Use casesFinancial institutions & fintech

  • Bank
  • Participation bank
  • Brokerage
  • Portfolio management company

Suitability and appropriateness assessment

Every client who opens an account, asks for advice, or wants a derivative or leveraged product needs the right test, answers that hold together, and the written warning the result calls for.

With the systemEach client's answers read, the history and forms gathered, the test scored by your table; the adviser decides.

Who decides
Investment adviser

Mobile app · test form · 4 November 2026Illustrative

Appropriateness test form · general client

Question 9. Which instruments have you traded before? Checked: shares, investment funds, futures and options.

→ trading history: shares and funds only, no derivatives

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

1. The work

Today

  1. ReadClients fill in the test form at account opening, in the app, or at the branch: investment horizon, risk and return preference, income and investable assets, age, occupation, education, and the instruments they have traded; a customer representative reads the answers and any notes.
  2. GatherFor each request they look up the client's class (general or professional), the trading history with you, earlier test results and warnings, and the signed risk disclosure forms, and read the product's documents for its risk class.
  3. ScoreThe form is scored by hand or in a spreadsheet against your procedure and the result compared with the product: the appropriateness test (uygunluk testi) for orders, the suitability test (yerindelik testi) for advice and portfolio management.
  4. WriteThe written warning, when the result calls for one, is typed from a template; an answer that contradicts the trading history is easy to miss.
  5. DecideThe investment adviser decides whether the service or the product can be offered, which warning goes to the client, and whether to serve a client who still insists after a warning.

With the system

  1. ReadThe SLM reads the client's form, free-text answers, and the representative's notes, and the product's documents, and returns structured answers, the product's features, and every answer that contradicts another.SLM
  2. GatherThe agent, with read-only access, pulls the client's class, the trading history, earlier results and warnings, and the signed forms, and lists what is missing or out of date.Agentic AI
  3. ScoreYour rule table, run as code, not a model, says which test applies, scores the answers as your procedure writes, and compares the result with the product's risk class.Your rule
  4. WriteThe SLM drafts the written warning in your template and the assessment note for the file, marking every contradiction for the adviser.SLM
  5. DecideThe investment adviser checks the file and decides, as before; compliance reviews every contradiction the system marks, and nothing reaches the client until a person sends it.Person decides

What lands on the desk

The adviser gets the client's answers already read, the history and the forms already pulled, and the test, the result, and the warning it calls for already set by your own rule table. The warning is drafted; the adviser decides and sends.

FIG. 1 · Assessment fileIllustrative
Mobile app · general client · request 04.11.2026Index options on the derivatives market

SLM · what the client says

  • RequestIndex options on Borsa İstanbul's derivatives market (VİOP)app request
  • ExperienceSays they have traded derivatives beforeform, question 9
  • PreferenceShort horizon; high risk acceptedform, questions 2–3

Agent · records found, read-only

  • Client classgeneral
  • flaggedDerivatives in the trading historytrading historynone; shares and funds only
  • metGeneral risk disclosure formdocumentssigned
  • Earlier warningsnone

Rule table · test and result

  • TestAppropriateness test: a general client placing ordersIII-39.1 art. 33
  • ResultNot appropriate for options, on the answers and the history
  • ThenA written warning; no one-to-one advice on this productart. 33/9

SLM · drafts

Ready in your template: the written warning, and the file note marking that the stated experience contradicts the trading history.

  • SLM
  • Agent
  • Code

Why a person decidesWhether to warn, refuse, or serve a client who insists is the adviser's, inside your procedure.

DecidesInvestment adviser

2. The system

For your technical team — skip it if you like.

How it works

In

  • Test forms from the app and the branch
  • Representative's notes
  • Product documents and risk disclosures

  1. SLM · understands

    Answer and product reader

    Fine-tuned to return the answers and the product's features

    Reads

    • Everything under In

    Out

    • Horizon, risk preference, income, assets
    • Knowledge and trading experience
    • Product features: leverage, complexity
    • Answers that contradict each other

  2. Agent · prepares

    Assessment-file agent

    Fixed workflow, read-only

    Reads

    • The answers and the product

    Out

    • General or professional client
    • Trading history with you
    • Earlier results and warnings
    • Signed forms, and what is missing

  3. Code · calculates

    Test rule table

    Your written procedure and scoring, run as code; no model

    Reads

    • The answers and the history
    • The product's risk class
    • The service requested

    Out

    • Which test applies, or none
    • The client's result against the product
    • Warning required, or service not allowed

