Arabic Multimodal Search Relevance Recovery Leader

The mandate

The company requires executive leadership to develop multimodal search across Gulf markets with rigorous Arabic acceptance evidence. The role must improve engagement while preserving dialect intent, balance visual and structured signals for property relevance, ensure amenity accuracy against inventory data, and build dashboards that resolve conversion attribution across customer, partner and business perspectives.

The interim must take the Dubai seat within three weeks for ten months. A permanent regional search appointment will open after the recovered evaluation suite governs two releases, with six weeks allocated for the successor to observe a peak-period decision and then chair the third release council under the interim’s documented handover.

The assignment closes when Arabic, English and code-switched search journeys have segment-specific quality and harm corridors; three consecutive releases satisfy offline, human-review and controlled-online gates; content conflicts are traceable to source; and the successor signs the residual dialect, image and inventory risk register. Aggregate click-through improvement cannot replace those conditions.

The leader may pause relevance launches, reweight the approved experiment portfolio, quarantine unreliable labels, change external annotation capacity and move up to AED 25 million within the search budget. Product withdrawal, permanent leadership hires, supplier commitments above AED 8 million and changes to marketplace ranking policy need council approval; the interim cannot suppress safety or accessibility attributes to lift conversion.

Advertising rank, hotel contracting, booking-engine replacement and general customer-service transformation sit beyond this brief. The work covers query understanding, image-text grounding, inventory consistency, relevance measurement and the decision machinery connecting these systems to regional product releases.

Why this seat is open

The failed expansion exposed a governance model in which global benchmark improvement could overrule regional language and content evidence. Once the responsible executive left, product and science leaders defended incompatible accounts of the result. Temporary authority is required to rebuild a trusted measurement system before the company appoints a lasting regional search owner.

What you will own

• Reconstruct the release impact by dialect, code-switch pattern, device, journey intent, image dependency and property-content quality rather than one regional average.
• Establish a protected Arabic evaluation set with provenance, reviewer agreement, cultural and accessibility cases, contamination controls and scheduled refresh rules.
• Decide how text relevance, visual similarity, structured inventory and verified partner content arbitrate when their signals conflict.
• Introduce launch gates joining offline retrieval quality, fluent human judgment, harm checks, guarded experiments, complaint signals and reversible rollback.
• Rebuild annotation operations around dialect competence, hard-case escalation, blind review, adjudication evidence and measurable label drift.
• Resolve the economics of heavier models, regional retrieval, re-ranking and image processing against latency, compute, conversion and customer-remediation costs.
• Transfer the evaluation registry, disputed decisions, supplier posture, launch calendar and talent assessment through a successor-led peak release.

Candidate qualifications

• Held executive or enterprise-director authority for Arabic search, recommendation, language AI or multimodal discovery serving multiple Middle Eastern markets.
• Recovered a failed relevance release by reconciling offline measures, fluent reviewer evidence, controlled experiments and customer outcomes.
• Built evaluation coverage across dialectal Arabic, Modern Standard Arabic, English and code-switching with defensible annotation quality.
• Governed text-image grounding where visual cues, structured attributes and merchant or partner content produced contradictory ranking signals.
• Made serving-cost and latency trade-offs across retrieval, ranking and multimodal models without sacrificing signed customer-protection constraints.
• Completed a permanent handover after resetting a multilingual AI product’s release, data and regional decision rights.