Operational web modules
A five-step customs declaration replacing a shared spreadsheet. Fuel depot management across three tanks. Contract management across four legal entities. Payment requests with multi-step approval chains.
AI systems for transport & logistics operators
One route, reported as a 1.4bn loss. Actually: 1.9bn profit.
One mistyped freight rate reversed the result of an entire route. Nobody cheated and nobody was lazy - the number simply sat somewhere no one was cross-checking. We build AI that runs inside your operation, so errors like that surface before they turn into decisions.
30 minutes · no charge · nothing to prepare
One route, one monthJune
As reported
−1.4bn
The system compared the rate in the report against the rate in the contract and found a single mistyped line.
After reconciliation
+1.9bn
Swing on a single route3.3bn VND
5 months
from the first working session to systems running in production
4 modules
used daily across six different departments
~60 USD
monthly AI infrastructure cost for all of it
1 person
built the whole thing - and they do not work in tech
The problem
You are not short of data. It is just scattered.
Accounting's spreadsheet, dispatch's chat group, paper dockets at the depot, declarations in the customs software, rates in an email thread, a supervisor's notebook. Each is right about its own piece. None is right about the whole picture - and the whole picture is what management has to decide on.
Five questions most boards cannot answer today:
01Which routes are actually profitable?
02Which costs are running over budget?
03Where exactly is the discrepancy, right now?
04Is there leakage nobody has noticed?
05When do I personally need to step in?
What we build
Not a chatbot. Systems that are actually in production.
This is the real inventory at a transport & logistics operator we work with, after five months. Every item has real users in a real department.
An operating dashboard for the board
Volume, receivables by age, fleet operations, per-vehicle spend, and a what-if table for choosing routes when fuel or freight prices move.
Automatic alerts over chat
Weekly collections reports, repair requests raised within two minutes, and balance warnings on fuel and toll accounts before a truck gets stuck.
Automated data pipelines
Hourly toll transactions, fuel invoices, multi-year GPS history for regulators, and per-order distance.
Your own business rules, written down
How volume is counted, how revenue is recognised, who approves what, fuel allowances. Things that used to live only in a few people's heads.
Documents and internal policy
Internal regulations written to statute, a payroll policy fit to file with the authorities, bilingual contract redlines, formal replies to customs.
Evidence
What the system saw that nobody had been able to see
These are real discrepancies found at one transport & logistics operator over five months. Every line carries an honest label: we do not merge "found" with "recovered". Figures are in Vietnamese dong, roughly 25,000 to the US dollar - so 1 billion VND is about 40,000 USD.
| Finding | Figure | Status |
|---|---|---|
| Toll transactions passed through the gates with no invoice collected | ~500m VND | Opportunity |
| A major customer over its credit limit, found in the very first report | 1.56bn VND | Opportunity |
| Internal fuel trading showing phantom profit from a mis-stated cost basis | 740m → ~138m | Happened |
| Tank gauge calibrated in millimetres but read in centimetres | one tank 3,755 L short | Happened |
| One month of fuel issues reconciled between the module and operations | 7 wrong plates, 4 missing dockets | Happened |
| Vehicles that should be overhauled or sold instead of carried at a loss | 25 vehicles | Opportunity |
| Fleet fuel consumption over allowance, consistently, every month | +18–34% / month | Opportunity |
| Six plates from sold vehicles reused on new ones, nearly written off twice | 6 vehicles | Happened |
| Two-driver orders losing money on every single job | ~7m VND / order | Happened |
| Operational revenue reconciled against the accounting ledger, by business line | 89–103% | Opportunity |
| Empty running on the return leg, all year round | 32–45% | Estimate |
Two billion VND is sitting at "seen, not yet recovered".
We say that plainly instead of folding it into an ROI figure. The first two rows alone are roughly 2 billion VND of real money the system pointed at - but collecting it is accounting's job, and nobody has confirmed how much came back. A software vendor would write "saved 2 billion". We write "found 2 billion, still to be collected". You should know that difference before you sign anything.
Five months
What we leave behind is not reports. It is the ability to build.
Everything above was built by an internal-control officer with no technical training. In April, anything hard got handed to an engineer. By July they were asking the system directly whether it could finish the job without calling one. By September they were running their own eight-item financial-optimisation roadmap and treating AI as the implementer.
Working sessions with AI, by month
The break comes at the end of June, when the real problem got framed: "automate the internal process, usable by someone who is not technical". Usage rose sevenfold and stayed there. *September counted to the 17th.
Apr – MayAI as an analystData in, analysis out. A pipeline pulling files automatically at 8:30 every morning. The output was still spreadsheets and documents.
Jun – JulAI running the reportingReports delivered into chat each morning. Then a web dashboard on hourly data, with per-person access control.
Jul – SepThe user designs, AI implementsThe process spec is written in-house, AI builds the module. Four modules shipped in two months. Domain records, user permissions and AI spend were all handled without us.
Every decision you sign today rests on a number. The only question is who checked it.
In the engagement above, the answer used to be: nobody.
Why it is different
Off-the-shelf software makes you change how you work. This runs on the process you already have.
01
It fills the holes in your data
Drivers and depot staff were never going to become diligent data-entry clerks - that is why the data always has holes. The system collects from several sources, cross-checks them, and fills the gaps with machine data instead of waiting for someone to type.
Demonstrated
Found uninvoiced spend, over-limit receivables and stock discrepancies without anyone entering an extra line.
02
It catches things before the money is gone
Access is tight by role, by legal entity and by border gate: a declarant sees only their own gate, the fuel depot never sees commercial reports. And all the data stays inside your systems, not ours.
