
The credit application reaches the decision complete, with the source of every figure
Today the credit analyst spends much of the time looking for documents across several systems, and the proposal they write depends on how much they managed to find.
Where a credit application stalls today
The request comes from the branch or the relationship manager, and from then on the credit analyst pieces the application together bit by bit: financial statements and tax returns from the document archive, the account's history from core banking, the reports from the central credit register and the collateral from yet another system. For an individual the documents change, with payslips and tax returns in place of financial statements, but the search is the same. This is where most of the time goes, and sometimes the application comes back from the decision stage because of an expired document that nobody had flagged.
The credit analyst receives the draft with the source of every figure and adds their own judgement
In the example, the agent reads the request, gathers from the bank's systems the documents required for that type of customer and facility, and flags any that are missing straight away, so the credit analyst can ask the customer for them from day one. It then extracts the data, reclassifies it according to the bank's criteria and drafts a proposal in which every figure points back to the document it comes from. The credit analyst checks it, corrects it and adds their own assessment, while the credit decision stays with whoever holds the authority to make it, as today.
- The agent compares the request with the documents required
- Reclassifies the data and cites the source of every figure
- The credit analyst checks the draft and writes the assessment
- Whoever holds the authority makes the credit decision
Illustrative example
What an assisted credit assessment needs
| Item | In this case |
|---|---|
| Data | In this caseRequests, financial statements or income, account history, the central credit register and collateral, with a single customer identifier |
| Modules | In this caseData Foundation, Data Intelligence, Company Brain, Agents |
| Who decides | In this caseThe credit analyst on the proposal, whoever holds the decision-making authority on the facility; the head of the credit department is the agent's owner |
| What we measure | In this caseThe time from request to credit decision, the completeness of applications at the decision stage and how many come back, using the same measurement taken before starting |
| AI Act risk class | In this caseIt depends on who is applying. Assessing the creditworthiness of an individual is a high-risk use under Annex III, while credit to a company does not fall under that point. That is why, before development, we define the classification with those at the bank who deal with risk and compliance |
| Where to start | In this caseWith Deployment on business lending, if the process has an owner; for lending to individuals, which calls for more controls, first a Mapping to weigh its timing and risk |
If the lending criteria are not written down
When two systems give two different exposures for the same customer, the agent inherits the same confusion as the credit analyst, so we start from customer master data with Data Foundation and the agent comes later. The same applies if the lending criteria change from one decision-maker to the next without being written down anywhere: before building, we put them in writing with whoever is accountable for credit, and this lengthens the project.
The other cases for banks are on the Banking & payments page, and all the sheets are in the use cases library.
Let's look together at a credit application that was sent back
At the first meeting we go through a returned application with a credit analyst, to see where it stalled and which data was missing.