D-CAT Group ↗ Contact ENTRDEESAZ
12 Questions Start a 4-week pilot
Start a 4-week pilot

Solutions · by industry · banking

Banking: deposits, arrears and branches read every morning, on your servers, without a credit decision

CIO and CISO (the runtime installs next to the warehouse; read-only gateways; credentials never leave the bank; row-level scope by region and segment) · head of retail banking (deposits, attrition, cross-sell) · head of credit monitoring and internal audit (arrears migration as patterns requiring review, never as decisions) · branch network director and head of digital channels (cost per transaction, wait time, onboarding). The buyer is the CIO with the retail banking head; the daily readers are segment managers and regional branch managers.

How the three agents on this page read one semantic modelOne semantic model on the left; three scheduled agents in the middle - Retail Banking Director Agent, Lending Portfolio Agent, Branch & Channel Operations Agent; each publishes findings, recommendations and a brief on the right.semantic modelimported from Axoria Data Studiodeposit_balancecasa_ratioaccount_countaccounts_closedattrition_rateRetail Banking DirectorAgentmonthly on the 2nd · 06:00FfindingRrecomm.BbriefLending Portfolio Agentmonthly on the 4th · 06:30FfindingRrecomm.BbriefBranch & ChannelOperations Agentweekly · Monday 06:00FfindingRrecomm.Bbriefevery number in a record carries an evidence address

Questions

The three questions a bank board asks every month

  • Is the deposit base moving from demand to time, and where?
  • Where is early arrears rising — which segment, sector and vintage — before it reaches the non-performing line?
  • What does a branch transaction cost now that transactions are leaving the branch?

In conversation

Sample questions people type into the conversation

deepen path — the finding's evidence is carried frozen; a new number needs a new query, and the agent says so

  • "Break the demand-deposit outflow down by segment and region."
  • "Which age band closed the most current accounts in August?"
  • "Show the 31–60 day bucket by origination vintage for construction SMEs."
  • "What does a teller transaction cost in the provincial cluster?"
  • "How current is the balance data?"

Personas

Three example agents

The agents below are examples for this industry, not a fixed set: Retail Banking Director Agent · Lending Portfolio Agent · Branch & Channel Operations Agent. Each one’s scheduled run is shown in the example sets that follow.

Set 1

Retail Banking Director Agent

Audience: the head of retail banking, segment managers, the deposits desk.

Set 2

Lending Portfolio Agent

Audience: head of credit monitoring, internal audit, the credit operations manager.

Set 3

Branch & Channel Operations Agent

Audience: the branch network director, the head of digital channels, the operations controller.

An agent's name says which measures it reads and for whom. It holds none of the role's authority: it does not decide, approve or act — and it never makes, suggests or scores a credit decision.

Agents for the topics you choose

An agent is a job description, a bound semantic model and a schedule — not code — so the topics it watches are yours to set: a payer, a service line, a campaign, a supplier. We write the first ones with you.

Build your agents with us →

Example sets

Three example sets, one run each

Each set is one agent persona and one scheduled run’s output — the brief, the recommendations and the findings they rest on — in the product’s own record envelope. The first set is open; the other two are collapsed. The model and the measures the sets assume are listed at the foot of the page.

IllustrativeSet 1Retail Banking Director Agent
Persona
Reads the deposit base, account attrition and product issuance for a mid-size retail and SME bank with 214 branches in one country; separates "deposits grew" from "demand deposits became time deposits", and "more cards" from "more cards used". · Audience: the head of retail banking, segment managers, the deposits desk. · Tone: formal, measured, executive-summary language. · Output language: en
Signals it watches
(1) CASA ratio — demand-deposit share of total deposits, month over month. (2) Account attrition by age band, segment and closure reason. (3) Product issuance by channel, with the channel share. (4) Salary-account inflows as a count of customers, not amounts. (5) Nominal growth read beside a ratio, never alone — amounts in this model are not inflation-adjusted.
Thresholds it was given
finding 15 % · critical 40 % · ratio metrics (CASA ratio, attrition rate, channel share) 8 % · minimum share 2 % of the retail base · severity rule: medium when a threshold is crossed, high when a ratio metric moves more than 1.25× its threshold or a count metric more than 1.5× the finding threshold, critical above the critical threshold; a finding with a breakdown cell takes the severity of the cell's own move; an ongoing topic keeps the severity it was given

