A BI or AI project is only as good as the data you feed it. If your customer spreadsheets, ERP exports and CRM contradict each other, the dashboard will display wrong numbers with a nice layout, and the AI model will learn your mistakes. This is the most common cause of failure we see in Moroccan SMEs, well ahead of the choice of tool or vendor.
Before committing MAD 100,000 or 500,000, there is a short and relatively cheap step: a data quality audit. This article explains what it really covers, what it costs in Morocco, what you can check yourself this week, and how to avoid an audit that creates a CNDP problem instead of solving one.
Why data quality decides whether a BI or AI project succeeds
A dashboard consolidates data from several places. If revenue per customer differs between the ERP and the sales director’s Excel file, the first presentation meeting revolves around one question: “which number is right?” Trust is lost in the first month and the tool is abandoned, whatever it cost.
With AI, the problem is quieter and more expensive. A sales forecasting or customer scoring model learns from your history. Empty fields, categories typed ten different ways, dates set to 01/01/1900: the model learns these anomalies as if they were reality, and its predictions are worthless. Nobody notices until production.
Then there is the money. Cleaning discovered in the middle of a project usually costs two to three times more than the same cleaning done beforehand, because it happens under pressure, with an integrator billing every extra day and a timeline already announced to management. The budgets we detail in our BI dashboard cost guide for Morocco and in our AI project cost guide assume the data is ready. This article covers what happens before.
The warning signs that your data is not ready
You do not need a tool to spot the first symptoms. They show up in meetings, emails and the files that circulate inside the company. These are the ones we find most often in Moroccan SMEs:
- Two departments give two different answers to the same question: monthly revenue, stock level, number of active customers.
- The same customer exists under three spellings (with “SARL”, with “S.A.R.L.”, without any suffix), with a different or missing ICE number depending on the system.
- Cities, sectors or product categories are typed as free text: “Casa”, “Casablanca” and “CASABLANCA” are three different values to a computer.
- An Excel file called “final version v3 definitive” circulates by email or WhatsApp and serves as the unofficial reference.
- The ERP export needs a full day of manual rework before anyone can read it.
- Amounts in MAD and in euros sit in the same column with no currency indicated.
- Only one person knows where the data lives and how to fix it. When they are on leave, reporting stops.
If you recognise three or more of these situations, run an audit before signing a BI or AI quote. It is not a sign of bad management: it is the normal state of a company that grew with tools chosen at different times.
What a data quality audit really covers
A serious audit measures six dimensions, source by source, on the data that will actually feed the future project. The point is not to analyse everything, but to analyse what matters.
The six dimensions
- Completeness: percentage of filled fields on critical columns (customer email, ICE, product code, delivery date, account manager).
- Consistency: does the same information have the same value and format across systems? Is the product catalogue identical in the ERP, the CRM and the e-commerce site?
- Uniqueness: duplicate rate on customers, products and suppliers. In an SME CRM that has never been cleaned, 5 to 15% duplicates is common.
- Accuracy: a sample is checked against reality. Calling thirty customers to verify their details, or physically counting a stock, tells you more than any script.
- Freshness: last update date, delay between the real event and its recording.
- Traceability: who enters what, where and when. Is there a designated source of truth for each piece of data?
The deliverables to insist on
The report should contain a map of sources (which system holds what, who owns it), a score per source and per dimension, the ten blocking issues ranked by impact, a costed estimate of the cleaning effort, and a clear recommendation: go ahead, go ahead with prerequisites, or clean first. If only a developer can read the report, it has failed. The owner or CEO must be able to decide with it.
What a data audit costs in Morocco and how long it takes
The ranges below reflect what we observe on the Moroccan market in 2026, excluding VAT. They vary with the number of sources, how easy the data is to access, and how much field verification is involved.
- Light audit (3 to 5 sources, one business line, a simple BI goal): 5 to 10 working days, MAD 20,000 to 50,000.
- Full audit (ERP, CRM, business spreadsheets, e-commerce, preparation for an AI project, automated profiling and field checks): 3 to 6 weeks, MAD 50,000 to 120,000.
- Daily rate of an experienced data consultant in Morocco: MAD 3,000 to 6,000, depending on profile and company size.
A simple rule: the audit usually represents 5 to 10% of the budget of the project you are considering. Below that, be careful: a “free audit” offered by a BI vendor is a sales visit, not a diagnosis. Above it, the scope is probably oversized for an SME. Some providers fold this work into a broader digital transformation audit, which makes sense if you are also questioning your processes and tools.
