A small factory used AI to build its own ERP in one month: Build or Buy, has the balance really shifted?

On November 16, Forbes Councils published a rather unassuming entrepreneur's first-person account. The author, Vanessa Yiannakou, runs a small factory that makes custom blinds. She said that, using open-source software plus an AI model as her development partner, she replaced the company's original accounting and ERP systems in less than a month, and also migrated more than two years of historical accounting records.

The results she listed were very concrete: she built a quotation tool for custom products, implemented production tracking on the shop floor, and customers can follow the entire process from inquiry all the way to final payment. Quoting that used to take at least ten days at the fastest now produces results almost instantly.

This is a company’s own claim, not an industry benchmark. But for the first time, it brings a question we have repeatedly discussed over the past six months down to a concrete small factory: when AI pushes the cost of writing code down to near zero, has the answer to the multiple-choice question of “buy an ERP system or build one yourself” really changed?

1. What Exactly Was Built in One Month

First, let's clarify the scope of this matter to avoid being misled by the headline 'building an ERP in one month.' What Yiannakou built is not a general system covering the whole company and capable of replacing SAP, but a custom system grown around its own processes—with three core parts: quoting, workshop production tracking, and customer progress visibility. The accounting part involved replacement plus historical data migration, and every line was reconciled against the old system.

< 1 month

The cycle from project initiation to replacing the existing ERP (per the company's own account)

The truly valuable information here lies in the two actions she emphasized: migrated accounts line by line (checking migrated accounts line by line), and testing the tool first with dummy orders before putting it in front of customers. In other words, what AI saves is the time spent coding, not the reconciliation and testing. The most accurate way to put it is with the line from the report: "Faster development won't make checking any less important; it will only make the consequences of skipping checks arrive sooner."

AI has lowered the barrier to building software, but it comes with no business knowledge and no quality checks. Handing raw business material straight to a large model—whether to write code or to make decisions—is to hide the cost of verification somewhere out of sight.

What requires the greatest vigilance is "if it runs, go live." Yiannakou's repeated emphasis on line-by-line reconciliation and running virtual orders first precisely shows that the core accounts cannot withstand "roughly right." The time saved by skipping validation will be reclaimed with interest at the first month-end closing or tax filing—and by then there will be no old system to fall back on.

II. An open-source foundation is the prerequisite for this to be possible

The fact that this small factory could get it done within a month rests on an often-overlooked precondition: it stood on the shoulders of open-source ERP. Look at the activity data for open-source commercial software as of November 20, and you'll understand why "building it yourself" suddenly became feasible in the past two years:

ProjectPermissionGitHub StarsSubmitted in the last 12 months定位
OdooLGPL-3.055,97617,381Full-module ERP suite
ERPNextGPL-3.041,46713,131Full-featured ERP, v16.60.0 (November)
CarbonAGPL-3.02,8503,593Manufacturing ERP + MES + QMS
DolibarrGPL-3.07,88810,765Lightweight inventory management + invoicing

The data source is the GitHub statistics refreshed by appsgit on November 20. The key point is not the number of Stars itself, but the "number of commits in the past 12 months"—Odoo had 17,381 commits in a year, and ERPNext had 13,131, showing that the foundation is alive and continuously evolving. What AI helps you write is the upper 20% of customization; the lower 80%—the accounting engine, inventory logic, and permission model—has been accumulated by these communities over many years.

20%

AI-assisted customization section (quotes/tracking/views)

80%

The accounting/inventory/permissions core assumed by the open-source base

This also explains another set of seemingly contradictory data. In its November SME ERP evaluation, the German consulting firm much. Consulting scored nine horizontal ERP systems against 14 criteria, and Odoo was the only one that hit all of them (14/14). In the same evaluation, a group of "AI-native finance software" products that had raised huge amounts of funding—Rillet ($100 million Series C, $1 billion valuation), Campfire ($103.5 million), DualEntry ($90 million), and Everest ($140 million)—either did only finance or only operations; none could independently run a German manufacturer.

So the idea that "AI-native startups can replace ERP" is an illusion. These new companies solve a certain link (general ledger, closing, reconciliation), while manufacturing enterprises need finance, procurement, production, and inventory to run on the same data model. AI makes point tools sharper, but it has not made the "integrated core" dispensable—on the contrary, it has increased the importance of the foundation.

III. A decision framework for CIOs: when to build in-house and when not to touch it

To read this personal account as "anyone can build an ERP in a month" is a dangerous misreading. But to read it as "AI doesn't change build-vs-buy at all" is equally stubborn. The real judgment framework is the following three points.

First, use "core vs. add-on" to draw the boundary.Accounting general ledgers, inventory ledgers, permission models, closing logic—such core systems that require correctness, auditability, and the ability to withstand year-end closing should preferentially use mature open-source or commercial ERP; quotation configuration, workshop dashboards, customer progress views—such add-ons that grow closely with one's own processes and change frequently—are exactly the home ground for AI self-building. What that small factory got right was precisely using the knife in the right place.

Second, write “reconciliation and testing” into the delivery definition of done.For AI-generated code, the acceptance criteria cannot be “it runs,” but rather “historical data matches line by line, virtual orders can run through, and exception paths have been reviewed by someone.” Turn these three things into a checklist, so that AI efficiency gains do not turn into hidden landmines.

Third, evaluate the "right to maintain," not just the "right to develop." Building it in a month is not hard; what's hard is whether this system can still be changed three years later, when the product line has changed, people have left, and AI tools have been replaced. Before building in-house, answer first: Who will maintain it? Where is the documentation? Is the underlying layer an exportable open-source standard? If the answer depends on one person or one AI tool, that's not in-house building—it's creating a new lock-in for yourself.

Back to the question at the beginning: did the balance move? It moved, but it didn't tip over. AI has drastically lowered the cost on the "build" side, making tools that used to be "too small to be worth doing" worth doing; but the weight on the "verify, maintain, and take responsibility" side hasn't decreased at all. The real new opportunity for SMEs is not to replace ERP with AI, but to use AI + an open-source foundation to build, one by one, those peripheral needs that have always been queued up and never gotten budget. First hand the core over to mature systems and leave creativity to your own processes—this is what that small blinds factory is truly worth learning from.


This article is compiled and analyzed based on publicly available information: the Forbes Councils entrepreneur’s first-person account “Miles Of Progress: How AI Is Rewriting The Build-Versus-Buy Decision” (2026-11-16, author Vanessa Yiannakou), appsgit’s activity statistics for open-source commercial software (data refreshed on 2026-11-20), and much. Consulting’s “Which ERP is really the AI operating system for German SMEs” (2026-11). Business outcomes are all self-reported and have not been independently audited; assessment scores represent the consulting agency’s views. Data as of 2026-11-22.

Zhongdi Information · Focused on Odoo and digital solutions for enterprise management

关于我们

​我们致力于帮助中小企业实现数字化转型,我们的团队由一群充满激情和创新思维的专业人士组成,他们具备丰富的行业经验和技术专长。

扫一扫获取顾问以及手册

Sign in to leave a comment
Largest-ever IPO rings the bell: 81.3 billion or 93.5 billion? The AI supplier's books are laid bare before CIOs for the first time