Run header

  • Scheduled monthly on the 2nd, 06:00, Europe/Istanbul
  • Period: August 2026 vs July 2026 (month over month — 21 vs 23 business days; balance ratios are point-in-time at month-end and unaffected; issuance counts are stated per month and the day difference is noted)
  • Data complete through 2026-09-01, measured via max(BALANCE_DATE) (month-end balances)
  • Scope: the agent runs under the retail banking scope (all regions, no loan-book metrics)
  • Model: the workspace's approved model, written onto the run
  • Cost: recorded on the run
  • Amounts in nominal ₺
Brief

August 2026: Retail Deposits +6.18 % Nominal While the Demand Share Fell to 34.08 %, Current-Account Closures 0.62 % → 0.91 % Led by the 18–25 Band, Card Issuance +18.40 % on the Mobile Channel

Assumptions strip

Month-over-month comparison (August vs July 2026) was used because amounts in this model are nominal and not inflation-adjusted; ratios carry the reading. Balance ratios are point-in-time at month-end. Loan-book metrics are outside this agent's scope and do not appear.

The retail deposit base grew 6.18 % in August in nominal terms, but the growth was time deposits (+13.63 %) while demand deposits fell 5.78 %, taking the demand share from 38.41 % to 34.08 %. Current-account closures rose from 0.62 % to 0.91 % of the opening base, with the 18–25 age band closing at 1.84 % and carrying 28.70 % of the closures on 14.20 % of the base; 61.00 % of closures carry the uninformative reason customer request. Card issuance rose 18.40 %, all of it and more through the mobile channel, which now issues 66.93 % of cards; whether the cards are used is not in the model. A sentence attributing the deposit shift to the bank's time-deposit rate was returned once by the claim check; the rewrite states the shift without a cause and passed, because the rate is not in the model.

covers: Finding 1, Finding 2, Finding 3

— Run: monthly, the workspace's approved model, data complete through 2026-09-01. Guard Ledger: held 0 after one coached rewrite — candidate f-01 draft carried drawn by the new time-deposit rate, gate 4 returned it as unevidenced causality, the rewrite dropped the clause and passed; not_selected 0; duplicate 0.

Recommendation 1

Break August's demand and time deposits down by customer segment and by region side by side, and put the number of accounts that moved between the two products beside them if the model can count it

based_on: Finding 1 · owner: head of retail deposits

If the demand outflow and the time inflow sit in the same segments, the money moved inside the bank; if the segments differ, the two are separate stories. Both balance cuts are in this model at month-end grain. A count of accounts that moved between products is not a metric here today; the deposits head reads the transfer log beside the record.

The findings this rests on

Finding 1

Deposit Mix Shift: Demand-Deposit Share of Retail Deposits 38.41 % → 34.08 % (−11.26 %) While Total Retail Deposits +6.18 % MoM — Time Deposits +13.63 %, Demand Deposits −5.78 %

severity highdirection downnovelty newtrust Abreakdown DEPOSIT_TYPE = Demand

Retail deposits stood at 196.0 B ₺ at the end of August 2026 against 184.6 B ₺ at the end of July (+6.18 %, nominal). The growth is all time deposits: 129.2 B ₺ against 113.7 B ₺ (+13.63 %), while demand deposits fell from 70.9 B ₺ to 66.8 B ₺ (−5.78 %). The CASA ratio therefore moved from 38.41 % to 34.08 % (66.8 of 196.0 B ₺ against 70.9 of 184.6 B ₺), a −11.26 % change in a ratio metric, 1.41× the 8 % threshold — severity high. The shift is present in all six regions; the metropolitan cluster carries the larger absolute move. This data cannot say whether the bank's own time-deposit rate drew the money — product interest rates are not metrics in this model, and the record makes no claim about them.

Recommendation 2

Put current-account closures on a weekly watch by age band and branch region, and ask the model owner whether the closure reason codes can be made finer than "customer request"

based_on: Finding 2 · owner: head of retail banking, with the data-model owner

A reason code that holds 61.00 % of the rows explains nothing. The weekly watch will show whether the 18–25 rate is a single month or a run; the agent stamps a repeat ongoing rather than alerting again. Finer reason codes are a source-system change, not an agent change.

