Use casesFinancial institutions & fintech

Risk-based pricing with price response

Your price covers risk and capital, yet you lose good customers over a few basis points and discount those who would have signed anyway.

Kinds of AI
Models
Who decides
Pricing committee
FIG. 1 · Price cardIllustrative
Personal loan · 36 monthsSalaried, bureau band B, existing customer

Model · candidate rates

RateExpected profit per offerP(accept)
Floorthin marginlowhigh
Floor + smallhighestmedium-high
Floor + midcompetitor quotes cluster herelowermedium
Floor + highadverse-selection zone; PD riseslowerlow

RecommendedFloor + small: the highest expected profit per offer.

Code · policy checks

  • metSegment-gap rulemet
  • metNo protected attributemet
  • metRenewal ≤ new-customer pricemet
  • Model
  • Code

Written limitRelationship manager discretion: a small band without approval; below it, a named approver.

DecidesPricing committee

1. The desk

Today

  1. PriceThe rate grid reflects expected loss and funding, but not how each segment reacts to price.
  2. DiscountBranch discretion leaks margin; discounts go to customers who would have accepted the list rate.
  3. LearnDeclined quotes are rarely stored, so nobody can say what a lower price would have won.
  4. DecideThe pricing committee signs a grid that cannot show how each segment would respond to a different rate.

With the system

  1. PriceThe cost floor is computed from funding, risk, capital, and cost to serve; the take-up and profit models show acceptance and expected profit at each candidate rate.Model
  2. DiscountThe optimizer recommends a rate and a discretion band per segment, inside the floor, the legal caps, and the policy checks; below-floor exceptions need a named approver.Model
  3. LearnEvery quote is logged, declined ones included; the take-up model is re-estimated on them, and realized take-up is checked against predicted each month.Model
  4. DecideThe pricing committee signs off the grid and the objective, as before; the relationship manager moves only inside the band.Person decides

What lands on the desk

The pricing committee sees, for one segment, how acceptance and expected profit move across candidate rates, where the floor and the ceiling sit, and which policy checks the recommended rate passes. The explanation line is what the customer can be told.

The pricing committee (the asset-liability committee or the credit committee) signs off the grid and the objective function; validation reviews the take-up and PD models. The RM moves only inside the discretion band, and below-floor exceptions need a named approver. The optimizer never prices an individual outside the approved grid.

FIG. 2 · Price cardIllustrative
Personal loan · 36 monthsSalaried, bureau band B, existing customer

Code · floor and ceiling

  • Cost floorFunding + expected loss + capital + operating cost: the floor rate
  • Policy ceilingLegal cap and internal max

Model · candidate rates

RateExpected profit per offerP(accept)
Floorthin marginlowhigh
Floor + smallhighestmedium-high
Floor + midcompetitor quotes cluster herelowermedium
Floor + highadverse-selection zone; PD riseslowerlow

RecommendedFloor + small: the highest expected profit per offer.

Explanation

Price reflects your risk band, term, and amount; not your tenure.

Code · policy checks

  • metSegment-gap rulemet
  • metNo protected attributemet
  • metRenewal ≤ new-customer pricemet
  • Model
  • Code

Written limitRelationship manager discretion: a small band without approval; below it, a named approver.

DecidesPricing committee

2. The system

For your technical team — skip it if you like.

How it works

In

  • Funding curve (transfer price)
  • PD and LGD, existing or built
  • The capital model
  • Cost to serve

  1. Code · calculates

    Cost-floor calculation

    Funding, expected loss, capital, operating cost, added in code

    Reads

    • Everything under In

    Out

    • The floor rate
    • Per EBA/GL/2020/06 section 6

  2. Model · estimates

    Price-response (take-up) model

    Logistic or GBM, monotone in price; exogenous price variation

    Reads

    • Quote history, declined and not-taken too
    • Rate, term, amount, channel, risk band
    • Central bank weighted average rates

    Out

    • Probability of acceptance
    • By rate, term, amount, risk, and channel

  3. Model · estimates

    Profit model

    Reads

    • Take-up
    • PD conditional on price (adverse selection)
    • The floor

    Out

    • Expected profit per offer at each rate

  4. Model · estimates

    Constrained optimizer

    Maximizes expected profit inside the policy constraints

    Reads

    • The profit curve
    • Floor, legal caps, segment-gap rule
    • No protected attributes or proxies
    • Renewals not priced above new customers

    Out

    • A recommended rate per segment
    • An RM discretion band

  5. Decision

    Pricing committee

    Reads

    • The proposed grid

    Out

    • The pricing committee signs off the grid and the objective
    • The relationship manager moves only inside the band

None: the model proposes inside the policy, 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

Price-response and profit models

  1. Target

    Accept or not accept, per quote; PD modeled jointly, because higher prices can attract worse risk.

  2. Data

    The hard part is declined quotes. If they are not stored, step one is logging every quote; where past prices were set by risk alone, price variation not driven by risk comes from randomized rate bands, rate-card change dates, or branch discretion.

  3. Method and challenger

    Logistic against GBM with a monotone price constraint; elasticity by micro-segment; re-estimated often under high inflation.

  4. Validation evidence

    Realized against predicted take-up by band, a profit back-test, fairness checks by segment, and scenario and sensitivity tests on funding and rate caps.

  5. Monitoring

    Champion/challenger rate grids; monthly realized take-up and margin by segment.

3. Value and conditions

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

Where the value comes from

  1. Margin on accepted loans, and less discount leakage.
  2. Conversion of good-risk applicants.
  3. Risk mix.

What your institution measures

  • Risk-adjusted return on capital (RAROC) per segment
  • Acceptance rate by risk band
  • Discount frequency and size
  • Share of exceptions
  • Realized against predicted take-up
  • Price complaints
Published elsewhere
In a randomized field experiment at a South African consumer lender, cutting rates below the lender's standard rates raised demand only gently, but demand sensitivity rose sharply at prices above them; the effect of loan maturity on demand was large relative to price sensitivity.

Karlan and Zinman, "Credit Elasticities in Less-Developed Economies", American Economic Review 2008, as summarized by Innovations for Poverty Action

Measured in the Decision Blueprint

Margin and acceptance by segment from your quote history set the baseline; your institution sets the target, confirmed in a price test.

Conditions

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

  • The EU Consumer Credit Directive 2 (from 20 November 2026) requires disclosure of personalized pricing and sets creditworthiness rules; pricing that uses a natural person's creditworthiness is high-risk under the EU AI Act.
  • Inertia pricing of renewing customers draws supervisory scrutiny.
  • In Türkiye, the central bank's (TCMB) maximum rates on cards and overdrafts and the loan-growth caps are constraints, not objectives, and their levels change often.
  • The biggest practical risk is missing declined-quote data.

Start here

We check what your quote history holds, declined quotes included, rebuild the cost floor from your funding and capital inputs, and find where past prices varied for reasons other than risk.

Request a Decision BlueprintWhat the Decision Blueprint covers
  • Models

    Application scorecard for new-to-bank applicants

    TodayMost applicants are new to you, the bureau file is thin, and your only outcomes are for the people you approved.

    With the systemA credit scoring model beside today's rules: a scorecard, with the default probability by band.

    Credit policy committee decides

  • 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

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/risk-based-pricing · Printed copy; the website is the current version.