From Cloud Storage to Production ML: Yassir’s 2023 Inflection Point
Yassir was founded in 2017 by Noureddine Tayebi and El Mahdi Yettou as a ride-hailing app in Algiers. By 2022, the company had raised a $150 million Series B and expanded into food delivery, grocery delivery, and financial services across 45 cities in Algeria, Morocco, and Tunisia — with further expansion into France, Canada, and Sub-Saharan Africa. The growth trajectory was clear. What was less clear was how to operate an increasingly complex multi-service platform without the operational efficiency gains that would allow the business model to sustain itself at scale.
The 2023 integration of Vertex AI — Google Cloud’s managed machine learning platform — marked the operational inflection point. Prior to Vertex AI adoption, Yassir’s machine learning development cycle took several weeks from model development to production deployment. Post-integration, that timeline compressed to days. The acceleration is not primarily a speed-for-speed’s-sake achievement; it means that Yassir can now iterate on recommendation algorithms, pricing models, and matching logic at a cadence that approximates real-world learning rather than quarterly batch updates.
The full Google Cloud stack underlying the Vertex AI deployment includes Google Kubernetes Engine (GKE) Autopilot, Cloud Run for containerized applications, BigQuery for data warehousing and analytics, Dataproc and Dataflow for data processing, Dataplex for governance, and Looker for business intelligence. This is not a single-tool integration — it is a comprehensive data infrastructure that makes Vertex AI’s ML capabilities operationally coherent. Models trained on BigQuery data, governed by Dataplex, and monitored through Cloud Logging and Trace create a feedback loop that supports continuous improvement rather than one-time optimization.
The ML Outcomes: What the Numbers Actually Say
Yassir’s Vertex AI deployment has produced measurable outcomes across three business lines, each with distinct mechanics and strategic implications.
Ride-hailing operations achieved the most tangible operational improvement: a 50-second reduction in driver assignment time. This is not a vanity metric. In the economics of ride-hailing, driver assignment latency directly affects cancellation rates — drivers cancel pickups that take too long to initiate, and passengers cancel requests that don’t match quickly. A 50-second improvement in assignment time represents fewer cancellations per completed ride, lower dead-time miles for drivers, and improved unit economics per trip. The cascade effect across millions of rides per year is substantial.
Grocery delivery produced the most significant projected revenue impact: an estimated 20% revenue increase through AI-driven personalized recommendations, combined with a 25% increase in average order value. The recommendation system learns from purchase history, basket composition, time-of-day patterns, and neighborhood-level demand signals to surface products with higher probability of conversion. These projections reflect Yassir’s own modeling of the recommendation system’s contribution — which is standard practice in retail ML — but even at conservative discounts to the stated figures, the directional impact is material.
Customer retention across all services improved by an estimated 20%, according to Yassir’s reporting. This is the compound effect of more relevant recommendations, faster service fulfillment, and the network effects of a super app where a user retained in grocery delivery is also a potential ride-hailing and fintech user.
The full set of outcomes — faster matching, higher basket values, better retention — illustrates why vertical integration into a super app model amplifies the value of ML investments. Each data layer informs every other service.
Advertisement
The Kawarizmi Acquisition: Extending the ML Stack into Retail Media
In March 2026, Yassir acquired Kawarizmi — a Paris-based adtech firm specializing in programmatic trading and data-driven media buying across Europe, Africa, the Middle East, and South Asia. The deal is the logical next step in Yassir’s data monetization strategy and deserves to be understood in that context, not merely as an adtech acquisition.
Yassir’s 100,000+ commercial partners across its delivery and grocery platforms generate a first-party data asset that most Algerian companies do not possess: rich behavioral purchase data, location data, and session-level engagement data tied to verified user identities. This data is worth considerably more than the revenue from delivery commissions if it can be used to connect brand advertisers to intent-ready audiences.
Kawarizmi’s programmatic expertise — algorithmic media buying, creative performance optimization, and audience reach across MENA diaspora markets — provides the technical infrastructure to build what CEO Noureddine Tayebi described as “a scalable retail media network.” The strategic model is similar to what Amazon built with Amazon Advertising: use e-commerce and delivery transaction data to create a high-signal advertising product that outperforms generic display advertising because it is grounded in real purchase behavior.
For Algeria’s broader tech ecosystem, the Kawarizmi acquisition signals the emergence of a new monetization category: first-party data as a business asset, not just an operational byproduct. Algerian companies with significant user bases — telecoms, fintech platforms, delivery services — are sitting on data that has structural advertising value if properly organized, governed, and connected to brand partners.
What Algerian Tech Companies Should Take From Yassir’s Playbook
Yassir’s ML deployment is not a one-off success story — it is a replicable pattern that other Algerian technology companies can apply to their own contexts.