The findings this rests on

Finding 2

Current-Account Closures 0.62 % → 0.91 % of the Opening Base (+46.79 %) — the 18–25 Age Band Closes at 1.84 % and Holds 28.70 % of August's Closures on 14.20 % of the Base

severity criticaldirection upnovelty newtrust Abreakdown AGE_BAND = 18–25

Of 2.94 M retail current accounts open at the start of August 2026, 26,760 were closed during the month (0.91 %) against 18,230 in July (0.62 %): a +46.79 % change in a ratio metric, above the 40 % critical threshold — severity critical. The 18–25 age band closed 7,680 of its 417,500 accounts (1.84 %), twice the network rate; the band holds 14.20 % of the base and 28.70 % of the month's closures. The closure reason code is in the model and does not settle the question: 16,324 of August's 26,760 closures (61.00 %) carry customer request. No customer is named in this record; the age band is the finest cut the agent opens. This data cannot say where the closed accounts went — there is no destination-bank field, and the record does not guess.

Also published in this run, with no recommendation resting on it

Finding 3

Card Issuance +18.40 % MoM, Carried by the Mobile Channel: Digital Issuance 24,100 → 32,650 (+35.48 %) While Branch Issuance 17,100 → 16,130 (−5.67 %) — Digital Share 58.50 % → 66.93 %

severity highdirection upnovelty newtrust Abreakdown CHANNEL = Mobile

Credit cards issued rose from 41,200 in July 2026 to 48,780 in August (+18.40 %, on two fewer business days). Mobile issuance rose from 24,100 to 32,650 (+35.48 %, 2.37× the 15 % finding threshold — severity high on the breakdown cell) and branch issuance fell from 17,100 to 16,130 (−5.67 %), taking the mobile share of issuance from 58.50 % to 66.93 % (32,650 of 48,780; +14.42 % in a ratio metric). Network issuance +18.40 % on its own would read medium, 1.23× the threshold. Issuance is not use: the first-transaction (activation) metric is not in this model, so whether the added cards are being used cannot be read here.

What this run could not see.

Product interest rates, the destination of closed accounts, card activation, and inflation-adjusted growth. The loan book is outside this agent's scope by design.

IllustrativeSet 2Lending Portfolio Agent
Persona
Reads the SME and retail loan book by days-past-due bucket, segment, sector and origination vintage, and the origination pipeline by stage; names patterns requiring review for the credit-monitoring team and never a customer, never a decision. · Audience: head of credit monitoring, internal audit, the credit operations manager. · Tone: formal; every rate carries its base; every pattern is "requiring review", never a judgement. · Output language: en
Signals it watches
(1) Early-arrears migration — the 31–60 day bucket as a share of the segment book, month over month. (2) Sector concentration inside a bucket. (3) Non-performing ratio, read only beside the earlier buckets. (4) Origination turnaround by stage, with the share of applications lacking timestamps stated. (5) Prepayment and restructuring counts where the model carries them.
Thresholds it was given
finding 15 % · critical 40 % · ratio metrics (bucket shares, turnaround days) 8 % · minimum share 2 % of the segment book · severity rule: medium when a threshold is crossed, high when a ratio metric moves more than 1.25× its threshold or a count metric more than 1.5× the finding threshold, critical above the critical threshold; a finding with a breakdown cell takes the severity of the cell's own move; an ongoing topic keeps the severity it was given

Run header

  • Scheduled monthly on the 4th, 06:30, Europe/Istanbul
  • Period: August 2026 vs July 2026 (month-end buckets are point-in-time; turnaround is measured on applications disbursed in the month — 21 vs 23 business days, stated)
  • Data complete through 2026-09-01, measured via max(BUCKET_DATE); the origination stage timestamps have their own edge, 2026-09-03
  • Scope: the agent runs under the credit monitoring scope (all segments, no customer identifiers)
  • Model: the workspace's approved model
  • Cost: recorded on the run
  • Amounts in nominal ₺
Brief

August 2026: SME Early Arrears 1.84 % → 2.61 % of the Book, Concentrated in Construction & Contracting — Origination Turnaround 6.8 → 9.4 Days on a Fifth More Applications — Mortgage Prepayments Not Loaded

Assumptions strip

Month-over-month comparison was used; buckets are point-in-time at month-end. The mortgage prepayment metric returned no rows for August and is recorded as a data state, not a finding. This agent names no customer and makes no credit decision; the patterns below are for human review.

