[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-body-can-you-defend-this-number-to-the-audit-committee":3},"\nThere is a question every senior finance leader has fielded from an\naudit committee, usually delivered mildly: where did this number\ncome from? For most of corporate history it had a reassuring answer.\nA named analyst, a known spreadsheet, a review chain of people who\ncould each defend their step. The answer was slow, but it was an\nanswer.\n\nThat answer is dissolving. Board packs and investor materials now\ncontain figures a model computed, narratives an AI drafted, and\nsummaries a copilot assembled from systems no single person has\npersonally inspected. One in four executives now report audit-detected\nAI errors reaching boards or external audiences, drawn from data most\nof them do not trust for AI use in the first place\n([the verification gap](/insights/the-verification-gap)). The volume of\nmachine-touched content in board reporting is rising; the ability to\ndefend it, mostly, is not.\n\n## Why the old assurance model broke\n\nThe traditional defence of a board number was human familiarity.\nThe pack was assembled by hand, so each figure had a person behind\nit, and the controller's confidence was a summary of the team's\nconfidence. The model scaled poorly but honestly.\n\nAutomation broke it in three places at once. Volume. A modern\nreporting cycle assembles hundreds of figures from dozens of\nsystems; nobody personally re-derives them. Opacity. A figure that\npassed through a model or an AI summariser carries transformations\nits presenter cannot narrate from memory. And variability. Content\ndrafted by generative tools can differ between runs, so what was\nchecked and what was published are not provably the same artefact.\nAdding more review meetings to this pipeline is the instinctive\nresponse and the wrong one: review without lineage is confidence\ntheatre performed on a larger stage.\n\n## The defensible pack\n\nWhat replaces familiarity is lineage, the property that every\nnumber in the pack can produce, on demand, the account of itself\nthat the committee question implies. Concretely, four requirements:\n\n- **Traceable sources.** The figure names the systems and data it\n  came from, as a record generated by the assembly process itself,\n  not a reconstruction performed after the question.\n- **Deterministic recomputation.** Same inputs, same number, every\n  time. The published figure can be regenerated exactly, which is\n  the difference between an error that can be diagnosed and one\n  that can only be apologised for.\n- **Model provenance where models contributed.** Which model\n  version, trained on what, approved by whom, performing how, with\n  the record attached to the output rather than filed in a\n  different department.\n- **One governed path.** Every figure that reaches the board\n  travels through the same lineage-preserving pipeline. The side\n  channel, the number pasted from a spreadsheet the night before,\n  is where defensibility goes to die, and closing it is a design\n  decision, not a policy memo.\n\n> \"Where did this number come from?\" should be a query with an\n> answer in seconds, not a project with an answer in weeks.\n\nThe offensive version of the case matters as much. Reporting teams\nspend a remarkable share of each cycle on assembly and reconciling\nversions of numbers, time that lineage infrastructure returns as\nthinking time. And a finance function that can answer provenance\nquestions instantly changes its relationship with the committee:\nscrutiny becomes cheap to satisfy, which is what trust looks like\noperationally.\n\nThe near-term forcing function is simply that the question is now\nbeing asked. A committee that has had one AI-flavoured error surface\nin front of it, and per the benchmark data a quarter of executives\nreport exactly that, does not return to not asking.\n\nTry the question on your own last board pack: pick any figure and\ntime how long it takes to produce its full account. If the answer\nembarrasses you, that is the gap Prophesee's Reporting module\ncloses, with lineage, deterministic recomputation and model\npassports behind every figure. [Make the pack defensible](/contact).\n",1786833835743]