Big data
Some links on this page are affiliate links: we may earn a commission at no extra cost to you. Our approach
Big data is mostly an infrastructure problem: when data is too large, too fast or too varied to load into a spreadsheet or a classic database, you need pipelines, distributed storage and suitable processing tools. This problem does not appear at the same size in every company. ## The threshold at which the question arises An SME processing a few million rows a month does not need Hadoop or Spark. A managed data warehouse, a few connectors and good modelling are enough. Big data becomes relevant when volumes exceed a hundred million rows, when data arrives as a continuous stream, or when several sources must be kept in full detail. ## The essential building blocks Three components structure almost every project: ingestion (bringing data in), storage and modelling (organising the data), and serving (exposing data to decision tools). Governance runs through all three: who sees what, how long you keep it, and how you ensure the figures are right. ## Start small, but cleanly The best entry point is the first use case: customer segmentation, stock forecasting, detection of suspicious transactions. It justifies the investment, validates the architecture and serves as a model for the next ones. Projects that start with technology often end up as a data lake full of files nobody uses. ## Examples suited to the Moroccan market A retail chain can combine till receipts, stock and loyalty data to adjust assortments by city. A delivery platform can analyse millions of orders to optimise areas and time slots. A bank or fintech can keep transaction histories to detect suspicious behaviour. In each case, the value does not come from the volume itself but from the decision it leads to.When this guide is for you
- You are piling up data from several systems (tills, e-commerce, CRM, ERP) with no way to cross-reference it
- Your reports run to several hundred thousand rows and your exports are becoming slow or incomplete
- You want to keep detailed history (clicks, logs, orders) over several years to analyse trends
- You are preparing a prediction or recommendation model that needs a reliable historical data foundation
- You must meet traceability or audit requirements on the data your business produces
Step by step
- Find a business use case — Pick a decision that needs cross-referenced data, such as customer segmentation, stock forecasting or detection of suspicious transactions.
- Audit the data and sources — Map sources, volumes, formats and quality problems, then choose a managed warehouse rather than a heavy technology stack unless the volumes justify it.
- Build the ingestion and storage — Set up pipelines and an object storage layer, with checks on volume and treatment costs.
- Model and document — Create the first tables, define the indicators and maintain a data dictionary so the figures can be trusted.
- Serve and govern — Expose the data to the decision tools, define who sees what, how long data is kept, and the legal basis for personal data.
What it costs (in MAD)
| Service | Indicative price | Good to know |
|---|---|---|
| Data audit and architecture | 20,000 – 70,000 MAD (approx. 1,800 – 6,300 €) | Mapping sources and choosing the foundation |
| Managed data warehouse (Snowflake, BigQuery) | 2,000 – 25,000 MAD / month (approx. 180 – 2,250 €) | Billed on the volume processed and stored, variable |
| Ingestion pipelines (tool or development) | 15,000 – 80,000 MAD (approx. 1,350 – 7,200 €) | Managed tool or data developer |
| First tables and data models | 20,000 – 90,000 MAD (approx. 1,800 – 8,100 €) | Data dictionary and defined indicators |
| Data lake on object storage | 500 – 6,000 MAD / month (approx. 45 – 540 €) | S3 or equivalent, cost highly dependent on volume |
| Data engineer support | 25,000 – 60,000 MAD / month (approx. 2,250 – 5,400 €) | A scarce profile in the Moroccan market |
Indicative ranges for the Moroccan market in 2026, excluding tax. Prices vary with scope and provider.
Mistakes to avoid
- Buying big data technology before identifying a use case that makes money
- Confusing big data with BI: for most SMEs, a warehouse of a few gigabytes is more than enough
- Neglecting data quality: a model fed with dirty data produces wrong decisions with confidence
- Letting storage and processing costs explode without monitoring volume and queries
- Collecting personal data without a legal basis or a defined retention period
Frequently asked questions
Does a Moroccan SME really need big data?
Rarely in the strict technical sense. Below a few hundred million rows, a managed data warehouse is enough. Centralising and cleaning up data scattered across tills, e-commerce and CRM still delivers real gains, even with modest volumes.
What is the difference between big data and business intelligence?
BI turns already organised data into dashboards to track activity. Big data is about collecting and processing large, varied volumes, often to feed BI or AI models upstream.
How much data is needed to count as big data?
There is no official threshold. The useful criterion is whether classic tools can no longer process the data in a reasonable time, or the variety of formats. If a few million-row CSV file fits in your warehouse, you do not need Hadoop.
Can a big data project be outsourced?
Yes, often for the initial setup and first architecture. Over time, the company should keep control of the data model and access rights, otherwise it depends entirely on its provider to answer the simplest question.
What skills are needed in-house?
At minimum, someone comfortable with SQL to write queries and check figures, and a data quality owner. A data engineer for the architecture and a data analyst for interpretation complete the team as volumes grow.
Updated on
Related guides
Version française: /services/data-ia/big-data/ · AR · NL