The problem

One customer, four versions

1

The same customer looks different in every system

Rachel exists in four bank systems. Each was built for its own job, so each stores her a little differently. No single system holds the whole truth — and that quietly causes duplicate customers, missed risk and slow service.

Rachel, as four systems see her

Core Banking
Name
Rachel A Smith
Email
missing
Status
Active
Customer Service
Name
Rachel Smith
Email
rachel.smith@…
Phone
378-613-7157
Identity (KYC)
Name
RACHEL SMITH
KYC
Verified
Expiry
Jun 2027
Risk System
Name
R. Smith
Risk
Watch
PEP
No
4 records could look like 4 different people — but they're all one Rachel.
So what

Fragmented data isn't a tech detail — it's business risk. The job of this platform is to turn these four into one trusted answer, without losing where any fact came from.

2

Step 1 · Load all four records, exactly as they arrived

The platform reads all four source systems and copies every record into one place. Crucially, it changes nothing and merges nothing yet — it just labels each record with the system it came from. The originals are kept as evidence the whole way through.

What happens behind the screen

4 source systemsCore Banking · CRM · KYC · Risk
Loaded & labelledtagged by source
Nothing mergedevery original kept
Why keep the originals? Later, if anyone asks "where did this value come from?", the platform can point to the exact source record. That's what makes the final answer explainable instead of a black box.
So what

Think of four boxes of documents arriving. We label every page by which box it came from and file it carefully — we don't shred anything.

3

Step 2 · Check whether each record can be trusted

Before any record influences a customer profile, it gets a quick health inspection — a set of simple checks. Records that fail are quarantinedQuarantined: set aside in a holding area so a faulty record can't mix into the trusted result. It isn't deleted — it waits for a fix or review., not silently accepted.

Eight plain-English checks (sample scores)

88%
97%
91%
99%
98%
Rachel's case: her Core-Banking record is missing an email, so it scores lower and gets flagged. A low score doesn't mean she's a bad customer — it means that record needs attention before we trust it.
So what

Like a vehicle inspection: passing means "safe to continue"; failing means "fix before you rely on it." Bad data never quietly reaches the final record.

4

Step 3 · Recognise records that belong to one person

This is called entity resolutionEntity resolution: deciding which records describe the same real-world person — even when there's no shared ID and the details don't match exactly.. It doesn't rely on one exact identifier. Instead it weighs several clues — name, birth date, phone, email — like a careful investigator, and produces a confidence level.

Example: do these two records match?

Rachel A Smith
Email: missing
DOB: 14 Mar 1986
Phone: 378-613-7157
Source: Core Banking
92%likely same person
RACHEL SMITH
Email: rachel.smith@…
DOB: 14 Mar 1986
Phone: 3786137157
Source: KYC
96%
100%
100%
So what

People rarely appear identically across systems. Combining several clues turns "four maybe-people" into "one Rachel, counted once."

5

Step 4 · When the machine isn't sure, a person decides

This is stewardshipStewardship: trained people who review the uncertain cases the automation flags, and make the final call. The machine handles the obvious cases; humans handle the doubt.. Obvious matches are automatic. Borderline ones are sent to a human review queue — so two different people are never merged by accident.

What a reviewer sees

Record A · CRM
Name: Rachel Smith
Email: rachel.smith@…
Address: 14 Lake Road
78%needs review
Record B · KYC
Name: Rachael Smith
Email: r.smith@…
Address: 14 Lake Rd
✓ APPROVE — same person
✕ REJECT — keep separate
So what

Automation does the heavy lifting; people own the judgement calls. That human gate is what keeps an automated system safe and trusted.

6

Step 5 · Pick the best value for every field

Now the grouped records collapse into one. This is survivorshipSurvivorship: for each field, choosing which source's value "survives" into the final record — based on which system is most trustworthy for that field, how recent it is, and any rules that protect it.: the Golden Record isn't the newest row or an average — each field is chosen on purpose, with a reason recorded.

Rachel's Golden Record — note where each value won

RS
Rachel Smith✓ Golden Record · 93% confidence · 4 sources
Rachel Smithfrom KYC
rachel.smith@…from CRM
378-613-7157from CRM
VERIFIEDKYC · locked 🔒
WATCHfrom Risk
"Locked" means a regulated value (like a verified identity check) can't be casually overwritten by a less-authoritative system. The specialist source owns it.
So what

Each field comes from the system that knows it best: identity from KYC, contact from CRM, risk from Risk. One record, assembled from the best of all four.

