX instructor who teaches by building AI business development and marketing automation himselfA program run by the Ministry of Science and ICT since 2020 to foster the AI industry and support SME digital transformation. The National IT Industry Promotion Agency (NIPA) matches buyer and supplier companies and handles voucher settlement. Together with the Data Voucher and Cloud Voucher, it is one of the three main vouchers supporting digital transformation.
Consider a hypothetical case: Company A, a side-dish and ready-meal manufacturer with 118 employees and annual revenue of about KRW 50 billion (≈ US$37,408,300). It carries 340 SKUs, and because shelf life is short, its waste rate ran around 7%. Meanwhile, stockouts on its best-selling items led to about 30 order cancellations a month. The idea of using AI for demand forecasting had come up for two years, but each time they saw the KRW 180 million (≈ US$134,700) quote, they shelved it.
- The production manager had dug through the ERP starting two months before the call for proposals and pulled three years of shipment and return data. When it showed that near-expiry disposal cost KRW 420 million (≈ US$314,200) a year, the CEO concluded, 'Let’s make this a project.'
- Of the three suppliers we met, one CEO said 'Our model gets 95% accuracy,' but when asked about the 340 SKUs and the new-product turnover cycle, had no answer. We dropped them and formed a consortium with Company B, which had 2 similar food-industry references.
- The application laid out a plan to add weather and promotion history to three years of shipment data, with a target of 'disposal rate from 7% to 4%,' and the finance director had already set aside a separate budget for KRW 20 million (≈ US$15,000) in data-cleanup labor that the voucher doesn’t cover.
- One month in, a problem surfaced: 20% of return data was entered by hand, so item codes didn’t match. To keep business teams engaged, the first 30 minutes of the sales team’s weekly meeting were fixed as a 'forecast review' slot.
- Six months later, the waste rate fell from 7.1% to 4.3%, saving KRW 170 million (≈ US$127,200) a year, and stock-out cancellations dropped from 30 to 11 a month. Even after the program ended, the company keeps it running with a KRW 12 million (≈ US$8,980) annual maintenance contract and one designated in-house forecasting owner.