Loading page…
A Korean ERP vendor killed its own hosted business at 160 customers, spent two years rebuilding as genuinely multi-tenant, and enforced one rule: if a decision needs a human hand, don't make it
₩100,000/mo example list price (Korean won) · Azure bill cut from ~₩9M to ~₩3M a month once elastic scaling worked · ~₩100M/year paid to Microsoft
"Any decision that puts a human hand in the loop — don't make it." A prospect asking ₩90,000 instead of the ₩100,000 monthly fee is not a pricing question, it is a manual process you run forever. They took one in 2014, and he is still the hardest account they have — and he will not leave even when told to cancel.
Fewer bars = easier, cheaper, or faster for an AI-assisted solo builder. Editorial judgments based on the case details.
In 2000, Younglimwon Soft Lab launched an ASP business — hosted software billed monthly, before anyone called it cloud. Customers connected to a solution running in an internet data centre and paid a monthly fee instead of a licence. It sold to 160 customers and then the company stopped selling it. The reason is the most valuable sentence in the talk. There was no licence revenue, but the work was still being done the on-premise way, per customer, by hand. "The more of these we sold, the more the company would fail." So they cut it off at 160 rather than scale a model that lost money on every additional unit. What followed was a rebuild rather than a migration. Professor Park Jun-seong drilled three rules into the team leads, and the speaker — who runs the cloud SaaS business at Younglimwon — lists them in order: it has to be multi-tenant, single code and single instance; it has to be self-service, with the customer's need to ask you anything driven as close to zero as possible; and provisioning has to be automated. They spent two years redesigning from scratch, including redefining who the target was and what the service would be, instead of lifting the existing on-premise product into a data centre. The infrastructure decision came in 2011, when they compared providers and chose Microsoft Azure, specifically for elastic scaling and because, as the speaker puts it, SaaS should run on PaaS — rent infrastructure and build the environment yourself and every ongoing platform advance becomes your problem to chase. The result at the time of the talk, in December 2019: the cloud ERP runs about 120 customers across three Microsoft regions, sold domestically through partners, with provisioning, metering, invoicing and collection automated end to end. The third thing he wanted to say is that Korea's domestic SaaS market has a ceiling, and that the business has to go global.
For a ₩100,000-a-month cloud ERP, the first customers are not won by campaigns; they are won by removing every reason a buyer would need to phone you.
The quick-reference card — acquisition channels, replication playbook, risks & traps, and our verdict — is members-only.
Join the Unicorn Club — $5/moCancel anytime, in one clickPaste into Claude Code / Codex and get a working version of this product end-to-end.
Two production-grade prompts per idea — a full build spec and an SEO growth plan — plus 10 tailored build specs a week. Members only.
Join the Unicorn Club — $5/moCancel anytime, in one clickEvery claim traces back to a listed source — open any of them and check the number yourself.
Data credibility: 🗣 Founder-Reported, with a caveat on the label: the speaker is the executive who runs the cloud SaaS business at Younglimwon Soft Lab, not the company's founder, and he is presenting his own operation at a Talk IT industry session published on 2019-12-10 alongside Professor Park Jun-seong. Every figure is first-party and unaudited: 160 ASP customers before shutdown, about 120 live SaaS customers, the ₩100,000 monthly fee used as an example in a negotiation anecdote, the Azure bill falling from about ₩9 million to about ₩3 million a month, roughly ₩100 million a year paid to Microsoft, ₩25 million a year for the marketing tool, six VM sets, three regions, and roughly twenty minutes from signup to a provisioning confirmation email. No revenue, no margin, no headcount and no Japanese customer count were disclosed. All amounts are in Korean won as stated — the talk gives no USD conversion and none is invented here. Two names are damaged by the auto-caption: the product name is truncated to a fragment with the portal given as "시스템 .co.kr", which we render as SystemEver, and the marketing SaaS vendor's name is unrecoverable, so only its cost, billing terms and Australian data centre are reported. The "last July" and "that September" references are relative to a December 2019 talk. Because the talk is from 2019, treat every number as a snapshot of that year rather than of today.
The money arrives as a bank auto-debit on a monthly cycle, and the policy around it is more instructive than the amount.
The flywheel here compounds on cost rather than on demand.
The no-feedback stretch was two years long and it happened twice over.
SystemEver is Younglimwon Soft Lab's cloud ERP, sold as a monthly subscription and reachable through the company's own service portal. The architecture is the product decision, and the talk states it as three non-negotiables drilled in from the outside: multi-tenant with a single code base and a single instance; self-service, meaning the customer's need to contact you for signup, use or maintenance is pushed as near to zero as it can go; and automated provisioning. Around that sits a surprisingly complete self-service surface. The portal introduces the service and lets a prospect build their own quote. Applying triggers automatic provisioning and, about twenty minutes later, an email telling the customer the system is installed. After that come metering of usage, tax invoice issuance, billing and collection — and the speaker is specific that they did not program those integrations themselves; they connected the cloud provider's existing APIs. The client side solves a problem most ERP vendors handle badly. One program is developed and four sources are generated from it at the same time: a PC install build, a web browser build, an Android build and an iOS build. A patch or an upgrade lands on all four at once. He says that took roughly two years of hard work and that the professor had pushed for it heavily. There is also an after-sales system they built themselves, covering everything from a service request to feedback, and a separately built content system that turns the company's implementation know-how into self-service material — knitted together with the ERP and the portal so a customer experiences one system. A free trial site sits in front of the whole thing.