7

Step 6 · Every value can be explained — that's lineage

LineageLineage: a record of where each value came from, why it was chosen, and what the alternatives were — like a receipt attached to every field. means nothing in the Golden Record is a mystery. If a regulator asks "why is Rachel's KYC status Verified?", there's a direct, documented answer.

Why each value won

FieldWinning valueFromWhy it won
Emailrachel.smith@…CRMMost trusted for contact
KYC statusVERIFIEDKYCRegulatory lock 🔒
Risk ratingWATCHRiskSpecialist source
Phone378-613-7157CRMMost recent valid
So what

Lineage turns a merged profile into a defensible one. Auditors and regulators get a straight answer for every field — no guesswork.

8

Step 7 · Two different scores people often confuse

This trips everyone up, so it's worth one screen. A record can be highly confident and still describe a higher-risk customer. They answer completely different questions.

Confidence: 93%

"How sure are we who Rachel is?"

93%  strong evidence
vs
Risk: Watch

"How much business concern does she carry?"

Watch  a judgement call
Read it as: "We're 93% sure this is the right Rachel, and the bank has placed her on Watch." High confidence is good. It does not mean low risk.
So what

Confidence measures trust in the data. Risk measures concern about the customer. Never present them as the same number.

9

Step 8 · Staff see one screen instead of four apps

This is the Customer 360Customer 360: a single, complete view of a customer — profile, identity, risk, history — all tied to one resolved person instead of scattered across systems. view. A relationship manager or compliance analyst no longer opens four applications to understand Rachel; everything hangs off the one trusted record.

Rachel's unified profile

Profile
KYC
Risk
Transactions
Lineage
Confidence93% · strong
Sources4 systems resolved
KYCVerified · current
Risk ratingWatch
Transactions7 linked to this customer
So what

Activity, identity and risk all follow the resolved customer — so decisions are faster and nobody works from a half-picture.

10

Step 9 · A compliance AI that answers from trusted data

This is the RAGRAG (Retrieval-Augmented Generation): the AI first retrieves the relevant trusted records, then writes its answer from them — so it's grounded in your data, not made up. layer. You ask a question in normal language; it finds the relevant Golden Record evidence, answers, and shows the sources it used. No database queries required.

Compliance officer asks…

Summarise Rachel's risk profile, PEP and sanctions status, and how many sources contributed.
Rachel Smith
• Risk rating: Watch
PEPPEP (Politically Exposed Person): someone in a prominent public role who needs extra compliance attention. Rachel is not one. status: No
• Sanctions: No
• 4 source records contributed.✓ Grounded in retrieved Golden Record evidence · sources shown
Question
Find best evidencefrom trusted records
Answer + sources
So what

It shortens the path to the right evidence — and because every answer cites its sources, you can check it. The AI assists; it doesn't replace the officer.

11

Step 10 · The same data, shown differently to staff and customers

A compliance officer needs internal risk detail. A customer should only see their own appropriate information. The platform serves both from the one trusted record, but filters what each audience can see.

Same record · two views

Staff · Compliance officer

Sees the full picture for investigation:

  • Risk rating & confidence
  • PEP & sanctions checks
  • KYC detail & source systems
Customer · Self-service

Sees only what's appropriate:

  • Their own name & contact
  • Their verification status
  • No internal risk, PEP or source data
Customer asks: What is my KYC status?
Your identity verification is complete. If we ever need updated documents, your bank will contact you.Customer-safe answer · internal fields hidden
So what

Two doors to the same secure room. Staff enter one; customers enter another and see only what they should. Right answer, right audience.

12

Step 11 · From fragmented records to one trusted foundation

That's the whole journey. Four conflicting records about Rachel became one explainable, safely-shared profile — and the same thing happened across the entire customer base.

89,198source records in
20,502trusted records out
The drop isn't lost data — it's duplicates merged into one. Every original is still there, with full lineage.
Lower costFewer duplicate profiles and manual searches.
Stronger complianceKYC, PEP, sanctions and risk in one explainable place.
Better serviceOne consistent profile across every channel.
So what

One customer, one trusted record, many safe experiences — for operations, compliance and the customer alike.