Redroot
by LapisAI Book a demo
Redroot (akar data) · governed bridge, warehouse to AI

Ask in chat. Get the certified number.

Redroot sits between your data warehouses and the AI tools your teams already use. Data owners certify each metric once. Redroot answers with it in Claude, Copilot and ChatGPT, and shows where every number came from.

Book a demo See how it works
Certified Every answer carries its definition, owner, filters, access policy and source.
Assistant Redroot connected
Berapa NPL per cabang Q3 dibanding Q2?

NPL gross wilayah Anda naik menjadi 2,41% di Q3, naik 0,18 pp dari Q2. Cirebon dan Bandung tertinggi.

Redroot
NPL gross, Q3 2026 2,41% +0,18 ppdibanding Q2 2026 (2,23%)
Q3 2026 vs Q2 2026Per cabangDifilter ke wilayah Anda: Jawa
Tersertifikasi·Pemilik: Credit Risk·2 jam lalu · 08.00
ReceiptID kueri q_7f3a91c2
Metrik
NPL grossTersertifikasirisk.npl_gross_ratio · v4

Baki debet kredit dengan kolektibilitas 3 sampai 5, dibagi total baki debet kredit, pada akhir kuartal.

Disertifikasi oleh Dewi Lestari, Credit Risk · 2 Okt 2026
Filter
PeriodeQ3 2026 (1 Jul – 30 Sep) dibanding Q2 2026 (1 Apr – 30 Jun)
PerCabang
KebijakanDifilter ke wilayah Anda: JawaPeran Regional credit manager · region_code = 'JAWA'
Sumber SQL Server · BANKDW_PROD bankdw.risk.loan_collectibility_q · bankdw.core.dim_branch 8 baris · 1,2 dtk · akun layanan hanya-baca
q_7f3a91c2 · 1.2 s · logged to audit
The receipt opens in place: definition, filters, access policy, source.
Answers inside your AI tools.
Reads from your warehouses.
ClaudeSQL ServerMicrosoft CopilotOracleChatGPTPostgreSQLCustom agents over MCPMicrosoft FabricAzure SynapseSnowflakenextBigQuerynextDatabricksnextMaxComputenext
01 · Show the receipt

Every number can be traced in one tap.

A careful finance director should be comfortable acting on the number in front of them. The receipt opens in place, in chat and in the console, with the same structure every time.

  • Definition and ownerThe business definition, the certified version, and who certified it and when.
  • Every filter, including your access policyPeriod, grouping, follow-up filters and the row filter your role applied.
  • The SQL that ranCollapsed by default, copyable, against a named source warehouse.
  • A query IDThe same ID appears in the audit log, so security can trace who saw what.
ReceiptQuery ID q_7f3a91c2
Metric
NPL grossCertifiedrisk.npl_gross_ratio · v4

Outstanding balance of loans in collectibility 3 to 5, divided by total outstanding loans, at quarter end.

Certified by Dewi Lestari, Credit Risk · 2 Oct 2026
Filters
PeriodQ3 2026 (1 Jul – 30 Sep) vs Q2 2026 (1 Apr – 30 Jun)
Grouped byBranch
PolicyFiltered to your region: JawaRole Regional credit manager · region_code = 'JAWA'
Source SQL Server · BANKDW_PROD bankdw.risk.loan_collectibility_q · bankdw.core.dim_branch 8 rows · 1.2 s · read-only service account
T-SQL 
-- Redroot · risk.npl_gross_ratio v4 · certified
SELECT b.branch_name AS branch,
  SUM(CASE WHEN l.coll >= 3 AND l.period = '2026-Q3' THEN l.outstanding_idr ELSE 0 END)
    / NULLIF(SUM(CASE WHEN l.period = '2026-Q3' THEN l.outstanding_idr ELSE 0 END), 0) AS npl_q3,
  SUM(CASE WHEN l.coll >= 3 AND l.period = '2026-Q2' THEN l.outstanding_idr ELSE 0 END)
    / NULLIF(SUM(CASE WHEN l.period = '2026-Q2' THEN l.outstanding_idr ELSE 0 END), 0) AS npl_q2
FROM bankdw.risk.loan_collectibility_q AS l
JOIN bankdw.core.dim_branch AS b ON b.branch_id = l.branch_id
WHERE l.period IN ('2026-Q2', '2026-Q3')
  AND b.region_code = 'JAWA'  -- policy: regional_credit_manager
GROUP BY b.branch_name
ORDER BY npl_q3 DESC;
This query ID matches the entry in the audit log.
02 · From warehouse to answer

