Case study · Ekip
The straightforward SIRET searches leave the queue first.
Ekip receives payment-terminal records with a shortened merchant name and a partial city. When the existing match failed, someone had to reconstruct the merchant and search manually.
A small case repeated often enough to cost time
The records Ekip received did not always contain a full legal name. A shortened merchant name and partial city could defeat the existing match, after which a person repeated the search manually.
The risk was not only finding nothing. A plausible name could point to the wrong company, so the system had to return a verifiable candidate rather than an answer presented as certain.
A plausible name can point to the wrong city or activity. The useful result is a candidate the operations team can verify, not a confident guess.
Before
Each failed match meant another Google and registry search, one merchant at a time.
What Basalt built
Basalt connected the existing CSV flow to SIRENE and added checks for the merchant name, city and activity.
After
The system now supports about 30% of requests and returns four to five hours a week. The team keeps the final check.
Key figures
30%
of requests supported
The share genuinely supported, not total resolution.
4–5 h
returned each week
Observed time returned on straightforward searches.
0
changes to the existing CSV format
The team continues with the same file.
02
A workflow around the existing file
The CSV format did not change. Basalt added SIRENE search and checks on name, city and activity so straightforward cases could leave the queue without giving the team another tool.
Every match retains the evidence needed for review. Operations keeps the final validation before using the SIRET.
Select a step
How the workflow works
Read
Receive the same CSV and separate each incomplete merchant record.
Search
Reconstruct useful terms and check possible matches against SIRENE.
Verify
Return the likely SIRET and its evidence to the operations team for the final CNTR check.
03
Why 30% is the right figure to publish
The system supports only about 30% of requests. Ambiguous or incomplete cases remain with the team. That boundary is part of the result.
Those straightforward cases still return four to five hours a week. The value comes from repetition, not a claim to resolve every search automatically.
Bring us one workflow that takes too much time.
In 30 minutes, we will see whether it is worth mapping.
Book 30 minutes- Start with a real task
- Look at the modules involved
- Identify the control points
- Choose a clear next step