Containing the risks of generic AI and creating value

The pursuit of operational efficiency and profitability is one of the main guiding compasses of the national banking sector.

Today, Moroccan banks, operating in a market that is mature in many respects, have made performance management a top priority. In this regard, they have adopted, among other things, generic artificial intelligence for peripheral tasks such as marketing or public FAQs.

It is indeed well established that generic AI enables banks to improve operational efficiency. On the other hand, the downside is that this cutting-edge digital tool is not without risk when it comes to Core Banking (customer data, credit decisions, etc.).

At this stage, the key challenge is therefore to determine how Moroccan banks can benefit from the immeasurable potential offered by generic AI to improve their operational efficiency and profitability, while at the same time containing the multiple associated risks.

Franco‑Moroccan entrepreneur and investor Alex Halib, co‑creator of the AI solution Lessan, provides several answers, particularly in relation to the potential risks of generic AI for Moroccan banking institutions.

Three problematic scenarios

“The risk arises when we touch ‘Core Banking’, namely customer data and credit decisions. At this level, using a generic AI such as ChatGPT poses three major and blocking issues,” explains the young tech entrepreneur, who draws on several international entrepreneurial experiences (Europe, North America, Africa).

The first major obstacle relates to illegality: sending customer data to an American server runs counter to regulations on data sovereignty.

According to our expert, the second major issue is linked to hallucination, since in terms of local regulation, American models, for example, invent laws that do not exist. “In such a case, for a Moroccan bank, this is not a bug, it is a legal risk,” warns our interlocutor.

Lastly, the third blocking issue is the black box problem, as a bank cannot base its risk management on a model whose updates it does not control.

In summary, it is important to keep in mind that generic AI models are ill‑suited to linguistic code‑switching and local dialects, which are widely used in everyday life.

From this perspective, the Lessan solution, which positions Morocco as a laboratory, makes it possible to overcome a universal obstacle, especially given that there are 500 million Arabic speakers and a very large number of people (several hundred million) who use Swahili or Hinglish in everyday life in India.

Lessan’s value proposition

To overcome the disabling situations already described, Alex Halib and his team have developed the Lessan solution.

“Our mission is to unlock AI for the core business of banks (credit, compliance, legal matters, etc.). Today, this core is locked,” he confides.

Concretely, Lessan’s value proposition is based on three pillars: true sovereignty, linguistic identity, and sector‑specific expertise.

Regarding true sovereignty (on‑premise), unlike cloud solutions, Lessan operates directly on the bank’s servers. This gives the technological solution a major advantage: no sensitive data ever leaves the bank’s “basement”.

As for linguistic identity, Lessan’s models are trained and developed with real‑world usage in mind, namely darija, arabizi, and French‑Arabic code‑mixing, rather than just standard literary Arabic.

Still on the linguistic front, it is worth noting that billions of people speak local dialects that tech giants largely ignore. Hence the usefulness of the infrastructure developed in Morocco, which is intended to be deployed in similar markets internationally.

With regard to sector expertise, the strength of the new solution championed by our interlocutor lies in its deep knowledge of Moroccan banking regulation.

An active “super intern”!

Asked for concrete examples in which a sovereign AI would significantly enhance banks’ operational efficiency and thus their performance, Halib explains, in essence: “To put it simply, sovereign AI acts like a ‘super intern’ who never sleeps. However, it does not replace the banker; it prepares the banker’s work.”

The analysis of a corporate credit file is a telling example. Today, an analyst can spend hours reading financial statements and re‑entering figures—something that clearly leads to a loss of efficiency and effectiveness.

With a view to drastically reducing the “time‑to‑decision”, the Lessan AI solution can, for example, process financial documents, extract key ratios, and draft a preliminary summary in just a few minutes.

Quite logically, it is then up to the human expert to validate the final decision.

While the Franco‑Moroccan entrepreneur, who grew up in the Kingdom and lived for many years in France and Canada (Ontario), considers that a bank’s true assets consist of its procedures and its regulatory framework, he also points out that legislative information is often scattered across thousands of PDFs. This, in his view, is detrimental to process efficiency and effectiveness.

From this standpoint, the Lessan solution (a start‑up in the making) is designed to transform a documentary base into a “readily accessible brain.”

For example, a bank’s legal officer can ask the AI solution: “What is the current procedure for a land‑title release according to the latest circular?” and obtain an immediate answer, fully sourced, with a direct link to the exact internal document.

In this sense, it must be acknowledged that this sovereign AI can also be likened to active knowledge management.

A sovereign AI, adapted to local realities, therefore represents genuine added value for banking activity, with limited risks.

Streamlining customer journeys

Alex Halib is categorical: AI makes it possible to streamline customer journeys where traditional systems fail due to language barriers or data‑format constraints. From this perspective, it is a powerful lever for automation and digitalisation across banking processes.

“The darija chatbot allows a customer to say verbally ‘Bghit nbloki la cart diali’ (‘I want to block my card’)* and have the action executed without going through a complicated menu,” he assures.

Another telling example of automation and digitalisation of customer journeys: while the adviser is facing the client, they can ask the AI, “Is this client eligible for the Fogarim loan?” and receive an instant answer, with the criteria checked.