The front of the SME arrears ladder moved in August: balances 31–60 days past due rose from 1.84 % to 2.61 % of a 62.4 B ₺ book, while the 61–90 day bucket and the non-performing ratio moved little. Construction & contracting holds 22.22 % of the early bucket on 11.79 % of the book. Whether re-aged restructured loans are inside the number cannot be read from this model. At the front door, SME applications rose 21.37 % and the mean time to disbursement rose from 6.8 to 9.4 days, measured on the 86 % of applications that carry stage timestamps. The mortgage prepayment count could not be updated: its August load did not run.

covers: Finding 1, Finding 2

— Run: monthly, the workspace's approved model, data complete through 2026-09-01 (buckets) and 2026-09-03 (origination stages). Guard Ledger: held 0; data state 1 — envelope e5 (prepayment_count, retail mortgages) returned 0 rows for 2026-08, source load not run; not_selected 0; duplicate 0.

Recommendation 1

Break the SME 31–60 day bucket down by origination vintage and by sector for July and August, side by side, and hand the record to credit monitoring as a review list of segments, not of customers

based_on: Finding 1 · owner: head of credit monitoring

If the added early arrears sit in the 2025-H2 vintages, the question is underwriting of that period; if they sit across vintages in one sector, it is the sector. Both cuts are in this model at segment grain; the customer grain is not in this agent's scope, and the review that follows is a human one.

The findings this rests on

Finding 1

Early-Arrears Migration in the SME Book: 31–60 Days Past Due 1.84 % → 2.61 % of the 62.4 B ₺ Book (+41.90 %) — Construction & Contracting at 4.92 % of Its Sector Balance, 22.22 % of the Bucket on 11.79 % of the Book — Pattern Requiring Review

severity criticaldirection upnovelty newtrust Abreakdown DPD_BUCKET = 31–60 × SECTOR = Construction & contracting

The SME book stood at 62.4 B ₺ at the end of August 2026. Balances 31–60 days past due rose from 1.148 B ₺ (1.84 %) at the end of July to 1.629 B ₺ (2.61 %), a +41.90 % change in a ratio metric, above the 40 % critical threshold — severity critical. The next bucket, 61–90 days, moved from 1.14 % to 1.27 % (+11.27 %, above the 8 % ratio threshold but below critical), and the non-performing ratio from 3.12 % to 3.18 % (+1.92 %, below threshold): the movement is at the front of the ladder. Construction & contracting carries 0.362 B ₺ of the 31–60 bucket (22.22 %) on 11.79 % of the book, and its own early-arrears share is 4.92 %. This record names no customer and makes no credit judgement; it is a pattern requiring review by the credit-monitoring team. This data cannot separate a new arrear from a restructured loan that has re-aged — the restructuring flag is not in this model.

Recommendation 2

Break August's SME turnaround down by stage (application → assessment → decision → disbursement) and by branch region, and ask the model owner why 14 % of applications lack stage timestamps

based_on: Finding 2 · owner: credit operations manager, with the data-model owner

Applications grew 21.37 % and the mean duration 38.24 %; the stage cut will say where the days went. The missing timestamps are a source-system fault and belong to the model owner; until they are filled, the mean is stated on 86 % of applications, and that is written on the record.

The findings this rests on

Finding 2

SME Origination Turnaround 6.8 → 9.4 Days (+38.24 %) on Applications +21.37 % — 14 % of Applications Lack a Stage Timestamp and Are Excluded

severity highdirection upnovelty newtrust Abreakdown SEGMENT = SME

Applications for SME loans rose from 12,400 in July 2026 to 15,050 in August (+21.37 %, on two fewer business days), and disbursements from 7,930 to 8,620 (+8.70 %). Mean days from application to disbursement, measured on loans disbursed in the month, moved from 6.8 to 9.4 (+38.24 %, 4.78× the 8 % ratio threshold — severity high; the 40 % critical threshold is not reached). The stage-timestamp fields are missing on 2,107 of August's 15,050 applications (14.00 %); those are excluded from the mean and the exclusion is stated rather than filled. The record reads the volume and the duration side by side and claims nothing about approvals — the decision outcome is not a metric this agent opens. This data cannot say which stage lengthened — the per-stage cut is in the model but was not opened in this run.

