BEHAVIOURAL ANALYSIS FOR LENDERS · BUILT FOR AMLR

Customer due diligence that sees the person behind the deal.

Key Insights makes lenders’ onboarding AMLR-ready and analyses how customers describe their own business and loan, compared with registry data. The result is better input for onboarding, credit decisions and early collections.

Fewer drop-offs Hard data + behaviour Built for AMLR

Decision support for credit and compliance teams. The final decision is always yours.

CASE · ONBOARDING · SEK 800K Example Company Ltd
Ready for review
REGISTRY DATA VIA PARTNERRetrieved
Company details, board, beneficial owners, PEP and sanctions, annual accounts
BUSINESSConsistent
Customer: “Renovations for private homeowners” Registry: construction, SEK 2.1m
FOREIGN EXPOSUREContradiction
“Everything is local” → later: prepayment to a supplier in Estonia
PEOPLE IN THE DEALFlagged
“Jonas handles that arrangement” not in the registries
AMOUNT VS TURNOVERReview
SEK 800k 38% of latest annual turnover
BEHAVIOURAL SIGNALS
Concrete about the business Distanced about the loan Changing story
RISK CLASS INPUT Elevated Your compliance team sets the class
AUDIT TRAIL 14 data points Logged with source and rule

Example with a fictitious company.

USE CASES

One engine, three use cases

Start where the need is greatest. The same analysis follows the customer from the first application to repayment.

01 ONBOARDING AND AMLR
KI AWARE

Onboarding that reads between the lines

Registry data is retrieved automatically and the customer confirms it instead of filling it in. Open questions are checked against the registries in real time, discrepancies are flagged and everything is documented for AMLR.

  • Discrepancy report and risk class input
  • Beneficial owners, PEP and sanctions via partner
  • Ongoing monitoring and reclassification
  • Works on top of your existing onboarding

Built for AMLR, (EU) 2024/1624

02 CREDIT DECISIONS

Decision support for a second look

Credit officers get structured insight into ownership, realism in forecasts and the ability to execute the plan, alongside the financial ratios. Borderline cases can get a second look instead of an automatic no.

  • Behavioural indicators that complement the ratios
  • ML models trained on your own outcomes
  • Evidence and quotes behind every insight
  • For lending to companies

Decision support, never automated decisions

03 EARLY COLLECTIONS

Prioritise where the conversation makes a difference

In early collections it is hard to see which customers can and want to pay. Analysing the customer dialogue gives a prioritisation based on ability to pay, willingness to pay and signals in the dialogue.

  • Prioritisation of cases 1–90 days past due
  • Input for the right approach in the dialogue
  • Traceable reasoning for every case
  • The same engine as in onboarding

Decision support for your collections team

The same engine works wherever people describe their business in their own words.

HOW IT WORKS

An engine that checks the customer’s words against the registries

The engine takes in registry data and the customer’s own answers, reconciles them and adds behavioural indicators. The result is a ready-to-review file for your credit officers or compliance team.

IN

Data and the customer’s answers

Registry data via partnerCompany details, beneficial owners, signatories, PEP and sanctions, annual accounts
The customer’s own answersOpen questions in our chatbot, your own form or your existing application
Outcomes over timeRepayments, new financing and registry changes
THE ENGINE

Reconciliation and behavioural analysis

Registry reconciliationEvery answer is tested against what has already been retrieved
Language analysisLevel of detail, consistency and how concrete the customer is
Behavioural indicatorsResearch-based signals on ownership, risk appetite and realism
Rule framework per AMLR Art. 20(1)Transparent rules for what is flagged and what is escalated
OUT

A ready file for your team

Discrepancy reportStructured differences between the customer’s answers and the registries
Risk class inputLow, normal, elevated or high. Your compliance team sets the class.
Behavioural signalsIn separate views for compliance and credit officers
Audit trailSource, time and applied rule for every data point

BETWEEN THE LINES

The engine reads the whole dialogue together with the registries

Every answer is checked against the registries and against the customer’s other answers. Contradictions, people who do not appear in the registries and language that changes character are flagged with a reference to exactly what the customer said.

WHAT THE ENGINE LOOKS AT
Level of detail Consistency across answers People and roles Locus of control Distanced language Plausibility against registries

Based in part on research on unexpected follow-up questions: Hartwig, Granhag et al. (2005) and Vrij et al. (2009).

