The verdict in three sentences
A wholesaler's order desk that re-keys 50 to 400 PDF purchase orders a day often ties up 1 to 3 people on a task with no added value. AI extraction connected to the ERP costs 10,000 to 35,000 EUR on the market, then 0.01 to 0.08 EUR per document, with target accuracy above 95 % on key fields. Human review stays in place for discrepancies, and 1 to 2 roles can be redeployed to customer relations.
Why re-keying costs more than you think
Each customer sends its own format: an order generated by its ERP, a spreadsheet saved as PDF, a scan of a signed document, sometimes a photo. The order desk assistant reads it, looks up the item reference, checks the price, enters quantities and the requested delivery date. Keying errors (2 to 4 % of lines on average) lead to returns, credit notes and disputes.
| Indicator (wholesaler, 150 orders/day, 2026 order of magnitude) | Manual re-keying | AI extraction + review |
|---|---|---|
| Time per purchase order | 4 to 8 min | 30 s to 1 min of checking |
| Time to land in the ERP | 2 to 24 h | 5 to 30 min |
| Lines with keying errors | 2 to 4 % | 0.3 to 1 % |
| Accuracy on key fields | Varies with workload | 95 to 99 % |
| Orders needing human review | 100 % | 10 to 25 % (gaps, new formats) |
| FTEs involved | 2 to 3 | 0.5 to 1 |
| Month-end peaks | Overtime | Absorbed without extra staff |
Custom solution cost versus market tools
IDP (intelligent document processing) platforms work well on invoices, but purchase orders often need business logic: mapping the customer's reference to your item code, checking contract pricing, splitting into deliveries.
| Option | 2026 cost (market, order of magnitude) | Strengths | Limits |
|---|---|---|---|
| Subscription IDP platform | 300 to 2,000 EUR/month + 0.05 to 0.15 EUR/page | Fast start | Limited item mapping |
| Custom solution | 10,000 to 35,000 EUR | Business rules, direct ERP integration | 8 to 14 week project |
| Cost per document (custom) | 0.01 to 0.08 EUR | Predictable, no tiers | Depends on the model |
| Customer ordering portal (alternative) | 15,000 to 40,000 EUR | Removes PDFs at the source | Slow customer adoption |
| Maintenance and new formats | 10 to 15 % of the project per year | Keeps accuracy | Essential |
At Kolonell, an extraction pipeline connected to your ERP falls within our custom application grid: 4 to 7 M FCFA (6,100 to 10,670 EUR) for medium complexity, 7 to 15 M FCFA and above (10,670 to 22,870 EUR) with an item mapping table, price checks and several sites. Maintenance: 35,000, 75,000 or 120,000 FCFA per month (50, 110 or 180 EUR).
Steps of a successful project
- Sample: 300 to 500 real orders covering your top 20 customers, who often represent 70 % of volume.
- Key fields: order number, customer, references, quantities, prices, delivery date, address.
- Mapping table: customer references to your item codes, enriched with every correction.
- Exception queue: any field below the confidence threshold or any off-contract price goes to review.
- ERP integration: order created as a draft, never confirmed without an explicit rule.
Mini case study
Julien, order desk manager at an electrical supplies wholesaler in Nantes (illustrative case), handles 120 orders a day, or 2,640 a month. At 5 minutes per order, that is 220 hours a month. After AI extraction, checking takes 1 minute on average, or 44 hours. Saving: 176 hours at 32 EUR loaded, so 5,632 EUR. Monthly costs: 106 EUR of processing (0.04 EUR per document) and 180 EUR of maintenance. Net gain: 5,346 EUR a month. For a project estimated at 22,000 EUR, payback comes in about 4 months, and 1.2 roles move to key account follow-up.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
FAQ
What accuracy can we really reach?
On key fields of orders generated by customer ERPs, 97 to 99 % is common. On poor-quality scans, expect 90 to 95 %, hence the exception queue.
Do we need to replace our ERP?
No. Extraction creates orders through your ERP's existing API or imports (Sage, Odoo, SAP Business One, Dynamics). The connector accounts for 15 to 30 % of the budget.
What happens with a new customer in an unknown format?
Recent models read a new format without a prior template. The first 10 to 20 orders go through human review, then accuracy aligns with the average.
Is our customers' data protected?
Require EU hosting, limited PDF retention (for example 90 days) and no reuse for model training.
How long until production?
Usually 8 to 14 weeks, including 3 to 4 weeks of parallel processing to measure accuracy before stopping manual entry.
Let's scope your project. Send us an anonymised sample of orders, your daily volume and your ERP: we will price the scope within the 4 to 15 M FCFA grid and give you a go-live date. Detailed quote within 48 h. WhatsApp +221 77 596 93 33. Or request your free quote.
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.