A monthly subscription, priced from six factors rather than from cost: the number of program modules, the frequency of service requests expected, a weighting, strategic points, competitors' prices and the market price. The example fee he uses when describing a negotiation is ₩100,000 per month. Three policy decisions sit on top of it, and each exists to avoid manual work. Billing is monthly only, with annual prepayment refused, because Korea's cloud development act requires a mid-term cancellation on a prepaid year to be pro-rated and refunded — and he says there are disputes in that calculation, so they simply do not create the situation. Collection is bank auto-debit alone; every other payment method was dropped. And discounting is not available, which is the pointed part: the company's rule is that any decision requiring human hands should not be made at all, and the stated posture to customers is that they can use the product as it is, or not use it. The one he openly regrets is grandfathering. When prices change, existing customers stay on the unit price from their contract date rather than moving to the new price, as most cloud services do. He would change it, and does not, because of the customers already on the old terms. Domestically the business is sold with partners, and the portal carries a partner management area alongside the internal and system administration areas.
Microsoft Azure, chosen in 2011 after comparing providers, for two reasons he names: elastic scalability, and the belief that SaaS should run on PaaS — if you rent raw infrastructure and assemble the environment yourself, every continuing platform advance becomes your responsibility to chase, and a small vendor cannot keep up. The topology has one honest compromise in it. The web tier started out cloud-based but the integrations they needed were constrained — "there are one or two things they won't do for us" — so they moved to VMs, and they now run six VM sets that scale up and down. The database tier is PaaS: Azure SQL Database. They use three Microsoft regions, Hong Kong at the start, then Japan for the Japanese business and Korea for the Korean one. A systems engineer watches the three region dashboards side by side; monitoring alerts reach him and the team by email and mobile. There is no 24-hour monitoring rota, because the business is currently Asia-only. The cost numbers are the most useful thing in the talk. Before July of the previous year the Azure bill was around ₩9 million a month. Around that September, once elastic scaling was fully working, it dropped to about ₩3 million — from ₩8-9 million down to ₩3 million for the same service. Annually he describes paying Microsoft on the order of ₩100 million, recently reduced. Separately, marketing runs on a SaaS tool costing ₩25 million a year, billed annually in advance with no mid-term refund, hosted in Australia; the auto-caption garbles its name. No revenue figure is given anywhere, so margin cannot be computed. Our read: with roughly 120 customers and a ₩3 million monthly infrastructure bill, the infrastructure line is no longer what determines whether this business works.
Copy the shutdown decision first. A hosted business that grows revenue while losing money per unit is not an early-stage problem to grow out of; it is a model that gets worse with scale, and the correct response is the one taken here — stop selling at 160 and rebuild. Most companies in that position sell the 400th customer and then explain the losses as investment. Copy the automation rule, which is the sharpest operating principle in this whole batch: if a decision requires a human hand, do not make it. That rule is what converts into monthly-only billing, bank auto-debit only, no discounts, self-service quoting, and twenty-minute automated provisioning. It reads as rigidity and it is actually an economic choice, because every exception granted is a cost that recurs monthly forever while the revenue from it is a rounding error. Copy the build-once-generate-many approach if you ship to several platforms, and copy the decision to wire metering, invoicing and collection to your cloud provider's existing APIs rather than writing that plumbing yourself. What to change is the grandfathering policy, which he flags himself as his one regret. Locking every customer at their contract-date unit price means your pricing can never be corrected and every legacy account permanently constrains the new ones. Our read: the part that does not transfer is the starting position. This worked because an established ERP vendor already had the domain knowledge, the partner channel and the balance sheet to fund two years of rebuilding with nothing coming back. A solo founder copying the automation doctrine into a new product will get most of the benefit; copying the two-year rebuild will simply run out of money.
The ceiling he raises himself is the biggest one: there is a limit to running a SaaS business inside Korea alone, so it has to go global, and the talk is about preparing for that rather than having done it. A ₩100,000-a-month product needs a very large number of customers to matter, and about 120 of them is a business that works but has not yet proven it can cross a border — the Japan region exists, but no Japanese customer count is given. Single-vendor dependency is the second. Provisioning, metering, tax invoicing, billing, collection and monitoring all run on one cloud provider's APIs by design. That is exactly why the cost fell from ₩9 million to ₩3 million when the provider finished a feature, and it is also why a pricing change or a deprecation there lands directly on this business with no alternative wired up. Third, the grandfathering policy compounds quietly. Every customer priced at their contract date is a permanent constraint, and a decade of that makes repricing practically impossible — which is precisely why he calls it his regret. Fourth, the no-exceptions posture has a visible casualty. The one customer they discounted in 2014 is described as the hardest account he has, one who will not leave even when invited to cancel. That is a small failure inside a policy that mostly works, but it shows what a single exception costs over five years. Our read on evidence: this is a company executive presenting at an industry session, describing his own operation. The numbers are first-party and unaudited, no revenue was disclosed at all, and everything is as of December 2019.
The four founder playbooks — first customers, first dollar, the flywheel, the silent stretch — are in the Unicorn Club, on this idea and on all 1,435 others. The story, the numbers, the sources and every later section stay open to everyone.
Join the Unicorn Club — $5/moCancel anytime, in one click