On timing, the deciding factor is not the provider but access to data. If your IT team can deliver exports within 48 hours, the audit moves fast. If the ERP integrator has to be involved for every extraction, add weeks. The cleaning that follows is a separate project: depending on the findings, it costs between one and three times the price of the audit.
What you can check yourself before calling a provider
These checks take half a day with Excel and exports from your tools. They do not replace an audit, but they give you an honest pre-diagnosis and shorten the provider’s work, and therefore their quote.
- Export the customer list from the ERP and from the CRM. In Excel, use “Remove Duplicates” on the name, then on the ICE number. Note the row count before and after. A gap above 10% is a serious signal.
- Take the question that matters most to you (for example revenue per customer over twelve months). Ask accounting and sales to produce it separately. Compare the two answers.
- List the ten indicators you want on your future dashboard. For each one, identify the source column and count empty cells with COUNTBLANK.
- Check that the product code is the same in the ERP, the sales price list and the website, on twenty references picked at random.
- For each indicator, write down who enters the data, in which tool and how often. If the answer is “it depends”, you have found a governance problem.
- Time how long the current monthly reporting takes, in person-hours. That is the hidden cost the BI project is supposed to remove, and a strong argument for the budget.
Bring these results to the first meeting. A provider who ignores them and starts from scratch is not the right one.
CNDP and customer data: the risks of a poorly framed audit
A data audit means a third party will access your customer, supplier and sometimes employee files: names, phone numbers, ICE numbers, sometimes national ID numbers or payroll data. Under Law 09-08 you remain the data controller and the provider acts as a processor. That calls for a few concrete precautions:
- Sign a contract that states the purpose of the access, its duration, confidentiality and security commitments, and the deletion of extracts at the end of the assignment.
- Never send a full database over WhatsApp or a personal email account. Prefer pseudonymised extracts, on-site work or controlled remote access.
- If the provider uses a profiling tool hosted outside Morocco, ask where the data travels and check the rules that apply to transfers on the CNDP website.
- Check that the processing involved has gone through the required formalities with the CNDP. Audits often reveal data kept for no reason, such as ten-year-old ID card scans. That is a useful finding, not something to hide.
Our CNDP and Law 09-08 compliance checklist details the formalities to review before you give anyone access to your data. A well-framed audit improves your compliance; an improvised one can damage it.
Clean before you digitise: ETL, deduplication and data governance
The audit leads to a cleaning plan. In our experience it works when it combines three layers, and fails when it stops at the first.
One-off cleaning
Customer deduplication through fuzzy matching on name, ICE and phone number; normalisation of cities, date formats and currencies; completion of missing fields, for example by verifying ICE numbers against the OMPIC register. This is done once, on the data that actually feeds the indicators or the model. There is no point cleaning ten years of history if the project covers the last twenty-four months.
The ETL pipeline
The cleaning rules are then automated in a pipeline that extracts, transforms and loads the data at regular intervals. Without this step, someone redoes the cleaning by hand every month and the BI project turns into a full-time job. This pipeline becomes the foundation of the dashboard, and later of the AI project.
Governance
A named owner for each reference dataset (the customer file belongs to sales administration, the catalogue to the product manager), mandatory fields and drop-down lists in the ERP instead of free text, and a monthly quality indicator reviewed in management meetings. Without governance, clean data degrades within six months and the investment is lost.
How to choose the right provider for this audit
The Moroccan market includes freelancers, IT services companies and consulting firms offering this kind of assignment with very different levels of rigour. A few questions help you sort them quickly:
- Ask for an anonymised sample report from a previous audit. Its readability tells you whether the deliverable will help you decide.
- Ask which tools will be used for profiling (SQL or Python scripts, open-source data quality tools) and how field checks will be carried out.
- If the provider also sells the BI or AI solution, insist that the audit is billed separately, with a “do not proceed” conclusion allowed and accepted.
- Check their knowledge of the local context: ICE numbers, mandatory invoice mentions, ERPs common in Moroccan SMEs, VAT, date and amount formats.
- Read the contract carefully on duration, access, confidentiality, return and deletion of data.
An audit offered at MAD 5,000 is a sales call. An audit at MAD 200,000 for an SME with four data sources is out of proportion. In between, the right provider is the one who tells you clearly whether your data justifies the investment, even when the answer delays their own project.
What now?
Arrowlancer runs data quality audits for Moroccan SMEs, then builds the pipelines, dashboards and governance that follow, as part of our Data service line. If you are hesitating to commit a BI or AI budget without knowing whether your data will hold up, tell us about your situation: a first assessment often settles the question within a few days.