Use casesFinancial institutions & fintech

  • Bank
  • Participation bank
  • Consumer finance
  • Payment and e-money

Garnishment orders and legal letters

Garnishment notices arrive every day, each with a seven-day clock; a missed deadline can make the bank answer for the debt.

With the systemEach letter read, the debtor's products listed, the answer drafted; the specialist blocks and signs.

Who decides
Legal-operations specialist

UYAP · 4 November 2026Illustrative

Istanbul 14th Enforcement Office · First garnishment notice (İİK art. 89/1)

The rights and receivables of the debtor M*** K*** held at your bank are attached. Unless you object within seven days of service, the debt is deemed yours.

→ last day 11 November · the pension is not garnishable

The document as it arrives. In red, what the system marks for your expert.

1. The work

Today

  1. ReadLetters from enforcement offices (icra daireleri), tax offices, the social security institution (SGK), courts, and prosecutors arrive through UYAP, electronic garnishment channels, and on paper; a legal-operations team reads each by hand: a garnishment notice, a second notice, a release, or an information request?
  2. FindFor each one, the team finds the debtor and searches deposits, participation accounts, the safe-deposit box, and other products, checking where each balance comes from.
  3. CalculateThey work out by hand what may be blocked, keeping protected money such as a pension out and the block within the ordered amount.
  4. WriteThe answer is written and sent to the enforcement office within seven days of service; a missed deadline, a late block, or a payment to the customer after the block falls on the bank.
  5. SignThe legal-operations specialist checks, blocks, and signs each file from the same growing pile, and decides whether the bank objects.

With the system

  1. ReadThe SLM reads each letter, scanned paper included, and returns its type, the file number, the debtor's ID, the ordered amount, and the last day to answer.SLM
  2. FindThe agent, with read-only access, matches the debtor in core banking, lists every product with its balance, and flags salary and pension accounts by where the money comes from.Agentic AI
  3. CalculateYour written rule table, not a model, computes what may be blocked: the order of earlier garnishments, the salary share, and protected payments.Your rule
  4. WriteThe SLM drafts the answer in your template, stating each protected amount with its legal reason and marking any doubt for the specialist.SLM
  5. SignThe legal-operations specialist checks the file, applies the block in your own system, and signs the answer, as before; whether to object is always theirs.Person decides

What lands on the desk

The specialist gets the letter already read, the debtor's products already listed, and what may be blocked already computed by your own rules, with each protected amount and its reason. The answer is drafted in your template; the specialist checks, blocks, and signs.

FIG. 1 · Garnishment fileIllustrative
UYAP · 04.11.2026 · İstanbul 14. İcra DairesiFirst garnishment notice

SLM · what the letter says

  • TypeFirst garnishment noticeİİK art. 89/1
  • File2026/18342 Esas
  • DebtorM*** K***, ID number masked
  • Last day11.11.2026, seven days from service

Agent · products found, read-only

  • metDebtor matched in core banking3 products
  • metTurkish-lira demand accountblockable
  • flaggedPension account5510 art. 93not garnishable
  • Safe-deposit boxnone

Rule table · what may be blocked

  • Demand accountThe whole balance, which is below the ordered amount
  • PensionNothing: a pension is not garnishable5510 art. 93
  • ProposalBlock the demand account up to its balance

SLM · answer draft

Ready in your template: the amount to block, the pension account not garnishable with its reason, no safe-deposit box.

  • SLM
  • Agent
  • Code

Why a person decidesThe specialist checks the file, applies the block in your own system, and signs; whether to object is theirs.

DecidesLegal-operations specialist

2. The system

For your technical team — skip it if you like.

How it works

In

  • UYAP and electronic garnishment notices
  • Scanned paper letters via text reading (OCR)
  • Court and prosecutor letters

  1. SLM · understands

    Letter reader

    Fine-tuned to return type, file, debtor, amount, and last day

    Reads

    • Everything under In

    Out

    • Type: notice, second notice, release, request
    • Enforcement office and file number
    • Debtor's ID and the ordered amount
    • Date of service and the last day to answer

  2. Agent · prepares

    Debtor-matching agent

    Fixed workflow, read-only

    Reads

    • The debtor's ID from the letter

    Out

    • The debtor matched in core banking
    • Every product and its balance
    • Salary and pension accounts flagged
    • Safe-deposit box: held or not

  3. Code · calculates

    Blockable-amount rule table

    Your written rules, run as code; no model

    Reads

    • Products, balances, and flags
    • The ordered amount
    • Earlier garnishments, in order

