Almadar Institutional knowledge scattered across email and WhatsApp in an office on King Fahd Road doesn't need to be recreated, it needs to be made findable.
The practical symptom is familiar: an employee spends twenty minutes locating a policy document, a contract term or a previous decision, or gives up and asks someone, consuming two people's time. Multiply by a workforce and the cost is substantial and entirely invisible in any budget line.
Retrieval over generation
The useful architecture grounds answers in your own documents rather than a model's general knowledge. A question about your leave policy returns your policy with the source document cited, not a plausible general answer about Saudi labour law. This distinction matters because an unsourced answer is worse than no answer when someone acts on it.
What it works well on
Policies and procedures, contract terms across a supplier or customer base, technical documentation, prior project deliverables, and the accumulated answers to questions the same functions field repeatedly. Finance and HR are usually the highest-value starting points because both spend meaningful time answering the same internal questions.
Access control as a design requirement
Not everyone should retrieve everything. Salary information, board papers, legal advice and commercially sensitive contracts need permission handling that mirrors your existing access structure. Getting this wrong creates a data exposure that manual filing quietly prevented, which is why access design comes before rollout and connects to data governance.
A common Saudi scenario
A Riyadh group's HR team answers the same fifteen questions repeatedly, on leave entitlement, end-of-service calculation, iqama renewal, expense policy, mostly in both Arabic and English. A grounded assistant over the HR document set answers these directly with the policy cited. Query volume to HR drops sharply, and the answers become consistent, which they had not been when different staff answered from memory.
Keeping the corpus current
A knowledge assistant is only as good as the documents behind it. Superseded policies must be removed rather than left to be retrieved alongside current ones, and ownership of document currency must be assigned. Without this the system confidently returns last year's policy, which is the failure mode that destroys trust fastest and is entirely preventable with basic governance.
Choosing the first document set carefully
Deploying across every document a business holds at once produces poor answers and undermines confidence early. We start with one well-maintained, high-query set, usually HR policies or finance procedures, prove the answers are accurate and properly sourced, then extend. Each extension is a deliberate decision about document quality and access rather than a bulk ingestion. This connects to data governance and, where the corpus includes personal data, to the PDPL questions addressed in AI governance.
Organizations operating bilingually gain disproportionately, since retrieval across Arabic and English documents removes the common situation where a policy exists in one language and the person searching works in the other.