Data state

Data state (not a finding): the retail mortgage prepayment_count metric returned no rows for August 2026 — the source table's monthly load did not run. No severity, no recommendation rests on it; the model owner is notified.

What this run could not see.

The restructuring flag, the per-stage duration (in the model, not opened), the decision outcome (outside scope by design), and August's mortgage prepayments.

IllustrativeSet 3Branch & Channel Operations Agent
Persona
Reads teller transactions, branch operating cost, queue wait time and the mobile onboarding funnel for the same bank; separates "transactions left the branch" from "the branch got more expensive per transaction", and "fewer completions" from "a different app version". · Audience: the branch network director, the head of digital channels, the operations controller. · Tone: plain and operational. · Output language: en
Signals it watches
(1) Cost per teller transaction by branch cluster. (2) Low-volume branches as a share of the network. (3) Queue wait time by cluster, beside transaction volume. (4) Mobile onboarding completion by step and app version. (5) Channel migration — the share of transactions by channel.
Thresholds it was given
finding 15 % · critical 40 % · ratio metrics (cost per transaction, wait minutes, completion rate) 8 % · minimum share 2 % of branches or of onboarding starts · severity rule: medium when a threshold is crossed, high when a ratio metric moves more than 1.25× its threshold or a count metric more than 1.5× the finding threshold, critical above the critical threshold; a finding with a breakdown cell takes the severity of the cell's own move; an ongoing topic keeps the severity it was given

Run header

  • Scheduled weekly, Monday 06:00, Europe/Istanbul
  • Period: August 2026 vs July 2026 (21 vs 23 business days — transaction counts are affected and the difference is stated; cost per transaction and rates are not)
  • Data complete through 2026-09-06, measured via max(TRANSACTION_DATE), 6 hours ago; branch cost postings close monthly and their edge is 2026-08-31
  • Scope: the agent runs under the branch operations scope (all branches, no balances, no loan metrics)
  • Model: the workspace's approved model
  • Cost: recorded on the run
  • Amounts in nominal ₺
Brief

August 2026: Cost per Teller Transaction 211.6 ₺ → 241.9 ₺ as Transactions Left the Branch, Metropolitan Queue Wait 11.2 → 15.8 Minutes, Mobile Onboarding Completion 71.30 % → 63.90 % on App Version 6.2

Assumptions strip

Month-over-month comparison was used (21 vs 23 business days; counts are affected, rates are not). Branch cost closes monthly and its edge is 2026-08-31; transactions run to 2026-09-06. Queue wait is measured only where a ticket dispenser exists — the provincial cluster is absent from that metric by construction.

Teller transactions fell 11.41 % in August while branch operating cost held, so each transaction cost 241.9 ₺ against 211.6 ₺ in July; thirty-seven provincial branches, 17.29 % of the network, handled fewer than 8,000 transactions in the month. The metropolitan branches moved the other way on time: queue wait rose from 11.2 to 15.8 minutes on 8.02 % fewer transactions. In the app, onboarding starts grew 7.94 % but completion fell from 71.30 % to 63.90 %, and the fall sits on version 6.2 at the identity-verification step. The sentence attributing the waits to understaffing was held; the roster is not in the model.

covers: Finding 1, Finding 2, Finding 3

— Run: weekly, the workspace's approved model, data complete through 2026-09-06 (transactions) and 2026-08-31 (cost). Guard Ledger: held 1 — candidate f-02 draft, reason unevidenced_causality (gate 4): clause because the branches were understaffed had no roster envelope, one coached rewrite kept a weaker form of it, then held; duplicate 1 — candidate f-04 (cost per transaction by cluster) merged into Finding 1 on fingerprint cost_per_txn|2026-08|provincial; not_selected 0.