Demonstrated
Fuel and toll balance warnings land in chat before a truck is stranded at a gate. A hole that let staff see orders from other border gates was also found and closed.
03
It is cut to the process you already have
We do not make the business change how it works to fit the software. Business rules are approved by management and then written down as company property - not knowledge trapped in one or two heads.
Demonstrated
300 business rules recorded and in force, each with the date and the person who confirmed it.
How we work
Three steps, and the last one is us stepping back
The goal is not for you to depend on us. It is for your team to run it, while we keep the layer underneath current.
Step 01
Survey
You bring the bottlenecks. We tell you which are solvable now, which need data you do not yet have, and which are not worth touching. No charge for this step.
Step 02
Deploy
The AI agent is installed inside your systems, on your data. Every business rule is approved by management before it takes effect.
Step 03
Operate
Handed over for your team to run. We refresh the model layer monthly, so the system gets stronger over time instead of ageing.
Cost
The expensive part of AI is not the AI
This is the real infrastructure figure for the system described above, after five months of continuous operation. The build itself is quoted separately, against real scope, after the survey.
AI infrastructure, per month
~60 USD
Around 1.5m VND, covering every module, dashboard, alert stream and data pipeline in production.
Versus traditional software
−70%
Build cost compared with conventional software development. Our own figure, not independently audited.
A division, not a promise
A year of that infrastructure costs about 18 million dong. A single line in the evidence table above - the toll charges that passed the gates with no invoice collected - is roughly 28 times that figure. The over-limit receivable is roughly 87 times it.
We still will not call this ROI. That is money seen, not money recovered, and the division leaves out the cost of the build.
The honest part
What did not work
An AI supplier who only tells you the wins is hiding the more useful half. These are the things that failed inside the very project above, taken straight from the internal closing report.
Cancelled
OCR on handwritten dockets
Ran three weeks, then stopped. Misread 10 of 28 images, wrong dates, invented plate numbers. The lesson: photographed handwriting is not yet good enough for bulk automation.
Unfinished
Classifying 13,000 product names
Quality never got there, so it was dropped. The same problem was later solved more cheaply with a self-learning catalogue inside the customs module.
Abandoned
Automatic link to the accounting package
Several approaches tried, all dropped. Replaced by something simpler: the user uploads the ledger, the system reads it from the data warehouse.
Still open
Automation running on a personal laptop
If the machine is off, the report does not go out. It happened repeatedly in the first two weeks. Being moved onto a company server.
Counter-intuitive
The live dashboard read worse than the static one
The user said it plainly: the static analysis showed the problem faster than the live version. More interactivity is not the same as more understanding.
A real cost
Accuracy always needed a human check
Plenty of figures were wrong at first and only came right after the user taught the system the referencing rules. That is genuine effort on the client side, and we have not measured it.
Common questions
The eight we get asked most
Our data is a mess. Do we have to clean it up first?
No, and that is not how it works anyway. It is precisely because the data is messy that automatic reconciliation finds things - the tank gauge in the wrong unit, the duplicated plates, the mistyped rate all surfaced the moment two sources were put side by side. If the data were already clean, most of the value would be gone before we arrived.
Does our data leave the company?
No. The system runs inside your accounts and your infrastructure, with access split by role, by legal entity and by border gate. In the engagement described above, a declarant sees only their own gate, the fuel depot never sees commercial reports, and the dashboard sits behind the company's own login.
We have no IT staff. Who runs it?
Your people. Everything in the examples above was built and is operated by one internal-control officer with no technical training and no programming. Our job is to build a foundation someone like that can work on, and then step back.
How long before we see anything?
The first financial finding in that engagement - a major customer 1.56bn over its credit limit - came out of the very first receivables report, in month one. The first web module was in production after about three months. The pace depends on where your data currently sits and who is allowed to open it.
What happens if we stop working with you?
The code, the data and the business rules live in your accounts, not ours. There is no key and no licence to renew to keep the system running. You also keep the rules that were written down, which is the hardest part to rebuild.
Does the AI make decisions for us?
No. It reconciles, finds and warns; people decide. Every business rule - how volume is counted, who approves which spend, what the allowance is - is approved by management before it takes effect, and recorded with the date and the name of whoever confirmed it.
We are small. Is this for us?
Possibly not, and we will say so at the survey. If the whole operation still fits in one spreadsheet run by one person who holds it all in their head, you do not need us yet. The value starts once the data has spread across several places and nobody can see the whole picture any more.
What does it cost?
The AI infrastructure for the system described above runs at about 60 USD a month. The build is quoted separately against real scope after the survey - we do not quote before we know what condition your data is in. The survey itself is free.
Who you deal with
Who you would actually be signing with
Wildcats AI Studio is the shared brand of two entities: VentureWild in Da Nang does the delivery, and Wildcats Global in the United States holds the brand and the network. Vietnamese clients contract and invoice with VentureWild.
The contracting entity
CONG TY TNHH VENTURE WILD
Tax ID0402350798
Registered officeFloor 8, Bach Dang Complex, 50 Bach Dang, Hai Chau, Da Nang
InvoicingVietnamese e-invoice, per current regulation
Brand and network
Wildcats Global
Registered inDelaware, United States
RoleShared brand, community and international clients
Vietnamese clientsStill contract and pay VentureWild
The tax ID is publicly verifiable on the Vietnamese tax authority portal.
Next step
Thirty minutes, and you will know whether there is anything worth doing
The first survey is free and needs no preparation from you. You describe where the process jams; we tell you straight which parts AI can solve in a few weeks, which it cannot, and which you should just do yourselves without us.
Tuấn Anh Nguyễn
Co-Founder · Wildcats AI Studio