Case study · Business, AI
One system. No ERP.
Kovaa is Qomero's B2B packaging and wood business, trading as KOVAA Packaging and PTU Wood. Four systems, built when needed.
Status: K-LionBook, K-AutoDesign, kovaa.vn: LIVEK-Ledger: R&D
Context
Kovaa is Qomero's own B2B packaging and wood business, selling through KOVAA Packaging and PTU Wood.
Problem
Sales ran on spreadsheets, paper receipts and Zalo. Reorders slipped and overdue debt surfaced late. Popular SaaS CRMs in Vietnam didn't follow the industry's sales pipeline.
Published scope
K-LionBook (CRM, live), K-AutoDesign (live), kovaa.vn (website), K-Ledger (R&D).
What we did
No ERP. One system per job, built in the order the work needed. AI only suggests, with a reason; the salesperson confirms.
Verified result
Overdue receivables −71%. Reorders and follow-ups caught +45%. Manual reporting −80%, work 3× faster. 15 salespeople use K-LionBook daily. (Kovaa internal figures)
Limits
Kovaa's internal figures. K-Ledger is in R&D; no result is claimed.
99 / 98
kovaa.vn Performance score, desktop / mobile
Source: PageSpeed Insights
15
people on K-LionBook
Source: Kovaa internal figures
−71%
overdue receivables
Source: Kovaa internal figures
+45%
reorders and follow-ups caught
Source: Kovaa internal figures
3×
faster sales work
Source: Kovaa internal figures
−80%
manual reporting
Source: Kovaa internal figures
Before the build
Kovaa ran on spreadsheets, paper receipts and Zalo messages.
The bookkeepers pieced together documents, two-way balances, stock books and cash flow from those sources.
Sales needed to see what a customer owed. Bookkeeping needed to see what the customer was promised. The two lived apart.
Reorders were missed, follow-ups dropped, overdue debt noticed late.
Why off-the-shelf failed
Kovaa chose not to buy an ERP. Each piece was built when the work needed it.
Kovaa looked at several SaaS CRMs popular in Vietnam. None followed the sales pipeline this industry runs on.
Even accounting stayed narrow: K-Ledger is not an ERP and does no tax accounting. It covers only the operational accounting Kovaa needs day to day.
What we decided
Most were business decisions before they were technical ones.
One job per system. No shared ERP. The website only helps customers find Kovaa; selling and packaging design live elsewhere.
The salesperson owns the data. AI in K-LionBook only rewords or suggests; the salesperson confirms. A task is created only when they click.
Every suggestion has a reason. Each carries its reason and evidence. AI may reword the reason, never invent one.
Each salesperson sees their own numbers. In sales health and the forecast, each person sees only their own row.
Done means the team uses it daily. A phase is finished only when the sales team uses it instead of the spreadsheet, on their own phones and laptops.
A quote belongs to one fixed design. In K-AutoDesign a quote is tied to a design version that cannot change. When AI is not enough, the request goes to a designer rather than more options, and design advice needs a person's approval.
Bad data is stopped at the source. In K-Ledger, business rules live in the database and the audit trail is append-only: nobody can edit it, administrators included. It is organised around contracts and follows Circular 133/2016/TT-BTC.
Speed is content. kovaa.vn was built to load fast on phones and written for Google (SEO) and AI search (GEO). Every site Qomero builds scores above 97 on PageSpeed.
Rebuilt after learning. A first K-Ledger in Django showed what was needed; the current version is rebuilt from it.
Architecture
Four systems, one job each.
Drawn from documented technologies; data flows are not inferred.
K-LionBook
The B2B sales team's CRM: customers, opportunities, contracts, orders, receivables, margin, suppliers and purchase orders in one place. Stack: Next.js, PostgreSQL, pgvector, Drizzle, Supabase.
K-AutoDesign
AI-assisted packaging design, from brief to production quote. Stack: Python and TypeScript, with a generative image model.
kovaa.vn
The website packaging buyers find Kovaa through. Stack: Next.js, Sanity CMS, Vercel.
K-Ledger
R&D. Contract-centred pre-accounting: documents, receivables, stock and cash. Stack: Python.
What the system handles
- A customer is due to reorder a SKU, has gone quiet, has an unanswered quote, an expiring contract or overdue debt: they appear on the attention list.
- For each customer, the system suggests the next step, with its reason and evidence.
- After a meeting, the salesperson writes a note in Vietnamese. AI suggests what to record; nothing is saved until they confirm.
- The forecast is split into committed, likely and potential.
- Follow-up tasks are created by named rules.
- A supplier order (PO2) is past its ready date, a customer order (PO1) risks late delivery, or a confirmed line has no supplier order: the system warns.
- A salesperson asks about the data and gets an answer with the records behind it.
- Ad reports are pulled from the ad platforms, read-only.
- A customer writes a brief that saves as they type, then gets AI design options. They request a quote tied to a design snapshot, or hand the brief to a designer when the options are not enough.
- (R&D) Each receivable and payable traces to its source invoice; receipts attach to line items; stock is read against working capital.
Result
99 / 98
kovaa.vn Performance score, desktop / mobile
Source: PageSpeed Insights
15
people on K-LionBook
Source: Kovaa internal figures
−71%
overdue receivables
Source: Kovaa internal figures
+45%
reorders and follow-ups caught
Source: Kovaa internal figures
3×
faster sales work
Source: Kovaa internal figures
−80%
manual reporting
Source: Kovaa internal figures
- Kovaa's B2B sales team uses K-LionBook every day.
- K-AutoDesign runs in production.
- K-Ledger is in R&D.
More case studies
ARVUMA
Buyers scattered across channels; no single order record.
Every channel's orders in one place.
Read the case study
Synolang
Learners couldn't see which criterion held them back.
AI on the repetitive work, honest when unsure.
Read the case study