The verdict in three sentences
For a Singapore forwarder handling 7,000 shipments a year, manual rekeying of bills of lading, commercial invoices and packing lists costs more than an AI extraction project priced at 25,000 to 45,000 USD. With 35 fields read at 95% accuracy and human review on the remaining 5%, the operations team reassigns 2 FTEs to customer service and customs filing. The project ships in 8 weeks and usually pays back in under 18 months.
What the AI actually reads in a shipment file
A typical file contains an ocean bill of lading (B/L), one or more commercial invoices, a packing list and sometimes a certificate of origin. Documents arrive as native PDFs, poor scans or phone photos sent over WhatsApp or email. A modern engine combines OCR, a language model and business rules: it classifies the document, extracts the fields, then checks consistency across documents (B/L gross weight against the packing list total, for example).
The 35 standard fields: B/L number, vessel, voyage, port of loading, port of discharge, shipper, consignee, notify party, container and seal numbers, container type, package count, gross and net weight, volume, goods description, suggested HS code, invoice value, currency, Incoterm, freight, invoice number, date, origin.
| File step | Manual entry | With AI extraction | Time saved per file |
|---|---|---|---|
| Document identification | 4 min | 10 s | 3 min 50 |
| B/L entry (15 fields) | 9 min | 40 s + 1 min review | 7 min 20 |
| Invoice and packing list (20 fields) | 12 min | 1 min + 2 min review | 9 min |
| Weight and value reconciliation | 5 min | automatic | 5 min |
| Downstream error correction | 6 min (average) | 1 min | 5 min |
| Total | 36 min | about 5 min 50 | about 30 min |
Across 7,000 files, 30 minutes saved equals 3,500 hours a year, roughly 2 full-time roles on a 1,760-hour basis.
2026 budget: three project levels
Figures below are 2026 order-of-magnitude estimates for Singapore, excluding the existing forwarding software licence.
| Item | Level 1 (essential) | Level 2 (standard) | Level 3 (advanced) |
|---|---|---|---|
| Document types | B/L + invoice | B/L, invoice, packing list | + certificate of origin, permits |
| Fields extracted | 20 | 35 | 50+ |
| Forwarding system integration | Excel export | API or automated CSV import | two-way API + checks |
| Review interface | basic | review queue per agent | dashboard + statistics |
| Project price | 25,000 USD | 35,000 USD | 45,000 USD |
| Monthly running cost (AI + hosting) | 400 USD | 650 USD | 1,000 USD |
| Timeline | 5 weeks | 8 weeks | 11 weeks |
Running cost depends on page volume: budget about 0.03 to 0.06 USD per page processed by the model, plus hosting. For 7,000 files averaging 6 pages, the bill stays under 650 USD a month.
What moves the quote
Scan quality matters a lot: phone photos need preprocessing (deskewing, contrast). The number of B/L layouts also counts, since each carrier (MSC, Maersk, CMA CGM, ONE) has its own template, although recent models generalise well. Finally, integration with the forwarding software is often the longest item when the vendor exposes no API.
Planning and governance
An 8-week plan for level 2: weeks 1 and 2, collect 300 real anonymised files and scope the fields; weeks 3 to 5, configure extraction and consistency rules; week 6, connect the forwarding system; week 7, parallel run on live flow; week 8, cutover and training. The confidence threshold is set field by field: below 90%, the field goes to the human review queue.
On data, documents stay on a server you control with access logging, which supports compliance with Singapore's PDPA.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Mini case study
Wei Ling, head of operations at a Singapore forwarder, handles 7,000 files a year with 6 data entry staff. Loaded cost per staff member: 4,200 USD a month, or 50,400 USD a year. She picks level 2 at 35,000 USD plus 650 USD a month in running costs (7,800 USD a year).
- Savings: 2 FTEs reassigned, or 100,800 USD a year of freed capacity.
- Net annual gain: 100,800 minus 7,800, or 93,000 USD.
- Payback: 35,000 divided by 93,000, about 5 months.
On top of that, weight and value errors caught before customs filing reduce penalties and demurrage, a gain Wei Ling left out of the calculation.
FAQ
What accuracy should we expect on poor scans?
On native PDFs, accuracy often exceeds 98%. On phone photos it drops to 88 to 92%, which is why the human review queue secures the 5 to 10% of uncertain fields.
Does the AI suggest the HS code?
Yes, it can suggest an HS code from the goods description, with about 80% correct suggestions on the first 6 digits. The declarant keeps the final decision, which remains essential with customs.
Do we need to replace our forwarding software?
No. The project plugs into the existing tool via API, CSV import or, as a last resort, an RPA bot. The connector is usually 20 to 30% of the budget.
How many files do we need to start?
A sample of 200 to 300 real files is enough to configure and test. Below 2,000 files a year, payback often exceeds 30 months in lower-wage markets.
What if a carrier changes its B/L template?
Language models adapt without retraining in most cases. A maintenance contract of 200 to 350 USD a month covers rule adjustments.
Let's scope your project. Send us 20 sample files and your annual volume: we will price the scope, an indicative budget between 25,000 and 45,000 USD and a 5 to 11 week timeline. 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.
