Frequently Asked Questions
Answers for banks, neobanks, BNPL providers and digital lenders, from exploring the problem to signing.
Alternative credit scoring explained
What is alternative credit scoring, and how is it different from a traditional credit score?
Alternative credit scoring predicts whether an applicant will repay using data other than a credit bureau file. A traditional credit score is built on repayment history, so it only works for people who have already borrowed. An alternative credit score can assess anyone, including first-time borrowers.
Credolab's alternative credit score uses privacy-consented device and behavioural interaction metadata. Where the bureau looks at the past, we read intent: how an applicant behaves while applying says a lot about whether they mean to pay you back.
What alternative data can lenders use for credit risk assessment?
The main sources of alternative data for credit scoring are open banking and cashflow data, rent and utility payments, telco data, digital footprint data, psychometric tests and device and behavioural metadata. Each answers a different question, and each covers a different share of applicants.
Credolab focuses on device and behavioural metadata because it covers every applicant who applies digitally, with no extra step for the customer and no personal data. It complements the other sources rather than replacing them.
How does behavioural data predict whether an applicant will repay?
Behavioural data predicts repayment because people who intend to repay behave differently from those who don't, long before the first instalment is due. Credolab measures patterns such as how an applicant fills in the application form, how they use their device and habits like charging their phone overnight.
Our models compare each applicant with the behaviour of confirmed good and bad borrowers in your own portfolio, and return the probability that they will miss a payment. Counts and patterns only; never content.
Why do traditional credit scoring models miss thin-file applicants?
Traditional credit scoring models need a credit history to score, so applicants with a thin file or no file get either no score or an unreliable one. Lenders then reject good customers or approve them blind, and both cost money.
Credolab fixes that data asymmetry: every digital applicant leaves a behavioural footprint, so every applicant gets a score. That is why our clients raise approval rates as well as cut defaults.
Where does alternative data fit in a bank's credit risk management framework?
In a credit risk management framework, alternative data sits at origination, alongside bureau and income checks, and feeds the same credit risk analysis. It adds a signal for applicants the bureau cannot see and a second opinion on those it can.
Credolab's score slots in as an extra variable, a separate decision rule or a challenger model. Because it typically correlates less than 5% with bureau, cashflow and socio-demographic data, it strengthens your existing credit risk assessment instead of repeating it.
Choosing a credit scoring solution
What should we look for in a credit scoring solution?
A good credit scoring solution should score every applicant, add signal your current data lacks, and fit your stack without a rebuild. In practice, ask five things:
- Coverage: what share of your applicants gets a score?
- New signal: how strongly does it correlate with the data you already buy?
- Proof on your data: can you test it on your own live applicants before you commit?
- Integration: how many developer days, and does your decision engine change?
- Privacy: does it process personal data, and who is the controller?
Credolab's answers: 100% of digital applicants, typically under 5% correlation with other sources, a production test in shadow mode, about two developer days, and no personal data, with Credolab always the processor.
Is Credolab credit scoring software, a data provider or a decisioning platform?
Credolab is a behavioural scoring and data layer, not a decisioning platform. We provide credit scores (CredoScore) and behavioural insights (CredoInsights) through an API, and your decisioning platform or automated credit decisioning system uses them to make the call.
Think of us as the intelligence inside your credit stack: you keep your decision engine, your rules and your credit policy, and we add the signal they are missing.
Is Credolab a credit bureau or a Consumer Reporting Agency?
No. Credolab is not a credit bureau or a Consumer Reporting Agency. We don't collect or maintain credit files or sell consumer reports; we license scoring models and analytics to lenders, much as FICO does.
We work alongside bureaus as partners, including TransUnion and Círculo de Crédito, to widen the population they can score.
How is Credolab different from fraud prevention and device intelligence tools?
Credolab predicts credit risk, not fraud. Our input data can look similar to a device intelligence vendor's, but our output is the probability that an applicant will miss a payment, from First Payment Default up to 90 days past due at 9 months on book.
That matters because many costly applicants are not fraudsters. They have valid IDs, normal devices and clean bureau files, and they simply never intend to pay. Fraud tools pass them; we flag them. If your device data already caught them, your risk team would not still be fighting false positives.
How does Credolab compare with cashflow underwriting and open banking data?
Credolab complements cashflow underwriting; each answers a different question. The bureau asks whether an applicant has paid before. Cashflow and open banking data ask whether they can afford to repay. Credolab asks whether they intend to repay.
Lenders get the strongest credit risk assessment by combining all three. Credolab also needs no extra step from the applicant, such as connecting a bank account, so it covers people who would drop out of an open banking flow.
Should we build a behavioural scoring model in-house instead?
You can, but most lenders find it slower and costlier than it looks. A behavioural credit scoring model needs a data collection SDK that stays compliant with app store policies, feature engineering over thousands of signals, and years of outcome data to know which signals actually predict default.
Credolab has refined its behavioural feature library since 2016, across more than 200 million people in 37 countries. You get a model built for your portfolio and recalibrated quarterly, while your data science team focuses on your decision strategy.
Credolab's scores and results
What exactly does Credolab's score predict?
Credolab's score predicts the probability that an applicant will miss a payment. You choose the outcome with us, from First Payment Default (FPD) through FPD30 and 30 days past due at 3 months on book, up to 90 days past due at 9 months on book.
The score is then judged by how well it separates applicants who went on to pay from those who didn't, on your own portfolio.
What results do lenders typically see with Credolab?