1. Start with a measurement-first data infrastructure before any ML model
The Vertex AI deployment did not produce outcomes because the models are inherently superior — it produced outcomes because Yassir had already built a coherent data infrastructure (BigQuery + Dataplex + Looker) that made model outputs measurable and actionable. Algerian companies attempting to deploy ML without first solving data quality, data governance, and measurement will run models that produce outputs nobody trusts. The infrastructure investment — data warehousing, logging, monitoring — is the prerequisite for ML value, not the afterthought.
2. Target matching and ranking problems first — the ROI is most direct
Driver assignment optimization and product recommendation ranking are ideal first ML use cases because their impact is directly measurable in operations and revenue. Every percentage point improvement in match quality translates to fewer cancellations; every recommendation that converts to a purchase has a dollar value. Algerian companies in logistics, e-commerce, and services should map their highest-frequency, highest-volume matching or ranking decisions and identify which ones currently run on simple rules that could be replaced by learned models.
3. Treat first-party data as a strategic asset requiring active governance
Yassir’s acquisition of Kawarizmi to monetize its first-party data reflects a recognition that transaction data from operations has commercial value beyond the transaction itself. Algerian enterprises accumulating purchase, location, and behavioral data — banks, telecoms, delivery platforms, e-commerce operators — should begin treating this data as a balance-sheet asset requiring formal governance (data classification, access controls, retention policies) rather than an operational byproduct. Without governance, the data cannot be commercially activated, and the window for building proprietary data advantages narrows as regulatory requirements (Algeria’s data protection framework, Law 18-07) become more strictly enforced.
4. Use managed ML platforms to avoid the infrastructure trap
Vertex AI’s value for Yassir is not just in the models — it is in the managed infrastructure that eliminates the need for dedicated MLOps engineering teams to maintain model serving, versioning, and monitoring. For Algerian companies without Yassir’s engineering headcount, managed platforms (Vertex AI, AWS SageMaker, Azure Machine Learning) allow small teams to reach production-grade ML without building the full infrastructure stack from scratch. The compressed development-to-production timeline — from weeks to days — is available to any Algerian company willing to invest in the cloud data foundation.
The Bigger Picture
Yassir’s Vertex AI deployment and Kawarizmi acquisition together tell a story about what the next generation of Algerian tech competition looks like. The companies that will define Algeria’s digital economy in 2028-2030 are not those with the most users or the most services — they are those with the most coherent data and ML infrastructure underpinning their operations.
Global super app platforms — Grab in Southeast Asia, Rappi in Latin America — have followed exactly this trajectory: start with operations, build data infrastructure, deploy ML to optimize operations, then monetize the data asset through advertising and financial services. Yassir is executing this playbook at a pace that would have been unrecognizable four years ago.
For every Algerian tech company watching Yassir’s trajectory, the actionable insight is not “we need Vertex AI” — it is “we need to treat our operational data as seriously as our product.” The ML capability follows from the data foundation. The data monetization opportunity follows from the ML capability. And the competitive moat deepens with each additional data point, model iteration, and user interaction.
Frequently Asked Questions
What is Vertex AI and why did Yassir choose it over alternatives?
Vertex AI is Google Cloud’s managed machine learning platform, providing infrastructure for training, deploying, and monitoring ML models without building the underlying MLOps stack from scratch. Yassir’s choice of Vertex AI reflects its existing Google Cloud infrastructure investment — BigQuery for analytics, GKE for containerized workloads, and the full Google Cloud monitoring stack. Switching to AWS SageMaker or Azure Machine Learning would have required migrating data infrastructure that was already deeply integrated with Google services. For companies not already committed to a cloud provider, all three major managed ML platforms (Vertex AI, SageMaker, Azure ML) offer comparable managed deployment capabilities.
How does the Kawarizmi acquisition change Yassir’s business model?
Prior to the Kawarizmi acquisition, Yassir’s revenue came primarily from delivery commissions, ride-hailing fees, and nascent financial services. The acquisition adds a retail media revenue stream: brands pay to reach Yassir’s 8 million users through targeted advertising informed by first-party purchase and behavioral data. This model has higher margins than transaction fees and creates a revenue source that scales with data quality, not just transaction volume. The Kawarizmi deal also extends Yassir’s reach to European and MENA diaspora audiences — Algerians, Moroccans, and Tunisians in France, Belgium, and Canada — who are digitally reachable but not geographically inside Yassir’s delivery zones.
What would it cost an Algerian startup to run a comparable Vertex AI deployment?
Vertex AI pricing is consumption-based, tied to compute hours for training and prediction endpoints for serving. A small-scale implementation comparable to an early Yassir recommendation use case — training monthly models on 1-2 million data points and serving predictions in real time — runs approximately $500-2,000 USD per month depending on prediction volume. The larger cost is the data engineering investment needed to make the data ready for model training: typically 2-4 months of senior engineering time to build reliable data pipelines. Google Cloud’s Startup Program provides credit packages that can substantially offset the initial costs for qualifying Algerian startups.
—