Recommendation 1

Break cost per teller transaction down by branch cluster and by branch for July and August, with each branch's queue wait beside it where the wait is measured

based_on: Finding 1, Finding 2 · owner: branch network director

Thirty-seven branches carry the low-volume end and forty-one carry the long-wait end; the branch cut will show whether any branch is on both lists. Cost is posted monthly and transactions daily; the two are read on their own grain. The staffing roster is not in this model, and the sentence the agent drafted about understaffing was held after one rewrite.

The findings this rests on

Finding 1

Cost per Teller Transaction 211.6 ₺ → 241.9 ₺ (+14.32 %) as Transactions Fell −11.41 % MoM While Branch Operating Cost Held (+1.27 %) — 37 of 214 Branches (17.29 %) Below 8,000 Transactions in the Month

severity highdirection upnovelty newtrust Abreakdown BRANCH_CLUSTER = Provincial

Teller transactions fell from 4.82 M in July 2026 to 4.27 M in August (−11.41 %; two fewer business days account for part of it, and the per-day count still fell). Branch operating cost moved from 1.02 B ₺ to 1.033 B ₺ (+1.27 %, below threshold), so the cost per teller transaction rose from 211.6 ₺ to 241.9 ₺ (1.033 B ₺ over 4.27 M against 1.02 B ₺ over 4.82 M; +14.32 %, 1.79× the 8 % ratio threshold — severity high). Thirty-seven branches, 17.29 % of the network and all in the provincial cluster, recorded fewer than 8,000 teller transactions in August. This data cannot say what those branches do besides teller work — advisory and sales activity are not counted in this star.

Finding 2

Queue Wait at the 41 Metropolitan Branches 11.2 → 15.8 Minutes (+41.07 %) While Their Transactions Fell −8.02 %

severity criticaldirection upnovelty newtrust Abreakdown BRANCH_CLUSTER = Metropolitan

Mean queue wait at the 41 metropolitan branches (19.16 % of the network) rose from 11.2 minutes in July 2026 to 15.8 in August (+41.07 %, above the 40 % critical threshold — severity critical) while their teller transactions fell from 1.31 M to 1.205 M (−8.02 %). Longer waits on fewer transactions is the shape a staffing change would leave, but the staffing roster is not in this model, and the record does not infer one. The wait metric comes from the queue-ticket system and covers only branches with ticket dispensers — the provincial cluster is absent from it by construction.

Recommendation 2

Put mobile onboarding completion on a daily watch by app version and by step, and hand the identity-verification step's rate on version 6.2 to the digital channels team

based_on: Finding 3 · owner: head of digital channels

The version cut isolates the drop; the step cut locates it. Both are in this model at day grain. Whether the step's fall is the app or the identity provider cannot be read here — the provider's response codes are not in the model.

The findings this rests on

Finding 3

Mobile Onboarding Completion 71.30 % → 63.90 % of Starts (−10.38 %) on Starts +7.94 % — App Version 6.2 Completes at 61.10 % Against 72.40 % for Version 6.1

severity highdirection downnovelty newtrust Abreakdown APP_VERSION = 6.2

Onboarding starts in the mobile app rose from 214,000 in July 2026 to 231,000 in August (+7.94 %), while completions moved from 152,582 to 147,609, taking the completion rate from 71.30 % to 63.90 % (−10.38 %, 1.30× the 8 % ratio threshold — severity high). The app-version cut is in this model and is where the move sits: 173,800 of August's starts ran on version 6.2 and 106,196 of them completed (61.10 %); 57,200 ran on version 6.1 and 41,413 completed (72.40 %). The step cut shows the drop at identity verification. The release date of 6.2 is not a fact in this model; the record states the version difference and nothing about the release.

What this run could not see.

The staffing roster, non-teller activity in the low-volume branches, the identity provider's response codes, and the release date of app version 6.2. Balances and the loan book are outside this agent's scope by design.

Every number carries an address

In the product each figure binds to an evidence address — envelope, row, cell, check digit — and a number without one is held, not published. The anatomy of a record →

Derived figures show their operation

A change, a share or a gap is computed by the engine and carries its operands; the model never divides. How every number is proved →

AI governance is the publication layer

Eight deterministic gates, the trust tier, badged interpretation, an assumptions strip that cannot be switched off, and the Guard Ledger. Where the model can and cannot reach →

FAQ

Four questions this page is usually asked

01Does anything leave the bank?