    Out

    • What may be blocked, per product
    • Protected amounts and the legal reason
    • The block capped at the ordered amount

  4. SLM · understands

    Answer writer

    The same SLM, in your answer template

    Reads

    • The letter
    • The computed amounts
    • Your answer template

    Out

    • An answer draft in your template
    • Each protected amount with its reason
    • Points of doubt marked for the specialist

  5. Decision

    Legal-operations specialist

    Reads

    • The file and the answer draft

    Out

    • Checks the file and applies the block in your system
    • Signs and sends the answer
    • Decides whether to object
How the agent's work is arrangedFixed workflow

Fixed workflow

Every letter goes through the same steps in the same order (match the debtor, list the products, hand over to the rule table), so a fixed workflow fits and nothing is left for the agent to choose.

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)

  1. Test set

    500–1,000 recent letters of every type, paper and electronic, labeled by your senior legal-operations specialists with type, file number, debtor, amount, and last day, plus 100–200 answers they grade against your template.

  2. Dataset

    Past letters and the answers your team sent, with the fields as they were filed; personal data masked in training copies.

  3. Continued pre-training (CPT)If needed

    Only if the base model fails the test on the vocabulary of enforcement law and official letters.

  4. Supervised fine-tuning (SFT)

    On reading the fields of each letter type and on your answer template.

  5. Reinforcement learning (RL)Skipped

    Skipped: reading fields and writing to a template are taught by fine-tuning, and no further reward is needed.

  6. Retrieval

    Retrieval supplies your current templates and the law texts an answer cites; the amounts come from the rule table, never from the model.

  7. Monitoring

    The specialist reviews every answer; a new letter format or a rule change triggers a re-test before the next release.

Where the agentic system stops

The legal-operations specialist checks every file, applies the block in the bank's own system, and signs every answer; whether to object is always theirs.

Show its tools, limits, and record
Tools and access
Reads letters from UYAP, the electronic garnishment channels, and document management; read-only core banking (customer search, products, balances, incoming payment types); writes only the case file and a block proposal for the specialist.
May not
Apply or lift a block, release or pay out money, decide an objection, contact the customer, give a legal opinion, or grant itself any access.
Who decides and the written limit
Legal-operations specialistThe legal-operations specialist checks every file, applies the block in the bank's own system, and signs every answer; whether to object is always theirs. The system never applies or lifts a block, never decides an objection, never contacts the customer, and gives no legal opinion: any doubt goes to the specialist.
Stops when
An uncertain debtor match, a letter it cannot read with confidence, a payment whose source cannot be told, or any legal doubt: the file goes to the specialist, marked.
Record
The letter, its date of service and last day, every product checked, the rule-table version and each computed amount, the draft, and who blocked and signed.

How agentic systems stay inside limits

3. Value and conditions

Where the time goes, and what your institution will measure.

Where the value comes from

  1. Answers sent within the seven days.
  2. Fewer wrong blocks on protected money.
  3. Specialist hours freed for the hard files.

What your institution measures

  • Share of letters answered on time
  • Wrong blocks on protected money
  • Letters that reach a second notice
  • Specialist time per letter

Measured in the Decision Blueprint

Today's on-time share and specialist time per letter, from your letter log, set the baseline; your institution sets the target.

Conditions

The rules, the data, and the known error, with its guard.

  • The deadlines are the law's: seven days from service to answer the enforcement office (İİK art. 89) and the tax office (Law 6183 art. 79); a second notice starts its own seven days.
  • Some money is protected: pensions and social-security benefits cannot be garnished except for SGK's own claims and alimony (Law 5510 art. 93), and a salary only above the debtor's living amount, by the share and order of İİK art. 83.
  • No machine-learning model is needed here: what may be blocked comes from your written rule table, a rule, not a model. Legal interpretation stays with your legal owners.
  • Letters and balances stay on your servers, and personal data is masked in training copies.

Start here

We take a month of letters with your legal-operations team, time each step from receipt to answer, collect your rule table and answer templates, and test the letter reader on your own letters before anything is built.

Who builds it Every engagement is led by an engineer who has built machine-learning models that real decisions run on since 2012, and language models and agentic systems in production since 2024. About eqlx

Request a Decision BlueprintWhat the Decision Blueprint covers

Equalize the gap.

Bring work that takes expertise. Leave knowing the first step.

A short description of the work is enough; no confidential records.

Start with a Decision BlueprintTalk to the engineer

https://eqlx.ai/use-cases/garnishment-orders · Printed copy; the website is the current version.