Building a B2B SaaS for Vietnamese online sellers is one problem. Measuring whether it is going in the right direction is a different one. The first nine months of Ordinex made both problems visible: things that worked better than we expected, and things we had to tear down and rebuild.
Why we are writing this
We are not interested in building-in-public posts that exist to share clean metrics. This one is useful to us because looking back helps us understand where we actually are, and it is useful to Scout users because they deserve to know how the product they are using gets built.
This period runs from when we started building seriously in mid-2025 through early January 2026. It was not nine months of clean progress.
The real numbers
Some numbers we can share from this period:
- Closed beta opened September 2025. We did not rush to scale it. The first cohort was around twenty users, all active sellers importing from 1688 and reselling on TikTok Shop or Shopee.
- User conversations in nine months: over 60. Not surveys. Actual conversations about daily workflows, pain points, and where the product fell short.
- Month-one retention for the beta group: above 70 percent. For an early-stage B2B tool, we treat this as enough of a signal to keep going.
- Features built and discarded: more than the features still in the product. Not something to brag about, and not something to be ashamed of either. Just accurate.
These numbers do not make us look like a rocketship. They are true, and true matters more than impressive right now.
What worked better than expected
The decision to focus on 1688 and Vietnamese sellers was right. We get asked occasionally: why not expand to Alibaba, Lazada Malaysia, or other markets? The answer is that expanding at this stage means doing nothing well. Vietnamese sellers importing from 1688 have enough specific problems to keep us busy for years. A narrow scope let us build faster and understand more deeply.
Putting real landed cost into the filtering step was the right product call. Most sourcing tools stop at "here is the price on the platform." Scout adds domestic China shipping to the consolidation warehouse, freight to Vietnam, customs and clearance fees, platform commissions, and shrinkage before surfacing results. More than a few users told us this is the reason they stayed on Scout rather than doing the math manually. The calculation sounds simple but has real layers, and it took us a few months to get the Margin Engine to a number reliable enough to trust.
Keeping the beta small was the right call. We wrote separately about this. The short version: opening wide early dilutes feedback and makes it impossible to know who has a real problem. A small group, where we know every user by name, means we hear true stories from actual sellers.
What did not go as planned
Getting reliable 1688 data took longer than expected. Product data from 1688 is not clean. Prices change by the day, the same item sold by different suppliers can vary widely in price, and some categories have inconsistent data structures. We expected three months to get a stable pipeline. It took closer to six. This is a technical debt that cannot be ignored.
Onboarding took more time than we wanted. Vietnamese sellers are used to their own workflows. Even when Scout solves a real problem, changing an existing habit takes time. Some beta users dropped off not because Scout was broken but because they were not ready to change how they operated. That is a behavior-change problem, not a feature problem.
We built features before we understood the real workflow. In the first two months, several things got built based on assumptions rather than observation. Most of those are gone from the product now. The specific lesson: do not build based on what users say they want when you first meet them. Watch what they actually do after two weeks of real use.
The defect detection loop was too slow early on. With only twenty people in beta, each bug touched fewer users. But it touched exactly the users we needed most. One instance of a miscalculated landed cost that ran for two days caused more damage to trust than we anticipated. Since then we added automated verification layers for the Margin Engine before any change ships.
What we learned about product and users
What nine months taught us is not mostly technical. Most of it is about reading signals from real users.
Sellers care about outcomes, not features. Nobody told us "I want a margin filter." They said "I finish importing and I have no idea whether I made money until the stock actually sells." That is the real problem. The feature is just one way to address it.
Good data has clear value, but only when paired with a next action. Sellers do not just want to know what a product's margin is. They want to know whether to import it or not. The step after each number matters more than the number itself. This shaped how we designed the interface in Margin Engine v0.2.
The import cycle of Vietnamese sellers is longer than a typical SaaS usage pattern. The average Scout user imports every three to four weeks, which is roughly one capital cycle. That means Scout's value does not show up daily the way an email tool or CRM does. It shows up each time they are about to place a new order. Designing for that cycle is different from designing for daily-use software.
Where we go next
At the start of 2026, we have three things to do in order.
First, finish the Margin Engine to price more accurately for categories with unusual cost structures: bulky goods, fragile items, products requiring certification. Margin Engine v0.2 shipped in February 2026 with meaningful improvements to how we compute volumetric freight.
Second, batch import into Scout instead of adding products one at a time. Batch import shipped in March 2026, so sellers can push a whole watchlist in one step rather than adding SKUs manually.
Third, open Orders beta for managing 1688 purchase orders inside the same flow as product filtering. The goal is not to replace a seller's ordering system. It is to close the gap between "this product hits margin" and "place the order." Those two steps should not sit in separate tools. Orders beta is on track for April 2026.
We have no plans to expand to additional markets or additional sourcing platforms in 2026. The list above is enough to do well.
Bottom line
Nine months taught us something more important than any feature: understanding the seller's problem correctly is harder than building a solution to it. And understanding it correctly only comes from real conversations, not from analytics data or assumptions made at a desk.
The next chapter for Ordinex is still the same core task: help Vietnamese sellers know which products are worth importing, and at what quantity, so their capital does not sit in a warehouse going nowhere. When that works well, the rest follows.