Skip to content
Mtaji TechnologiesMtaji Technologies

Insights · Analysis · for financiers

Questions a credit committee should ask of a score it did not build

Bureaus, data vendors and platforms all offer scores. A committee that adopts one still answers for every decision made with it.

Mtaji Insights

Published 31 October 2026. 2 min read.

Written by the Mtaji Insights desk and checked against Mtaji’s product terms and publication rules before release. Editorial standards

In brief

  • A committee that adopts a score it did not build still answers for every decision made with it.
  • Ask what the score reads and on what consent, and whether it can explain a single result.
  • Ask how it was validated, who monitors it once live and who can switch it off. The financier decides.

Financiers are offered more scores than ever, from bureaus, data vendors and platforms like ours. Each promises to see creditworthiness a traditional file misses. A committee that adopts one takes on a model it did not build, and stays answerable for every decision made with it. These questions are worth asking first, of any score, the M-Score included.

A score is someone else’s judgment, compressed

Every score encodes choices: which data to read, how to treat what is missing, what outcome counts as failure. Those choices can widen access or quietly narrow it, and a model trained on a history of who was lent to can learn to repeat it. The committee does not need to rebuild the model. It needs to know enough to direct it, challenge it and stop it.

What does it read, and with whose consent?

Ask for the data sources and, for each, how it was obtained and on what lawful basis. For business credit, commercial performance (orders, deliveries, settlements) is relevant in a way that phone usage, contact lists or social media are not. Consent should be specific and given separately, not folded into terms nobody reads. Kenya’s data protection law gives people rights over decisions made by automated means, and a sole proprietor is a person.

Can it explain one result?

Accuracy across a portfolio is not enough. Take one applicant and ask why the score came out as it did, and what would change it. If the vendor cannot answer in terms a business owner would follow, the committee will not be able to answer a regulator, a customer or its auditors. The M-Score tells a business what it measures and the next step that would improve it. It does not publish how the parts are weighed.

How was it tested, and against what?

Ask what the model was validated on: repayments observed in a financed portfolio, or something else. A young model may have only modeled performance, and a responsible vendor says so plainly rather than quoting a result without its basis. Ask how it was tested for unfair outcomes across the groups your institution serves, and what was found.

Who watches it once it is live?

Models drift as the economy and the population they score change. Ask who monitors performance, how often, what triggers a review and who can switch the model off. Ask how a business can request a human review. A score with no monitoring plan is a decision the committee has stopped making without noticing.

Who decides?

The last question is the simplest. The score informs; the financier decides, under its own credit policy, and answers to its regulator for the decision. Any arrangement that blurs that line, in the contract or in practice, deserves a second look.