Between 1999 and 2015 more than nine hundred subpostmasters in the United Kingdom were prosecuted for theft, fraud and false accounting on the strength of figures produced by the Post Office’s Horizon accounting system. The shortfalls the system reported were not real. The convictions were quashed on a blanket basis by an Act of Parliament in 2024, and the public inquiry published the first volume of its final report on 8 July 2025, having examined more than two million pages of evidence.
It is usually described as a software scandal, and the software is the least transferable part of it. The transferable part is what an organization does when its own systems produce a number that suits it.
Data becomes evidence only when somebody tests it
A report generated by an internal system carries an authority that has nothing to do with its accuracy. It is precise, it is formatted, it arrives from a source the organization paid for, and it is produced by a process no individual within the business fully understands. Those properties make it persuasive rather than reliable, and the two are easily confused under pressure.
The people relying on Horizon output had no incentive to interrogate it. The figures supported the conclusion the organization had already reached, the alternative explanation implicated a system central to its operations, and the individuals in a position to challenge it were the ones being accused.
None of that requires bad faith. It requires only that the party which owns the system also owns the question, which is the ordinary situation in most internal investigations.
The convictions were quashed on a blanket basis by an Act of Parliament in 2024, sixteen years after the prosecutions began. The inquiry examined more than two million pages of evidence.
Independence is the product
Where an internal system supplies the central fact in a dispute, the useful question is who has examined it who does not report to the people relying on it.
An expert instructed to confirm a conclusion is worth very little once the conclusion is challenged, because the instruction is disclosable and the reasoning will be read in the light of it. The value of external litigation support lies precisely in the fact that it can return an unwelcome answer, and an organization that has never received one from a given adviser should wonder why.
The point applies well before any proceedings begin. Most internal investigations that later become litigation began as an operational enquiry conducted by the function that owned the process under examination, and the record created at that stage is the record that gets disclosed.
What tends to go wrong first
- The system output is treated as the finding rather than as an input to one. A discrepancy report identifies where to look, and it does not establish what happened there.
- Nobody establishes what the system does when it is wrong. Known defects, error rates and remote access by the vendor are ordinary features of enterprise software, and they are rarely documented in a form an investigator can use.
- Preservation is arranged too late to matter. Logs rotate, environments are rebuilt and the material that would have settled the question is gone by the time anybody realises it mattered.
- The people who could challenge the finding are the subjects of it. Where that is the structure, the absence of challenge tells you nothing at all.
An expert instructed to confirm a conclusion is worth very little once the conclusion is challenged. The instruction is disclosable, and the reasoning will be read in the light of it.
The cost of getting this wrong compounds quietly
The Horizon prosecutions ran for sixteen years. Each one made the next easier to bring, because a pattern of convictions is itself treated as corroboration, and the institutional confidence in the underlying system grew as a direct consequence of the harm it was causing.
That compounding is the feature worth recognising in a smaller organization, where the same structure produces a series of dismissals rather than prosecutions. A method that has produced findings before is trusted more each time it is used, and nobody revisits the first one.
Where an organization should place the check
Independent examination is disproportionate for most internal matters and essential for a narrow class of them. The threshold worth adopting is whether the outcome could result in dismissal, referral to a prosecutor, or a claim, since those are the situations in which the evidence will be read by somebody adversarial.
In that class, corporate investigations and financial investigations earn their cost by producing a record that survives being challenged rather than one that satisfies the organization commissioning it. Our Book of the Month shelf has The Secret Sentry, which is a long study of the same failure in a different setting, where the volume of material collected was consistently mistaken for understanding of it.
The question for a board
Where the central fact in your last serious internal case came from a system the organization operates, establish who outside the owning function examined it, and what they were asked. Where the answer is that nobody did, the finding has not yet been tested.
Falcone International
Bring us a question like this one
We handle corporate investigations, due diligence, financial investigations and duty of care — usually for people who need something established quietly, and established properly, before it turns into a problem.