ºìÌÒÊÓƵ

The Post Office Horizon scandal is the most consequential failure of IT project governance in British history. More than 900 people were wrongly convicted and at least 13 suicides have been linked to the scandal. And Horizon was a project management failure as much as a legal one. 

For readers unfamiliar with the case, the Post Office Horizon IT Inquiry provides a detailed account of how software defects, governance failures, organisational culture and the suppression of challenge combined to create one of the most significant miscarriages of justice in British history. 

Further reading: Post Office Horizon IT Inquiry: Official Inquiry Website; The Guardian's summary of the inquiry findings: Five things we learned from the report. 

This series has examined cognitive bias, groupthink, false memory, data illiteracy, Bayesian persuasion and AI accountability. Every one of those failure modes is present in the Horizon case. This article is not a legal review; it is a governance case study for project professionals.

What happened

From the early 2000s, sub-postmasters began reporting unexplained accounting shortfalls generated by errors in Fujitsu’s Horizon IT system. The Post Office, which had invested heavily in Horizon as a flagship technology project, chose consistently to treat these shortfalls as evidence of fraud by individual sub-postmasters rather than systemic error in the system.

The statutory inquiry found that the Post Office told sub-postmasters individually that no one else was having the same problem. This was demonstrably false and known to be false by those making the claim. The Institute of Directors’ subsequent analysis found that the Post Office board was vulnerable to groupthink and unable to gain a meaningful understanding of the organisation’s underlying culture.

The cognitive failure audit

Optimism bias: the prior investment in Horizon made acknowledging systemic failure cognitively intolerable. 

Confirmation bias: evidence of system errors was consistently interpreted as evidence of human wrongdoing. 

Groupthink: the leadership culture made challenge not merely unwelcome but career-limiting.

The illusory truth effect: repeated internal assertions that Horizon was robust became, over years, the organisation’s shared reality, regardless of the evidence accumulating from individual subpostmasters.

Illusory truth effect: repeated claims seem truer; Authority bias amplified everything: institutional power and legal resources sustained a false narrative against the testimony of hundreds of people with no comparable means of resistance.

What AI would have made worse

The Horizon scandal pre-dates widespread AI deployment in organisational governance. But its cognitive failure pattern is precisely what AI amplifies. An AI system trained on the Post Office’s internal data in 2005 would have learned that shortfalls correlate with individual error rather than system failure, because that was the interpretation consistently applied to the evidence in its training data. It would have produced authoritative-sounding outputs confirming the organisation’s preferred narrative, with the additional persuasive force of algorithmic authority.

This is not hypothetical. It is the structural risk that every organisation deploying AI in governance environments faces: systems that learn the organisation’s cognitive biases and reproduce them at scale. The project professional’s job is to build the structural challenge mechanisms that the Post Office so catastrophically lacked.

What the project profession must learn

Horizon is a warning about what happens when governance structures are captured by the cognitive biases and vested interests of the people running them. The technology was the vehicle. The failure was human. That will remain true in any AI-enabled governance environment. The tools change. The cognitive vulnerabilities do not.

Practice checklist

Name the Horizon lessons in your team: The cognitive failure modes in the Horizon case are directly applicable to any major IT or change project. Use it as a teaching case in team governance induction.

Build escalation independence: Governance structures that allow project executives to dismiss or suppress escalations from below replicate the Post Office failure pattern. Independent escalation routes are a structural requirement.

Treat system errors as system errors: When multiple independent sources report the same problem, the default interpretation should be systemic. Individual blame is a later question, not the first one. 

Check AI for confirmation bias: AI trained on historical organisational data will learn and reproduce the cognitive biases embedded in that data. Actively test AI tools against scenarios where the organisation’s preferred narrative was wrong. 

Protect those who raise concerns: The people who kept reporting problems in the Horizon case were treated as suspects. A governance culture that turns its whistleblowers into defendants will produce the same outcome. 

 

You may also be interested in:

• Visit ºìÌÒÊÓƵ bookshop: How to Learn Lessons Effectively
• What is lessons learned in projects management?
• The ‘why’ not the ‘what’: How to make lessons learned easy