No. The runtime installs next to the warehouse; gateways are read-only; credentials stay on the bank's servers; the model receives sealed query results, never tables, and never a customer identifier.

02Does the agent score customers or flag them for credit action?

No. It reports bucket shares by segment, sector and vintage as patterns requiring review. It cannot open the customer grain, and it never labels a pattern with a legal or regulatory word.

03How is row-level scope handled?

Each agent runs under a named scope frozen onto every record — retail banking, credit monitoring, branch operations in the sets above — and the scope is re-checked when a person opens the record.

04What if the source load fails?

The metric returns no rows and the run records a data state, visibly, with no severity and no recommendation on it — never a zero, never a finding.

Free data discovery study

See which of these signals your own warehouse can carry.

Send us the name of your data platform and the domain you argue about most. We read your semantic model with you and answer in writing: which of the signals on this page your model carries today, which need a change in Axoria Data Studio, and what a four-week pilot would measure. No cost, no sequence — a person replies within two working days.

Request a free data discovery study → Start a 4-week pilot

What the examples assume

The data, the model and the measures behind the three sets

Nothing in a set rests on a source outside the model described here. Open what you want to reconcile.

01Assumed data sources — data a bank already has

Core banking (accounts, balances, transactions by channel, openings and closures with reason codes), the loan book (balances by product, segment, sector and origination vintage, days-past-due buckets), the origination system (application, decision and disbursement timestamps), the card system (issuance by channel), the branch queue system (wait times), the mobile-app event log (onboarding steps by app version), branch cost postings. Typical warehouse grain: one row per account per day for balances, one per transaction, one per loan per month for buckets.

02Assumed semantic model — the minimum for the three personas
Measures
deposit_balance (demand, time) · casa_ratio (engine-derived) · account_count · accounts_closed · attrition_rate (engine-derived) · cards_issued · loan_balance · balance_by_dpd_bucket · dpd_bucket_share (engine-derived) · npl_ratio · applications · disbursements · turnaround_days (mean, from stage timestamps) · teller_transactions · branch_operating_cost · cost_per_transaction (engine-derived) · queue_wait_minutes (mean) · app_onboarding_starts · app_onboarding_completions · completion_rate (engine-derived).
Dimensions
date · branch · region · branch cluster (metropolitan / provincial) · customer segment (retail / SME) · age band · product · channel (branch / mobile / web / call centre) · loan sector · origination vintage · days-past-due bucket · closure reason code · app version · onboarding step. Marked personal and never broken down by: customer, account number, relationship manager.
Data edge
measured by the engine via max(BALANCE_DATE) for balances and max(TRANSACTION_DATE) for transactions; the model declares both. Amounts are nominal ₺ — the model carries no inflation adjustment, so the agents compare month over month and read ratios, and say so.
03Assumed KPIs — what bank executives track
KPIDefinitionUnitTypical bandUsual breakdown
Net interest margin(interest income − interest expense) / earning assets%US banks ≈ 3.0–3.4 % recently [vendor citing regulator data]segment · product
Cost-to-income ratiooperating expense / operating income%< 60 % considered good [vendor]business line · branch
Return on equity / assetsnet income / equity or assets%US commercial banks ROE ≈ 11–13 % [vendor]segment
Non-performing loan rationon-performing loans / gross loans%no single band; regulator dashboards publishproduct · segment
Cost of riskloan-loss provisions / average loansbpscompany-specificproduct · vintage
CASA ratiocurrent + savings deposits / total deposits%company-specificbranch · segment
Loan-to-deposit ratioloans / deposits%company-specificentity
Capital adequacy (CET1)regulatory capital / risk-weighted assets%regulatory minima plus buffersentity
SME origination turnaroundapplication to disbursementdayscompany-specificproduct · channel
Products per customeractive products / customersratiocompany-specificsegment
Digital active users / digital sales shareshare of customers or sales via digital%company-specificchannel
Customer attritionclosed accounts / opening base%company-specificsegment · branch
Net promoter scoresurveyindexcompany-specificsegment · channel
Days-past-due bucket migrationbalance in a bucket / segment book, month over month%company-specificsegment · sector · vintage

Related

Where these pages lead

The departments these personas belong to

On this site