Use casesFinancial institutions & fintech

  • Bank
  • Participation bank

Deposit pricing and balance retention

Every maturity is a chance to lose the balance or overpay to keep it; branches ask for special rates every day, and nobody can say which depositors would have stayed anyway.

With the systemA deposit rate grid that shows, by segment and tenor, how the rate moves balances kept at maturity and their cost.

Who decides
Asset-liability committee (ALCO)

Branch special-rate request · 4 November 2026

Time deposit · retail, large depositor · maturing this week

Long-standing customer. The branch manager writes: another bank has offered a higher rate; we request a matching rate to keep the balance.

→ above the branch band · the named approver decides

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

1. The work

Today

  1. PriceTreasury proposes deposit rates by tenor and segment from the funding need, the transfer price, and competitors' rates; the asset-liability committee (ALCO) sets the grid without knowing how each segment reacts to a rate change.
  2. ExceptionBranches ask the treasury pricing desk for special rates for large depositors before maturity; each request is judged on the phone or by email, and the extra rate often goes to depositors who would have rolled over anyway.
  3. LearnWhether a deposit rolled over, left, or moved to another product is rarely tied back to the rate offered, so nobody knows what a lower or higher rate would have kept.
  4. DecideThe ALCO signs a grid that cannot show how balances kept at maturity, and the cost of funds, would change at a different rate.

With the system

  1. PriceThe funding value of each tenor is computed in code from the transfer price, the reserve requirement cost, and the cost to serve; the retention models show, for each segment, how balances kept at maturity move with the rate.Model
  2. ExceptionFor a branch request, the model shows how likely this balance is to stay at the grid rate and at the rate asked; inside the band the branch decides, above it the request goes to the named approver on the pricing desk.Model
  3. LearnEvery maturity is logged with the rate offered and what the depositor did; the models are re-estimated on them, and a control group kept on today's grid shows what the new grid changed.Model
  4. DecideThe ALCO signs off the grid, the bands, and the objective, as before; no rate leaves the grid without the named approver.Person decides

What lands on the desk

The pricing desk sees, for one depositor, how likely the balance is to stay at the grid rate, at the top of the band, and at the rate asked, and what each costs against the funding value. The approver decides; the card only shows what the extra rate is likely to buy.

The asset-liability committee (ALCO) signs off the rate grid, the branch bands, and the objective; validation reviews the models. The branch moves only inside the band, and a rate above it needs the named approver on the treasury pricing desk. The optimizer never sets a depositor's rate outside the approved grid and never changes a rate already agreed.

FIG. 1 · Exception cardIllustrative
Time deposit · short tenor · maturing this weekRetail, large depositor; branch asks for a special rate

Code · funding value and grid

  • Funding valueTransfer price for the tenor, reserve cost, cost to serve: what the balance is worth to treasury
  • Grid rateThe ALCO's rate for this segment and tenor, and its branch band

Model · rate against balance kept

RateExpected funding valueP(roll over)
Grid ratemoderatemedium
Grid + smalltop of the branch bandhighesthigh
Rate askedabove the band; little more retentionlowerhigh
Below gridbalance likely to leavelowlow

SuggestedGrid + small: the balance is likely to stay at the top of the band.

Code · policy checks

  • not metRate asked within the branch bandnamed approver
  • metSegment-gap rulemet
  • metFunding target for the tenormet
  • metNo protected attributemet

Reasons

Long relationship, balance rolled over at every recent maturity, salary paid into the account; the gap to competitor rates matters less in this segment.

  • Model
  • Code

Written limitThe branch moves inside the band; above it, the named approver on the pricing desk decides; the ALCO signs the grid.

DecidesAsset-liability committee (ALCO)

2. The system

For your technical team — skip it if you like.

