The verdict in three sentences
A Singapore freight forwarder that re-keys 900 bills of lading and commercial invoices a month into its permit system spends about 180 hours on data entry and accumulates value, HS code and weight errors that cost penalties and storage days at PSA terminals. An AI extraction pipeline (reading PDFs and scans, extracting fields, consistency checks, human validation, export for TradeNet permits) is priced at 50 000 to 120 000 SGD by Singapore integrators in 2026 (market benchmark), while Kolonell's mid-complexity grid sits at 4 000 000 to 7 000 000 FCFA (6 100 to 10 670 EUR), delivered in 10 to 12 weeks. The realistic target is 95 % accuracy on key fields and 12 minutes saved per document, with validation by a declarant remaining mandatory.
Where time is lost in a customs file
The declarant receives documents by email or WhatsApp, often as skewed scanned PDFs, sometimes as photos. They copy shipper, consignee, container number, weights, description, value and currency, then look up the HS code in the Singapore Trade Classification. Every re-entry is a chance for error.
| Field or step (2026 estimate, Singapore forwarder) | Manual time | With AI extraction | Estimated manual error rate |
|---|---|---|---|
| Shipper and consignee identification | 2 min | 10 s | 1 to 2 % |
| Container and seal numbers | 2 min | 10 s | 2 to 3 % |
| Gross and net weight, package count | 2 min | 15 s | 2 to 4 % |
| Description, quantities, value, currency | 5 min | 1 min (check) | 3 to 5 % |
| HS code suggestion | 4 min | 1 min (validation) | 4 to 6 % |
| BL and invoice consistency check | 3 min | Automatic | Gaps often missed |
| Total per document | 18 min | 6 min |
AI does not replace the declaring agent's liability: it suggests, the declarant validates. Fields with low model confidence are highlighted and must be confirmed before export.
2026 budget and pipeline architecture
The pipeline combines OCR tuned for poor scans, an AI model that structures fields, business rules (weight per container, currency consistency, HS code list) and an export in the format your permit software needs for submission through the Networked Trade Platform.
| Item | 2026 range | Comment |
|---|---|---|
| Custom extraction pipeline, Singapore integrator (market benchmark) | 50 000 to 120 000 SGD | Similar scope, local rates |
| Custom extraction pipeline, Kolonell grid (mid complexity) | 4 000 000 to 7 000 000 FCFA (6 100 to 10 670 EUR) | OCR, extraction, validation, permit export |
| Multi-office or multi-country extension, Kolonell grid (complex) | 7 000 000 to 15 000 000+ FCFA (10 670 to 22 870+ EUR) | Several hubs and customs regimes |
| OCR and AI API for 900 three-page documents (estimate) | 180 to 350 SGD/month | Roughly 0.07 to 0.13 SGD per page |
| International customs SaaS (market benchmark) | 800 to 2 500 SGD/month | Generic, limited custom rules |
| Kolonell Standard maintenance | 75 000 FCFA/month (110 EUR) | Model tuning, new document formats |
| Annotation of 500 reference documents (internal effort) | 8 to 12 days of a declarant | Essential to reach 95 % |
Documents contain names, addresses and customer numbers: Singapore's PDPA applies, overseen by the PDPC. Keep files in a controlled environment and use an API contract with no data reuse. Note that the Productivity Solutions Grant only covers pre-approved local solutions.
10 to 12-week plan
| Week | Deliverable | Exit criterion |
|---|---|---|
| 1 to 2 | Inventory of formats (shipping lines, suppliers) | 20 BL and invoice templates listed |
| 3 to 5 | Field extraction and consistency rules | 90 % accuracy on the sample |
| 6 to 8 | Validation screen and permit export | Declarant validates a document in under 6 min |
| 9 to 10 | Pilot on one flow (container imports) | 95 % on key fields |
| 11 to 12 | Extension to other flows, training | Real gain measured per document |
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Wei Ling, head of operations at a freight forwarder in Singapore (illustrative example), handles 900 documents a month. Saving 12 minutes per document frees 900 × 12 = 180 hours a month, about 4 500 SGD at a loaded cost of 25 SGD an hour. The main gain is in errors: cutting the error rate from 3 % to 1 % avoids 18 errors a month, each costing on average 400 SGD (penalty, storage day, file rework), or 7 200 SGD a month. With a Kolonell build at 5 500 000 FCFA (8 380 EUR, about 12 500 SGD) and around 450 SGD a month of API and maintenance, the net monthly gain is about 11 250 SGD: payback in under 2 months. With a local integrator at 80 000 SGD, payback would be around 7 months.
FAQ
Can AI read poor-quality scans?
Yes, in most cases, but accuracy drops from 95 % to 80 or 85 % on blurry phone photos. The pipeline flags these documents for assisted rather than automatic entry.
Can data go straight into TradeNet?
The pipeline produces a file or pre-fill in the format your permit software expects, which the declarant checks before submission. Direct integration depends on your declaring software's API and is decided at scoping.
How many documents are needed to train the system?
An annotated sample of 300 to 500 real documents is enough to reach 95 % on key fields. It represents 8 to 12 days of work from an experienced declarant.
What does it cost to run?
For 900 three-page documents, expect 180 to 350 SGD a month of OCR and API, plus 75 000 FCFA (110 EUR) of Standard maintenance. Cost per document stays under 0.50 SGD.
Is the AI's HS code suggestion reliable?
It is 80 to 90 % reliable on regular goods from your history, much less on new products. That is why it remains a suggestion, validated by the declarant who carries the liability.
Let's scope your project. Send us 20 sample documents and your monthly volume: we will price your extraction pipeline between 4 000 000 and 7 000 000 FCFA (6 100 to 10 670 EUR), delivered in 10 to 12 weeks. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