One subject area at a time.

Data engineers connect and model. Data owners certify. Everyone else asks in the tool they already use.

Redroot console · Connections › New
Add connection wizard, test step: network, authentication, read-only role and schema checks running.
Model Builder imports: Power BI measures and dbt metrics mapped to the model, with conflicts marked.
Review and certify: the NPL gross change in plain language, the definition diff with AI-proposed lines marked, and the evaluation result.
hanya Jawa Barat
Redroot
NPL gross, Jawa Barat, Q3 2026 2,73% +0,28 ppdibanding Q2 2026 (2,45%)
Q3 2026 vs Q2 2026Per cabangJawa BaratDifilter ke wilayah Anda: Jawa
Tersertifikasi·Pemilik: Credit Risk·2 jam lalu · 08.00
03 · AI proposes, people certify

Nothing AI wrote is quoted as fact.

Model Builder drafts metrics, dimensions and synonyms in English and Indonesian. Each stays labelled AI draft until a named person certifies it, and the stamp follows the metric everywhere it is used.

Before a model version goes live, an evaluation gate runs your gold questions. Below the threshold, publishing is blocked until the model is fixed, or an owner overrides with a reason that is logged.

Cost of credit (CoC)Certified

Loan-loss provision expense over the last 12 months, divided by average outstanding loans over the same months.

metric: risk.cost_of_creditAI
numerator: SUM(provision_expense_idr)AI
denominator: AVG(outstanding_idr)AI
synonyms_id: [biaya kredit, CoC]AI
Certified by Dewi Lestari, Credit Risk · 9 Oct 2026, 10:42 · evaluation 97.9%
RedrootOutside certified scope

I can't answer that from certified data yet.

There is no certified forecast for NPL. Redroot answers from certified metrics only, and forecasts need a method and an owner before they can be certified.

Closest certified metrics
NPL grossNPL netLoan at risk (LaR)
What would be needed

A forecast metric for NPL with a defined method and horizon, owned and certified by Credit Risk.

Your question is sent with the request.
Request logged G-0142Logged as a gap for Credit Risk with your original question. You will be notified when the metric is certified.What will NPL be per branch in Q4?
04 · Refuse helpfully

When it can't answer, it says so.

A question outside certified scope gets a plain reason, the nearest certified metrics, and what would be needed to answer it. One tap logs the question as a gap, so data owners see what people are asking for.

05 · Govern

Access follows the person asking.

  • Roles from your identity providerMap Entra ID or Okta groups to roles. No separate user list.
  • Row filters and column masksApplied to every query in chat and console. Personal data stays masked.
  • View as, before you publishSee exactly what a person would get, beside the unrestricted view.
  • Audit log to your SIEMEvery question, policy and query ID, kept for 400 days and streamed.
Redroot console · Access › View as
Access and policies, View as: the answer Bayu Saputra, a branch manager, would get, beside the unrestricted admin view. Rows from other branches are filtered out and the debtor name and NIK columns are masked.
Audit log
Audit log with a query open: user, AI tool, role, policies applied and the same receipt the user saw in chat.

Start with one warehouse and one subject area.

Connect read-only, certify the metrics one team asks about most, and let that team ask in Claude, Copilot or ChatGPT. Add subject areas as owners certify them.

Book a demo Deployment and security