Use casesFinancial institutions & fintech
The lease file
A dealer sends a proforma invoice; you must know if the lessee is good, the price is right, and the machine is not already financed elsewhere.
- Kinds of AI
- ModelsSLMsAgentic AI
- Who decides
- Credit committee
- Where it runs
- Your servers, banking data kept in Türkiye
Model · lessee risk
SLM · documents
- flaggedSerial on the proforma differs from the insurance policy (one digit)
- Documents received11/13
- not metMissing: signature circular, delivery note
Agent · double financing
- metFKB registry: no contract on this serial
- metOwn book: no match
- Model
- SLM
- Agent
Why a person decidesAbove the branch limit: the credit committee decides.
DecidesCredit committee
1. The desk
Today
- ReadAn analyst reads a dozen PDFs (proforma, invoice, insurance, registration, authorized-signatory circular) and calls the dealer.
- PriceAsset price is judged by experience; residual value is a table.
- CheckDouble financing is caught only if someone checks the registry by serial number.
- DecideThe credit committee decides on the file the analyst assembled by hand, while the dealer waits.
With the system
- ReadThe SLM reads every document, often scanned, into one structured file with serial numbers, amounts, and dates, and flags mismatches between documents.SLM
- PriceModels score the lessee's risk, check the asset price against comparables with the residual value at the end of the term and its uncertainty band, and score the dealer.Model
- CheckThe agent checks the lease-contract registry of the Association of Financial Institutions (FKB) and the vehicle registry by serial number, searches your own book for duplicates, requests missing documents from the dealer, and drafts the memo; any identity mismatch goes to the fraud team.Agentic AI
- DecideThe credit committee decides, as before; the vendor is paid only after a named person signs delivery and acceptance.Person decides
What lands on the desk
The credit committee sees the lessee, the asset, and the dealer on one page, with the registry and duplicate checks already run. A one-digit serial mismatch is caught before funding, and the next step names what would clear it.
Model · lessee risk
Model · asset price and vendor
- flaggedProforma price above the comparable range: explain or renegotiate
- flaggedDealer early defaults above the portfolio average
SLM · documents
- flaggedSerial on the proforma differs from the insurance policy (one digit)
- Documents received11/13
- not metMissing: signature circular, delivery note
Agent · double financing
- metFKB registry: no contract on this serial
- metOwn book: no match
Agent · next step
Corrected insurance policy and an on-site photo of the serial plate.
- Model
- SLM
- Agent
Why a person decidesAbove the branch limit: the credit committee decides.
DecidesCredit committee
2. The system
For your technical team — skip it if you like.
How it works
In
- Proforma, invoice, insurance policy
- Registration or trade registry papers
- Signatory circular, tax certificate
- Financial statements, often scanned
SLM · understands
Document reader
Reads
- Everything under In
Out
- Parties, asset identifiers (serial, VIN)
- Amounts, VAT, and dates
- Cross-document mismatches flagged
Model · estimates
Lessee PD model
Probability of default: SME scorecard or GBM
Reads
- Lessee financials
- The bureau
Out
- A risk band
- Its reasons
Model · estimates
Asset and residual value model
GBM or hedonic regression, uncertainty band, expert overlay
Reads
- Brand, model, year
- Comparables and market indices
Out
- Price check against comparables
- Residual value at end of term
Model · estimates
Vendor risk score
Reads
- Dealer history: early defaults, fraud
Out
- A dealer band
Agent · prepares
Registry-check agent
Parallel checks
Reads
- All of the above
Out
- FKB lease registry and vehicle registry
- Own-book duplicate and invoice fingerprint
- Missing documents requested from dealer
- Draft memo, routed to the authority level
Decision
Credit committee
Reads
- The complete file
Out
- The credit committee decides
- The vendor is paid only after a named person signs delivery and acceptance
How the agent's work is arrangedParallel
Parallel
Registry, duplicate, document, price, and dealer checks are independent, and all must come back before the file is complete.
How the models are trained and tested
Your expert writes the test first. A step the test does not need is skipped, and we say why.
Show the training steps
Small language model (SLM)
Test set
100–200 past lease files graded by analysts (fields, mismatches caught).
Dataset
Past files with their final checked values.
Continued pre-training (CPT)Skipped
Skipped unless the base model fails the test on vocabulary.
Supervised fine-tuning (SFT)
On the lessor's document types.
Reinforcement learning (RL)Skipped
Skipped: extracted fields are checked against the documents and the test set.
Lessee, asset, and vendor models
Target
Lessee PD as in the application scorecard; asset value: the realized or remarketing price; vendor score: early default or fraud per dealer.
Method and challenger
A scorecard against a gradient-boosted model for the lessee; GBM against hedonic regression for the asset, with an expert overlay for niche equipment.
Validation evidence
Asset value judged on error against realized resale; residual value back-tested against returned-asset prices; evidence prepared for your own validation.
Where the agentic system stops
The credit committee decides the lease, or below its limit the approver your authority matrix names; the vendor-program manager decides dealer limits and suspensions; the fraud team handles every identity mismatch.
Show its tools, limits, and record
- Tools and access
- Read-only registries (the FKB lease-contract registry, the vehicle registry where available), the bureau, and the own book; may request documents from the dealer by template.
- May not
- Approve, pay the vendor, or change the price.
- Who decides and the written limit
- Credit committeeThe credit committee decides the lease, or below its limit the approver your authority matrix names; the vendor-program manager decides dealer limits and suspensions; the fraud team handles every identity mismatch. The agent never holds credit authority, and the vendor is paid only after a named person signs delivery and acceptance.
- Stops when
- Any serial or VIN mismatch, registry hit, or invoice duplicate goes to the fraud team; a price above the comparable band goes to an analyst; documents missing at the deadline leave the file waiting.
- Record
- Every registry query, match result, and document version.
3. Value and conditions
Where the time goes, and what your institution will measure.
Where the value comes from
- Time to decision on dealer deals: dealers send business to the fastest lessor.
- Analyst hours per file.
- Double-financing, phantom-asset, and residual-value losses avoided.
What your institution measures
- Time to decision
- Files per analyst
- Document errors caught before funding
- First-12-month default by dealer
- Residual-value forecast error vs. realized price
Measured in the Decision Blueprint
Baseline time per file and the mismatch rate on recent files set the starting point; your institution sets the target.
Conditions
The rules, the data, and the known error, with its guard.
- Türkiye: Law 6361 and the banking regulator's (BDDK) rules for leasing companies; the access route to the FKB registry (API or manual) is confirmed first.
- EU: the AI Act treats lessee assessment as high-risk only for natural-person lessees.
- Price data is thin for niche assets, so the file shows the uncertainty band.
- Dealer–lessee collusion: the vendor model and physical serial checks address it.
Start here
We take recent dealer deals, time each step from proforma to decision, and count the document mismatches your analysts caught and missed.
Similar desks
- ModelsSLMsAgentic AI
Early warning on the loan book
TodayThe borrower is current; other banks are already cutting its limits, and you see its quarter only at renewal.
With the systemA file, not a spreadsheet: signals checked in parallel and gathered in one place.
Watch-list committee decides
- ModelsSLMsAgentic AI
Merchant onboarding and monitoring
TodayNew merchants sign up daily; one merchant set up to defraud or launder can cost more than many good ones earn.
With the systemEach new merchant checked against your policy before anyone signs. Nothing is automatic.
Onboarding officer decides
Equalize the gap.
Bring one decision. Leave knowing the first step.
A short description of the decision is enough; no confidential records.
https://eqlx.ai/use-cases/leasing-file · Printed copy; the website is the current version.