  4. SLM · understands

    Warning writer

    The same SLM, in your warning templates

    Reads

    • The assessment file
    • Your templates

    Out

    • The written warning, when required
    • Assessment note for the file
    • Contradictions marked

  5. Decision

    Investment adviser

    Reads

    • The assessment file and the drafts

    Out

    • Offers or refuses the service or the product
    • Sends the warning the result calls for
    • Decides on a client who insists after a warning
    • Compliance reviews every marked contradiction
How the agent's work is arrangedFixed workflow

Fixed workflow

Every request takes the same steps in the same order (pull the class, the history, the earlier results, and the forms, then hand over to the table), so a fixed workflow fits and leaves the agent nothing 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

    300–600 recent client files labeled by your senior advisers and compliance with the answers, the contradictions, and the product features, plus 100–200 warnings they grade against your templates.

  2. Dataset

    Past forms, notes, product documents, and the warnings sent; personal data masked in training copies.

  3. Continued pre-training (CPT)If needed

    Only if the base model fails the test on capital-markets product vocabulary.

  4. Supervised fine-tuning (SFT)

    On your test forms, product documents, and warning templates.

  5. Reinforcement learning (RL)Skipped

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

  6. Retrieval

    Retrieval supplies your current procedure and the product documents; which test applies and the result come from the rule table, never from the model.

  7. Monitoring

    Compliance reviews a sample of files every month; each new product or change to the procedure triggers a re-test of the table and the drafts before it applies.

Where the agentic system stops

The investment adviser decides, inside your written procedure, whether the service or the product is offered, sends the written warning the result calls for, and decides whether to serve a client who still insists after one; compliance reviews every contradiction the system marks.

Show its tools, limits, and record
Tools and access
Reads the account-opening and document systems; read-only client class, trading history, earlier tests and warnings, and the signed forms; writes only the assessment file.
May not
Change a client's answers, class, or result, open an account or accept an order, give investment advice, send a warning, contact the client, or grant itself any access.
Who decides and the written limit
Investment adviserThe investment adviser decides, inside your written procedure, whether the service or the product is offered, sends the written warning the result calls for, and decides whether to serve a client who still insists after one; compliance reviews every contradiction the system marks. The system never changes an answer, a class, or a result, never opens an account, accepts an order, or gives advice, and never contacts the client.
Stops when
Answers the trading history contradicts, a professional class without its documents, missing or out-of-date answers, a leveraged product without the demo account and the signed declaration, or a product it cannot place in a class: the file goes to the adviser, marked.
Record
The form and its channel, every record pulled, the rule-table version, the test and the result it gave, the drafts, the warning sent, and who decided and sent it.

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. The right test on every request, scored the same way.
  2. Contradictions found before an order, not in an inspection.
  3. Adviser hours freed for the clients who need a conversation.

What your institution measures

  • Requests with the test and the warning on file
  • Contradictions found before the service
  • Files compliance sends back
  • Time from request to decision

Measured in the Decision Blueprint

Today's files missing a test or a warning, the contradictions found later, and the time per request, from your compliance reviews, set the baseline; your institution sets the target.

Conditions

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

  • The appropriateness test (uygunluk testi) asks whether a client has the knowledge and experience to understand the risks of a product offered or requested; in order handling and public-offering intermediation it is owed only to general clients, and investment fund units and listed government debt are exempt when the client asks for them and is told so. If the result is negative, or the information is missing, the client is warned in writing and gets no one-to-one advice on that product; the firm may still serve a client who insists (SPK Communiqué III-39.1 art. 33).
  • The suitability test (yerindelik testi) comes before a portfolio-management or investment-advice agreement and weighs the client's objectives, financial situation, and knowledge and experience; a service the result does not support cannot be offered, and without the information the service cannot be offered at all and the client is told so in writing (SPK Communiqué III-37.1 art. 40).
  • Leveraged trading has its own rules: no portfolio management or investment advice on it for the clients it is offered to, a demo account before a general client opens an account, and a signed declaration of its risks at opening (III-37.1 arts. 27/B–27/Ç); every general client receives the general risk disclosure form before the framework agreement (III-39.1 art. 25).
  • No machine-learning model is needed here: which test applies and how it is scored are written in the rules and in your procedure, which the board approves (III-39.1 art. 33/5), and run as code. The SLM reads and drafts; it never scores a client.

Start here

We take a month of requests with your advisers and compliance, check which carried the right test and warning, collect your procedure, forms, and product list, and test the reader on your own client files 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/suitability-assessment · Printed copy; the website is the current version.