Vendor splitting hides behind a shared registration address

Contract splitting is the practice of dividing purchases so that each award falls below the threshold that triggers competitive tendering. Vendor splitting, as this report uses the term, is the variant where the awards go to several nominally separate companies that one actor controls, so the pattern does not show up as repeat awards to a single supplier. The structural tell is shared infrastructure: one registered address across several vendor registrations, the same director under different company names, or phone numbers and email domains that trace to the same point. The difficulty is not knowing what to look for. It is doing the comparison at scale across records that are formatted differently, held in separate databases, and connected only by manual cross-referencing.

The standard approach is a spreadsheet: pull the vendor registrations, extract the address fields, sort and filter for matches. That works when the data is clean. It fails when addresses are formatted inconsistently ("Suite 4B" versus "Unit 4B" versus "4B"), when records are spread across jurisdictions with different filing conventions, or when the analyst is working across dozens of vendors and hundreds of contract records at once. The matching problem is not complex, but it is tedious enough that meaningful overlaps get missed under time pressure.

The deeper problem is that address overlap alone is rarely conclusive. The analyst needs to assess whether the overlap is explained by a shared registered agent, a co-working space, or a mail forwarding address, all of which are legitimate, or by a control relationship between ostensibly independent vendors. That assessment requires reading across several fields at once: director names, incorporation dates, contact details, contract award timelines, and the relationship between contract values and the tendering threshold. That is where human attention runs out before the dataset does. The comparison also has a hard limit: the awards are made by the contracting authority, so a working scheme usually needs someone on the buying side, and no comparison of vendor records can show that on its own.

What the AI contributes to this problem is not database lookup. In this workflow it is given no access to company registries or procurement portals. What it contributes is the ability to hold a structured set of extracted records in context and reason across them at once: comparing fields, flagging inconsistencies, and assessing whether the combination of signals is more consistent with independent vendors or a coordination structure.