Donum Operating Principles
How Donum thinks about reports, AI, data, trust, and the operator experience. The principles that shape what we ship and what we refuse to ship.
On reports
A report is an artifact, not a tool. It has a beginning, a middle, and an end. It should be readable once and still useful. If a stakeholder has to log in to understand it, it is not a report. It is a dashboard.
On AI
AI drafts. Humans own. Donum uses language models aggressively to produce narrative, but no numeric claim is generated by a model. Every metric in a Donum report is pulled from a named source, extracted as structured data, and then wrapped in narrative. Generation is constrained, never free-form, on anything that matters.
On trust
Trust is built by being right in detail, consistently, over time. Not by claiming to be right. Donum ships citations, audit logs, and verification passes because a report you cannot audit is a report you cannot defend.
On data
Customer data belongs to the customer. Donum does not train on customer data, does not share it between workspaces, and does not retain it beyond the documented retention window. The ai.txt file at the root of this site codifies this publicly.
On speed
A report that arrives late is worth less than one that arrives on time, regardless of polish. The operator who gets their monthly report on the first of the month can make decisions before the month ends. Donum optimises for on-time delivery and treats schedule reliability as a first-class metric.
On brand
Generated content that looks generic reads generic, no matter how good the numbers are. Donum's Brand DNA system is not a feature; it is a refusal to ship output that feels like it came from a third-party tool. Every report looks like the operator's brand produced it.
On edit surface
Operators should be able to override the machine. Every section of every Donum report is editable. The editor is a peer to the generator, not a fallback.
On scope
Donum is a reporting platform. It is not a BI tool, not a dashboarding product, and not a general-purpose AI assistant. Every feature we ship either makes a report better, faster, or more trustworthy. Features that do not meet this test get cut.
On approvals
Automation without a stop-the-line control is fragile. Donum defaults to human approval for high-stakes delivery paths (investor updates, board reports) and allows full auto-send for routine reports. The operator decides; the system defaults to safe.
On API and interoperability
Reports produced by Donum are available as data (JSON), as files (PDF, HTML, PPTX), and as MCP tool calls. We assume every report we generate will eventually be read by another system. We design for that.
On pricing
Credit-based pricing maps to the work the system does. Per-seat pricing punishes collaboration. Per-connection pricing punishes the operator for pulling from more sources. We priced ourselves the way we would want to be priced.