AI document checks for leasing applications: cost and time saved

The verdict in three sentences

A lender handling 2,000 leasing applications a month at 40 minutes of checks each burns more than 1,300 analyst hours, roughly eight full-time people. AI document checks (extraction, consistency checks, forgery detection) bring that down to 10 to 15 minutes per file, as long as a human analyst still signs off every decision. In Dubai or London, budget roughly USD 80,000 to 220,000 (about GBP 65,000 to 175,000) and 3 to 4 months for a serious build that satisfies UK GDPR or the UAE data protection rules.

Where the 40 minutes go

A typical asset finance or vehicle leasing file contains ID or passport, a trade licence or Companies House record, three to six months of bank statements, two years of accounts for a business, proof of income, the supplier quote and bank details. The analyst rekeys, cross-checks, then chases what is missing.

That last row matters. The aim is to equip the analyst, not to remove human judgement. Under the FCA Consumer Duty in the UK, and under the Central Bank of the UAE's consumer protection rules, you must be able to explain a decline and show who approved what.

What you need to build, and what it costs

Four building blocks: an upload portal or API for customers and introducer brokers, an extraction engine (OCR plus a language model), a rules engine reflecting your credit policy, and a review screen for analysts with a full audit trail. Integration with your loan management system often costs as much as the AI itself.

At Kolonell, this falls into the complex custom application band, 7,000,000 to 15,000,000+ FCFA (EUR 10,670 to 22,870+) depending on scope. Application maintenance follows our 35,000, 75,000 or 120,000 FCFA a month grid (EUR 53, 114 or 183); dedicated hosting is quoted separately.

Data protection: the non-negotiables

Bank statements and ID documents are personal data. In the UK you need a Data Protection Impact Assessment under UK GDPR and, because automated processing feeds a credit decision, safeguards under Article 22: a human must be able to review and override. In Dubai, the DIFC Data Protection Law or the federal PDPL applies depending on where you are licensed, with restrictions on transferring data abroad. Three design choices make compliance easier: in-region hosting for documents, pseudonymisation before anything reaches the AI model, and automatic purge of declined files after your retention period.

Faster decisions are also a sales argument. Dealers and equipment suppliers who introduce business get an answer in 24 hours instead of 4 days.

Mini case study

Sarah, COO of an asset finance lender in London, handles 2,000 applications a month. Before: 2,000 × 40 min = 80,000 minutes, or 1,333 hours, the equivalent of 8.3 analysts at 160 hours. After: 2,000 × 12 min = 400 hours, or 2.5 analysts. She moves 5 people to collections and broker relations. At an estimated loaded cost of GBP 4,200 a month per analyst, that frees GBP 21,000 a month of capacity. With a GBP 110,000 build and GBP 3,000 monthly running costs, the net gain is GBP 18,000 a month: payback in just over 6 months, before counting fraud avoided. A single doctored set of accounts caught on a GBP 60,000 equipment lease covers well over a year of running costs.

FAQ

Can the AI decline an application on its own?

It should not. The AI flags anomalies and proposes a score, the analyst decides. In our projects 100% of declines are confirmed by a person, with a note stored in the audit trail.

What forgery detection rate is realistic?

Crudely edited statements are caught more than 90% of the time through metadata and totals analysis. More sophisticated fakes require a library of statement templates per bank, built up during the first 3 months.

Do documents have to stay in the UK or the UAE?

Not always, but transfers abroad need a legal basis and safeguards. Keeping documents in-region and sending only pseudonymised data to the model usually saves several weeks of legal review.

How many files are needed to train the system?

A sample of 300 to 500 anonymised files is enough to calibrate extraction on your formats. Accuracy on key fields typically exceeds 95% after 6 to 8 weeks of tuning.

Can we start small?

Yes. A first release limited to consumer vehicle leasing goes live in 8 to 10 weeks, then business applications with accounts follow.

Let's scope your project. Tell us your application types, volumes and current loan system, and we will scope a first release deliverable in 3 to 4 months. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.