[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-body-can-you-defend-this-number-to-the-audit-committee":3},"\nThe most important question in a board meeting is often the\nsimplest. Where did this number come from?\n\nFor years, finance teams could answer it confidently. A report was\nbuilt by analysts, reviewed by managers and signed off through a\nclear process. The work could be traced back to the people who\nproduced it.\n\nToday, that is changing. Board packs increasingly include figures\ngenerated by automated workflows, forecasts produced by models and\ncommentary drafted by AI tools. Reporting is faster than ever, but\nmany organisations have lost something along the way. The ability\nto explain exactly how an answer was produced.\n\nAnd the errors have started arriving. One in four executives now\nreport audit-detected AI errors reaching boards or external\naudiences, built on data most of them admit they do not trust\n([the verification gap](/insights/the-verification-gap)). Audit\ncommittees have noticed. They are no longer interested only in\nwhether a number looks reasonable. They want to know where it came\nfrom, what data it relied on and whether the result can be\nreproduced.\n\nThose are not technical questions. They are governance questions.\n\n## The challenge is not AI. It is traceability.\n\nMost finance leaders are not worried about AI generating more\ninformation. They are worried about defending it.\n\nData flows through multiple systems. Models transform it. AI tools\nsummarise it. Reports are assembled automatically. Somewhere along\nthe way, the connection between a final figure and the steps that\ncreated it becomes difficult to see.\n\nWhen that happens, confidence starts to replace evidence. That is\na risk. A number should not be trusted because people believe it\nis correct. It should be trusted because the organisation can\ndemonstrate why it is correct.\n\n## More reviews will not solve it\n\nThe usual response is to add controls. Another review meeting,\nanother sign-off, another layer of approval.\n\nBut reviews are only as effective as the information available to\nthe reviewer. If nobody can easily show the source data, the\ntransformation steps and the models involved, additional reviews\ndo little more than increase cost and effort.\n\nThe answer is not more process. Every number should arrive at the\nboard with its own evidence attached.\n\n## What a defensible number looks like\n\nA finance team should be able to answer four questions about any\nfigure in a board pack, immediately.\n\n- **What data produced it?** The source systems should be recorded\n  automatically as the report is created, not reconstructed later\n  when questions are asked.\n- **Can it be reproduced?** The same inputs through the same\n  process should give\n  [the same result, every time](/insights/same-inputs-same-number).\n  A number that cannot be reproduced cannot be properly\n  investigated.\n- **What models contributed to it?** Where AI or analytical models\n  play a part, the organisation should know which model, which\n  version, who approved it and how it has been performing, in a\n  standing record per model\n  ([the Model Passport](/insights/the-model-passport)).\n- **Did it follow the approved route?** Every figure should travel\n  the same controlled path to the board. The greatest risks rarely\n  come from the official process. They come from the spreadsheet\n  emailed the night before the meeting.\n\n## Trust comes from evidence\n\nThe benefits go beyond compliance. Finance teams spend a\nremarkable amount of time reconciling reports, checking versions\nand explaining differences between numbers. Much of that effort\nexists because the reporting process has no built-in traceability.\nWhen every figure carries its own audit trail, scrutiny becomes\ncheap to satisfy. That is what trust looks like in practice.\n\nWhen a director asks where a number came from, the answer should\ntake seconds. Not days.\n\n## A simple test\n\nTake your most recent board pack. Pick a number at random. Ask how\nlong it would take to show where it came from, what data produced\nit, how it was calculated, and whether you could recreate it\nexactly.\n\nIf the answer embarrasses you, that is the gap\n[the Prophesee Finance Suite](/solutions/finance/reporting) closes.\nEvery figure carries its evidence, regenerates identically on\ndemand, and names the models behind it, because the record is\nproduced by the system as it works, not assembled for the meeting.\nThe ability to generate information is no longer the advantage.\nThe ability to defend it is. [Make the pack defensible](/contact).\n",1786984935415]