Lenders using Credolab typically approve more applicants and reduce defaults at the same time: up to 32% more approvals, up to 50% fewer defaults and up to 45% more predictive power when our scores are added to their existing models. Results depend on your segment, market and how you use the score.
Some examples from our clients:
- A Brazilian neobank cut First Payment Default by 32%.
- CrediOrbe in Colombia raised approval rates by 30%.
- A US salary advance provider reduced defaults by 25%.
Is Credolab's AI-powered credit scoring explainable?
Yes. Credolab's models are built on logistic regression, not black-box AI, so they are statistically explainable and auditable. Machine learning helps us engineer features from behavioural metadata; the scorecard itself stays transparent.
Each model is validated by its ability to separate good and bad borrowers on your own portfolio, which is what your model risk and validation teams will review. Because no personal data goes in, the score cannot take race, gender, postcode or any other personal attribute into account.
How much performance data do we need, and how soon can a model go live?
You can start using Credolab's pre-built scorecards, built on our proprietary behavioural feature library, as soon as you have a first set of matured loan outcomes. Your tailor-made CredoScore model follows once your portfolio has enough outcomes to train on, and we recalibrate it every quarter.
We never deploy a model blind: every model is validated on your own applicants before you use it in production. Our data scientists agree the right timing and cut-offs with your risk team.
Markets and integrations
Does Credolab work in our country?
Yes. Credolab collects data from Android, iOS and web applications, so we work wherever your applicants apply digitally, in any country. All you need to do is embed our SDK in your app or website to start collecting data. We have already scored applicants in 37 countries across the Americas, Europe, the Middle East, Africa and Asia-Pacific, so our behavioural feature library reflects very different markets.
Every model is calibrated on your own portfolio, so it reflects your customers rather than a global average. Where local privacy law requires data residency, we store metadata in-country.
Does Credolab work for US lenders?
Yes. US lenders use Credolab to catch First Payment Default that bureau and fraud checks miss: "non-starters" and "never-repayers" who pass KYC, look fine on the bureau and still never make a first payment. Our FPD score flags them at application, before you fund the loan.
Credolab also scores credit-invisible consumers, often the only way to properly assess applicants with no US credit file. And it works with the FICO Platform with no workflow or API changes.
Which decisioning platforms does Credolab work with?
Credolab's scores are available through leading credit decisioning and orchestration platforms, including the FICO Platform, Provenir, Synapse Analytics, Taktile and TaranDM, as well as through credit bureau partners.
On the FICO Platform you bring your own Credolab licence: existing Credolab clients receive scores and features through the existing connectors, and FICO clients new to Credolab just integrate our SDK. No workflow or API changes either way.
Data, privacy and compliance
Does Credolab pool data across clients or run a consortium?
No. Credolab does not run a consortium or share data between clients. Your data is kept in its own partition, and every velocity count, identity check and model is computed on your data alone.
Your applicants' behaviour trains your model, not a competitor's.
Does Credolab use a third-party device ID?
No. Credolab's Device ID is proprietary, built on the device's own fingerprint rather than on IMEI, advertising IDs or similar third-party identifiers. It lets you recognise repeat devices across applications without depending on someone else's identifier.
Which privacy laws does Credolab support?
Credolab is designed to help lenders comply with the privacy laws of the markets they lend in, including GDPR, CCPA, LGPD in Brazil, LFPDPPP in Mexico, PDPA in Singapore and the Data Privacy Act in the Philippines.
We collect no personal data, act as your data processor under a Data Processing Agreement, and store metadata in-country wherever the law requires it. If a user opts out, our SDK collects nothing.
How does Credolab fit the EU AI Act?
Under the EU AI Act, credit scoring is a high-risk use, and the lender deploying the overall decision system carries primary responsibility. Credolab's score is one component of that system, not a standalone automated decision, and we act as technology provider and data processor.
Our models are logistic regression scorecards, statistically explainable and auditable, which makes the documentation and oversight the Act requires easier for your team.
Where is Credolab hosted?
Credolab runs in the cloud, with each client's data held in its own partition. There is nothing to install or maintain on your servers, and we handle updates and model recalibration.
Where local privacy law requires data residency, we host metadata in-country.
Buying and getting started
What do procurement and InfoSec teams usually ask for?
Credolab's procurement pack covers what bank and fintech InfoSec, legal and vendor-risk teams usually request: our ISO/IEC 27001 certificate, a Data Processing Agreement, our privacy notice, security questionnaire answers and SDK documentation.
Metadata is encrypted at rest and in transit. Most answers are already on our Trust centre; for the rest, your account manager will send the full pack.
What happens after we sign?
After signing, onboarding follows four steps:
- Integrate: your developers add the Credolab SDK, typically in about two days, guided by our Developer Area and your Customer Success Manager.
- Collect: the SDK starts collecting metadata from consenting applicants while your current process keeps making every decision.
- Validate: once loans mature, we validate the score on your portfolio and agree cut-offs with your risk team.
- Go live and improve: you use the score in production, and we recalibrate your model every quarter.
Your Customer Success Manager stays with you throughout, including on app store policy compliance.
More answers on our product pages
On the hub, this block shows as plain links with no repeated answers, so each answer has one home and one schema block.
- Homepage: what Credolab is, how it differs from a bureau, what data it uses, who uses it, growing without bureau hits
- Product: how the SDK works, integration time, platforms, CredoScore vs CredoInsights, fit with existing models
- Use cases: thin-file applicants, First Payment Default, champion/challenger, use after onboarding
- Pricing: subscription tiers, how to evaluate, engineering cost
- Trust centre: personal data, controller vs processor, consent, ISO 27001, data residency, Google Play policy