Where do the model artefacts live?

In Saudi Arabia, with the features and the training records. A hosted model, when used, receives masked features and the boundary is documented.

What do you hand over for an examination?

The model file, drift and performance reports, the tuning record and a decision log the examiner can sample directly.

Do you replace the core systems?

No. We build the governed layer beside them and read from the systems of record, so the bank keeps its core intact.

Can you start with one product line?

That is the usual entry point. One product line, one governed release, then the pattern is repeated across the rest.

Riyadh, banking

Riyadh banks buy models they can defend to a supervisor

Credit, fraud and onboarding decisions are supervised activity in Saudi Arabia. The question is never whether the model scores well, it is whether the decision can be traced back to its inputs.

Book the Data Foundation Assessment Everything we do in Riyadh

A model without lineage is an opinion. We build the lineage first.

SAMAPDPLNCA ECCZATCA

Key takeaways

What you need to know, in a minute

Reason codes on every automated decision

The customer file shows which features drove the outcome and which source document each one came from.

Mixed Arabic and English documents

Onboarding packs arrive in both languages and in poor scans. Extraction is built for that reality, with confidence recorded per field.

Risk and finance read one number

Reconciliation checks run inside the pipeline, so a break is raised with an owner instead of being discovered in a committee.

Retail and corporate

Onboarding and KYC file assembly

Identity, address and corporate documents read, cross-checked and assembled with every extracted value traceable to its page.

Financial crime

Alert tuning that leaves a record

Threshold changes, the reason for each change and the effect on alert volume are kept as a file, not as tribal memory.

Service

Arabic complaint and call handling

Transcripts, categories and root-cause themes produced in Arabic as the governing version, because that is what the customer wrote.

Use cases

Where this applies in practice

What usually stalls these programmes

No model register

Without a named owner and a documented review cycle per model, the first supervisory question stops the programme.

Production data in test environments

Masking in non-production is set up before the first pipeline, not after an audit finding.

Vendor scores with no explanation

A score that cannot be decomposed cannot be defended. We keep the decomposition inside the bank.

Banking and finance in Riyadh, questions we get

Other sectors in Riyadh

Government and public sector

Bad data, bad AI. We fix the data layer under the service first.

Open page

Healthcare

One patient, one record, one version of the truth.

Open page

Telecom

Fix the subscriber record, then the models get interesting.

Open page

Retail and commerce

Catalogue first, customer second, forecast third.

Open page

Energy and utilities

An asset register that matches reality is the whole project.

Open page

Jeddah

Clean the intake and the rest of the automation stops fighting you.

Let's put your AI ambition on the table

Sixty minutes with a senior Bilytica partner. We stress-test your plans against what actually works in your sector, agree which moves are worth funding first, and you leave with a written 90-day plan and honest numbers, whether you work with us or not.

Start the conversation