ONBOARDING DIALOGUE · EXAMPLE COMPANY LTD Amount applied for SEK 800k
KI Aware Describe how a typical deal works, from when a customer finds you to when you get paid.
Customer · answer 1 We do renovations for private homeowners in Stockholm, mostly bathrooms and kitchens. Customers find us through referrals and we invoice when the job is done.
KI Aware Do you have customers or suppliers outside Sweden?
Customer · answer 2 No, everything is local.
KI Aware What will the money be used for?
Customer · answer 3 Materials for a bigger project starting in the spring.
KI Aware · follow-up Could you tell me more about the project and who you buy the materials from?
Customer · answer 4 It’s a partnership. The supplier is in Estonia and wants prepayment, Jonas handles that arrangement. We’ve done similar things before.
THE ENGINE’S ANALYSIS

File for the credit officer

CONSISTENTAnswer 1 vs registry
The business matches the registered industry, turnover and number of employees.
CONTRADICTIONAnswer 2 vs answer 4
“Everything is local”, but a prepayment is going to a supplier in Estonia.
UNKNOWN PERSONAnswer 4 vs registry
“Jonas” appears to control the arrangement but is not among the board, signatories or beneficial owners.
AMOUNTApplication vs accounts
SEK 800k equals 38% of the latest annual turnover, for a project that is not described further.
LANGUAGEAnswer 1 vs answers 3–4
Concrete and detailed about the business itself, but vague and distanced about the loan: “it’s a partnership”, “Jonas handles that”.
SUGGESTION UNDER YOUR RISK POLICY Elevated risk. Verify the supplier and Jonas’s role before deciding. The credit officer makes the assessment.

A short or nervous answer is not enough for a flag. The engine first asks a follow-up question and then weighs the whole dialogue against the registries.

GETTING STARTED

From the first meeting to live operation

You see what the engine finds in your own cases before anything in your process changes.

  1. 1

    Mapping

    We review your onboarding flow against AMLR and agree which steps the engine takes care of.

  2. 2

    Shadow run

    The engine runs on 25–50 of your existing cases, alongside your own process.

  3. 3

    Integration

    We connect the engine to your existing application, or you use our chatbot.

  4. 4

    Live operation

    Your credit officers and compliance team get ready files. You make the decisions.

BUILT FOR REGULATED LENDERS

Security and compliance from the start

OUR PRINCIPLE

A person always decides

Key Insights is decision support, not automated decision-making. Every flag shows its basis, with source, rule and quote, so that a person can weigh it.

AMLR-ready from day one

The onboarding process and data handling are built for AMLR, (EU) 2024/1624.

Traceable at every step

An audit trail with source, time and applied rule for every data point and decision.

Built to ISO 27001/27002

AES-256 encryption at rest, two-factor sign-in and role-based access.

GDPR by design

A data processing agreement with documented sub-processors. The platform has been reviewed by a lawyer.

Data stays in the EU

All data is stored and processed in Stockholm, Sweden.

Designed for the AI Act

Transparent logic, documented rules and human review of every file.

ABOUT US

The numbers show what a company has done. The person behind it decides what it will do.

10 July 2027 AMLR applies across the EU, including the requirement to understand the purpose of each customer relationship.
≈ 23,000 Swedish companies, 1.7% of all, had links to people in criminal networks in 2020–2023.
≈ 6% of the portfolio is written off each year at Swedish SME lenders.

Sources: AMLR, (EU) 2024/1624 · Brå (Swedish National Council for Crime Prevention) · Key Insights review of Swedish SME lenders’ 2024 annual reports.

Annual accounts show how last year went. A credit report shows whether someone has missed a payment. How a company handles the years ahead is still decided by the person running it.

The person behind the business makes the decisions, sets the plans and handles the setbacks. That is also where the purpose of the deal becomes clear: what the money will actually be used for, who is in control and where the business is heading. Understanding the purpose of the customer relationship is exactly what AMLR requires you to assess and document.

Yet most onboarding flows are built on rigid forms and checklists. Customers fill in details that already exist in the registries, credit officers search for discrepancies by hand and the process becomes long for everyone. Checks that tick off registries and lists rarely meet today’s requirements, while money laundering keeps finding new ways in, often through companies that look perfectly ordinary on paper.

The result is a balancing act. Lenders must follow the rules without making it hard to become a customer, stop money laundering without screening out good customers, and keep losses down without rejecting companies that lack a long track record.

The ones who most often get caught in the middle are skilled entrepreneurs.

A young company, an unusual business model or a temporary dip is enough for a case to stall in manual review or be rejected automatically. Often the customer gives up before anyone has understood who the entrepreneur is and what the company is meant to become.

That is why Key Insights exists. We make the person behind the deal visible, in a way that can be reviewed and documented. Customers describe their business in their own words, and the engine checks it against the registries and highlights what does not add up. The process stays simple for the customer and thorough for you, so you can say no to the wrong customers sooner and yes to the right ones more often.

WHO WE ARE

Credit expertise and behavioural science in one team

Martin Branzell

Martin Branzell

CEO AND FOUNDER

First employee at Krea Technology, Sweden’s and Finland’s largest business-loan broker. Seven years in business development and bank partnerships across Sweden, Finland and the Baltics, with a seat on the management team.

Sverker Sikström

Sverker Sikström

CO-FOUNDER, AI AND TECHNOLOGY

Professor of cognitive psychology whose research field is AI-based language analysis. Leads the development of the engine and of TalkToAlba, where the same technology is already in live use.

CONTACT

See the engine read your own cases

Book a demo and we will walk through your onboarding flow and show what the engine finds in a selection of your cases.

Book a demo
Martin Branzell, CEO and Founder martin@keyinsights.se