How it works

In

  • Transfer price curve from treasury
  • Reserve requirement cost
  • Cost to serve by channel

  1. Code · calculates

    Funding value by tenor

    Transfer price, reserve cost, and cost to serve, added in code

    Reads

    • Everything under In

    Out

    • The funding value of each tenor
    • Each component shown

  2. Model · scores

    Rollover model

    Logistic or survival model vs. GBM, monotone in the rate gap

    Reads

    • Maturities with the rate offered
    • Market and competitor rates at the time
    • Segment, tenor, balance, relationship

    Out

    • Probability the deposit rolls over
    • Share of the balance kept

  3. Model · scores

    New-money response model

    Balance flows against the rate gap; champion and challenger

    Reads

    • Inflows and outflows by segment
    • Rate changes on your grid and the market's

    Out

    • Expected new balances at each rate
    • By segment and tenor

  4. Model · scores

    Grid optimizer

    Lowest cost of funds for the balances the ALCO targets

    Reads

    • Retention and response curves
    • Funding targets by tenor and currency
    • Segment-gap rule; no protected attributes
    • The exception band

    Out

    • A rate per segment and tenor
    • A branch discretion band

  5. Decision

    Asset-liability committee (ALCO)

    Reads

    • The proposed grid

    Out

    • The ALCO signs off the grid and the objective
    • The branch moves only inside the band
    • Above it, the named approver on the pricing desk

None: a rate grid the ALCO writes plus a model is enough; branch requests travel through your own approval flow, and no agent and no language model are needed.

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

Rollover and new-money response models

  1. Target

    Rolled over or not at maturity, and the share of the balance kept; new balances by segment as rates move.

  2. Data

    Maturities with the rate offered, what the depositor did, and the market rates at the time; branch exception requests and their outcomes. The hard part is that rates were set by segment, so variation comes from grid change dates, branch exceptions, and price tests the ALCO signs off.

  3. Method and challenger

    Logistic or survival model against GBM with a monotone constraint on the rate gap; validated out of time; re-estimated often while policy rates move fast.

  4. Validation evidence

    Realized against predicted rollover by band, a back-test of balances through past rate changes, calibration, stability by segment, and documentation for your independent validation.

  5. Monitoring

    A control group of branches or segments kept on today's grid; monthly rollover, cost of funds, and exceptions by segment against it.

3. Value and conditions

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

Where the value comes from

  1. Cost of funds by tenor and segment.
  2. Balances kept at maturity.
  3. Fewer special rates paid to depositors who would have stayed.

What your institution measures

  • Cost of funds by tenor and segment
  • Rollover at maturity by segment
  • Exception frequency and size
  • Realized against predicted rollover
  • Balance outflows after rate changes

Measured in the Decision Blueprint

Cost of funds and rollover by segment from your deposit history set the baseline, against a control group kept on today's grid; your institution sets the target.

Conditions

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

  • The central bank (TCMB) sets reserve requirements on deposits and changes them by announcement (for example its macroprudential press release 2026-26 of 1 July 2026 on foreign-currency deposits): the funding value reads the current rules as inputs, and their levels stay off the page.
  • The grid prices new and maturing deposits only; the optimizer never changes a rate already agreed, and every rate above the branch band goes to the named approver.
  • The biggest practical risk is history with little rate variation, because rates were set by segment: the models lean on grid change dates, branch exceptions, and price tests the ALCO signs off, and a control group keeps the effect measured, not assumed.
  • Loyal depositors who never ask can end up with the lowest rates; the grid checks maturing balances against the rate offered for new money, as your policy writes.

Start here

We check what your deposit history holds, maturities and the rates offered included, rebuild the funding value from your transfer price and reserve inputs, and find where past rates varied for reasons other than segment.

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.

  1. You send a short descriptionNo confidential records, credentials, or production data
  2. The engineer checks whether it fits, and what it would needA model, a small language model, an agent, or none
  3. A proposed first step, or an honest “not a fit”A technical conversation with the engineer
Start with a Decision BlueprintTalk to the engineer

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