Use casesFinancial institutions & fintech

Regulation to policy

A new circular arrives, and nobody knows which sections of your policies it changes.

Kinds of AI
SLMsAgentic AI
Who decides
Compliance officer
FIG. 1 · Regulation to policy change listIllustrative
New circularCustomer complaint handling

  • Answer complaints within a shorter deadlineArt. 5(2)

    • Complaints Policy §3.4 "response time"quote matchedChange
    • Branch Procedure BP-12 §2quote matchedChange

  • Record channel and category for each complaintArt. 6(1)

    • Complaints Policy §5.1quote matchedReview
  • Checker1 link droppedits quote was not in the source
  • Unsupported links0
  • SLM
  • Agent

Written limitDraft only; nothing changes until the compliance officer signs.

DecidesCompliance officer

1. The desk

Today

  1. ReadCompliance reads each new text from the banking regulator (BDDK), the central bank (TCMB), the financial intelligence unit (MASAK), or the data protection authority (KVKK).
  2. MapDozens of internal policies are searched by hand for the sections the text touches.
  3. ListThe mapping lives in emails; business owners learn of a change late.
  4. Sign offThe compliance officer signs off, and at inspection nobody can show why a section was or was not changed.

With the system

  1. ReadThe SLM splits the new text into obligations per article: what changed, the effective date, and who it applies to, each with its citation.SLM
  2. MapRetrieval over your policy library and the same SLM propose candidate sections for each obligation, quoting both texts.SLM
  3. ListThe agent writes a draft change list in the compliance workspace; its checker drops or revises any link whose quote is not in the source, and flags possible gaps.Agentic AI
  4. Sign offThe compliance officer signs off every item and each policy owner approves their own section, as before; the record shows why each section changed or did not.Person decides

What lands on the desk

The compliance officer starts from a list in which every obligation is quoted from the circular and every affected section is quoted from your own policy. Obligations with no matching section are flagged as possible gaps, and the checker shows what it dropped.

FIG. 2 · Regulation to policy change listIllustrative
New circularCustomer complaint handling

  • Answer complaints within a shorter deadlineArt. 5(2)

    • Complaints Policy §3.4 "response time"quote matchedChange
    • Branch Procedure BP-12 §2quote matchedChange

  • Record channel and category for each complaintArt. 6(1)

    • Complaints Policy §5.1quote matchedReview

  • Annual report to the boardArt. 9

    • No section founda possible gapGap
  • Checker1 link droppedits quote was not in the source
  • Unsupported links0
  • SLM
  • Agent

Written limitDraft only; nothing changes until the compliance officer signs.

DecidesCompliance officer

2. The system

For your technical team — skip it if you like.

How it works

In

  • The new regulation text
  • Its predecessor

  1. SLM · understands

    Regulation reader

    Article segmentation and change detection

    Reads

    • Everything under In

    Out

    • Obligations per article
    • What changed, when, for whom
    • A citation for each

  2. SLM · understands

    Policy mapper

    The same SLM: retrieval, then a mapping task

    Reads

    • The obligations
    • Your policy library

    Out

    • Candidate sections
    • Quotes from both sides

  3. Agent · prepares

    Checking agent

    Check and revise

    Reads

    • The candidates

    Out

    • A draft change list
    • Every citation and quote checked
    • Unsupported links revised or dropped

  4. Decision

    Compliance officer, then policy owners

    Reads

    • The change list

    Out

    • Each item approved, edited, or rejected
How the agent's work is arrangedCheck and revise

Check and revise

The risk is a confident link that is not in the text; a separate checking pass that verifies each citation and quote, with revision until it passes, targets exactly that.

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

    30–60 past regulations with the mappings compliance actually made, graded by compliance staff (recall of affected sections, precision of links, citation correctness).

  2. Dataset

    Past regulation-to-policy mappings and change logs.

  3. Continued pre-training (CPT)If needed

    On Turkish regulatory language, only if the base model fails legal vocabulary on the test.

  4. Supervised fine-tuning (SFT)

    On obligation extraction and the mapping format.

  5. Reinforcement learning (RL)Skipped

    Skipped by default; a reward is possible (citation verified, link accepted by compliance) once enough reviewed mappings exist.

  6. Retrieval

    Retrieval supplies the current policy text; the model is not retrained when policies change.

Where the agentic system stops

Nothing changes automatically.

Show its tools, limits, and record
Tools and access
Read-only regulator sources and policy library; writes only a draft change list in the compliance workspace.
May not
Edit a policy, publish, or give a legal interpretation.
Who decides and the written limit
Compliance officerNothing changes automatically. The compliance officer signs off every item, and each policy owner approves their own section change. Legal interpretation stays with the institution's legal owners.
Stops when
A citation fails twice: the item goes to the compliance officer marked "unsupported"; an obligation with no candidate section is flagged "possible gap".
Record
Source versions, every candidate, checker results, and the approver's decision per item.

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. Time from publication to an approved change list.
  2. Fewer missed sections.
  3. An inspection-ready trail.

What your institution measures

  • Days from publication to approved change list
  • Affected sections found later (misses)
  • Share of links accepted by compliance
  • Compliance hours per regulation

Measured in the Decision Blueprint

A replay of recent regulations against your past mappings sets the baseline; your institution sets the target.

Conditions

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

  • No legal advice: eqlx prepares the technical mapping.
  • Regulator texts change after publication (corrections), so the source version is pinned.
  • Runs on your servers, because the policies are confidential; banking data stays in Türkiye.

Start here

We replay a few recent circulars against the mappings your compliance team actually made, and see which sections the reader finds and which it misses.

Request a Decision BlueprintWhat the Decision Blueprint covers
  • ModelsSLMsAgentic AI

    SME credit referrals

    TodayReferred files wait while analysts rebuild the firm from four systems and last year's committee notes.

    With the systemOne file with the rating, its reasons, and the authority level.

    Credit manager decides

  • 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

Equalize the gap.

Bring one decision. Leave knowing the first step.

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

Start with a Decision BlueprintTalk to the engineer

https://eqlx.ai/use-cases/regulation-to-policy · Printed copy; the website is the current version.