Product · AI Sampling

The AI builds your sample. You refine it by talking.

Klerum's AI turns a raw loan trial into a defensible review population that meets your coverage and tier targets, then lets reviewers shape it through conversation rather than a thirty-checkbox modal.

How it works

From a raw trial to a defensible sample, in four steps.

01

Upload the loan trial

Drop the bank's trial balance as CSV. The AI proposes a field mapping for any headers it does not recognize, and for repeat uploads it caches the mapping so it learns the bank's core format.

02

The AI recommends a sample

One click rolls the trial up by borrower and the AI recommends a review population against your scope and the effective profile, with a per-borrower rationale and an estimated coverage hit.

03

Refine it conversationally

Edit in plain English: “drop anything under $250k,” “add two more CRE credits,” “swap out the campground loan.” The selection updates and the conversation is saved.

04

Promote to the review population

Promote the final list in one step. For borrowers that exist only in the trial with no prior relationship, Klerum creates the relationship automatically.

The AI recommendation

A sample with a rationale on every line.

One click rolls the trial up by borrower and the AI proposes the review population against your scope and the effective profile, with a per-borrower rationale and a running coverage estimate.

AI sample recommendation with per-borrower rationale
Selected 19 of 61 relationships at 56.6% coverage — each borrower annotated with why it is in the sample against the chosen profile.
Conversational refinement

Refine the sample by talking to the model.

Add, drop, and interrogate candidates in plain English instead of a thirty-checkbox modal. The selection and coverage update as you go, and the conversation is saved with the engagement.

Conversational refinement transcript with the model
“Drop anything under $5M outstanding, and add the two largest Special Mention credits we're missing.” The model edits the sample and reports back what it changed.
The AI learns the data

A clean trial, however the bank exports it.

Upload the bank's trial balance and the AI proposes a field mapping from its headers, normalizing any export into a consistent set of fields so sampling, roll-ups, and coverage math work the same across every institution. The mapping is cached per bank, so the next upload starts from the confirmed one.

AI field mapping on a loan trial upload
100 loans, $222M outstanding — the AI maps each of the bank's headers to a canonical field for review.
borrowerCodeborrowerNameloanNumberloanTypeoriginationDatematurityDatecommitmentAmtoutstandingBalriskRatingloanOfficer

Today Klerum ingests CSV trials. Excel ingest is on the roadmap.

See Klerum on your own portfolio.

A 30-minute demo, no deck. Upload a sample trial and watch the AI propose a review population against your methodology.