AI-Driven Revenue Recovery for Global Airport Services Operator

QuantSpark helped a global airport services operator identify £1.5M of potential revenue leakage within a four-week feasibility study, using a generative AI-driven algorithm to reconcile invoiced revenue against what should have been billed. The study processed 42,000 invoiced lines, found a 16% discrepancy rate, and proved the approach worked across three airlines and two international airports, turning an opaque billing problem into a defensible, categorised list the client's own audit teams could act on.

The operator was losing an estimated millions of pounds a year through underbilling: charging customers for less than the services actually delivered. Two things were driving it. First, the record of what had actually been delivered, the service capture data, was itself patchy and error-prone. Second, the contractual billing logic that should have converted delivery into an invoice was not being implemented consistently, so agreed terms and actual charges drifted apart. Historically, recovering this lost revenue depended on manual reconciliation. That process was slow, non-exhaustive and inconsistent between contracts, so it could only ever catch a fraction of the leakage and could not scale across the operator's wider operations.

QuantSpark's response was to scope a tightly time-boxed, four-week feasibility study rather than commit to a full build against an unproven hypothesis. The pipeline had three linked stages. First, impact document extraction: multi-modal large language model techniques pulled prices, clauses and conditions directly out of contract documents, including manually scanned PDFs that would defeat a conventional text-extraction approach. Second, service capture matching: natural language processing techniques dynamically mapped what had actually been delivered against the contractual terms just extracted, effectively answering "what should this have cost, given the contract, and what was it actually billed at?" for every line. Third, discrepancy identification: the algorithm reconstructed what the invoice should have been and compared it line by line against the invoice that was actually issued historically. Every discrepancy found was then categorised by its underlying driver, for example an agreed price change that had not been reflected in billing, versus a straightforward billing logic error, so the output fed directly into the client's own financial and commercial audit investigations rather than landing as an undifferentiated list requiring manual triage.

The technology involved sits in two categories: a multi-modal LLM layer for document extraction, and an NLP layer for matching delivered service against contracted terms, wrapped in an automated reconciliation and categorisation step. No specific proprietary system is named here because none was disclosed in the public source, and none should be inferred.

The headline £1.5M is best read as the value of the discrepancies the study surfaced across the 42,000 lines processed, not as cash the client had, at the point the study concluded, already recovered. That distinction matters for anyone citing this case study: the categorisation step is precisely what makes the number defensible, because it lets a human audit team work through each flagged discrepancy against its actual contract and pursue recovery only where that audit confirms the finding. Realised recovery sits downstream of the feasibility study, inside the client's own audit process, and is not itself part of the reported result.

What makes the case compelling is less the size of the number and more what it proves: that a generative AI pipeline can take a process that was manual, non-exhaustive and inconsistent, and turn it into something systematic, categorised and demonstrably applicable across multiple airlines and airport sites within a month, at a scale, 42,000 lines, that manual reconciliation was never going to reach.