Your people approve AI work every day. Do they check it?

Ivanooo Direction shows, team by team, how often people question the AI when it matters, and how often they just click approve. In build: early pilots with a small number of companies.

A confident wrong answer pulls people with it: accuracy fell 15 points below working alone when an AI was confidently wrong, and people felt more confident (Shaw and Nave, Wharton, 1,372 people). 79% of AI failures in real conversations left no visible sign (Potts and Sudhof, Stanford, 2026). People checked the AI's facts in 8.7% of conversations and refined its output in 85.7% (Anthropic, 9,830 conversations).

How it works

Connect: read-only access in the enterprise AI plan you already pay for (designed for Microsoft 365 Copilot, ChatGPT Enterprise and Claude Enterprise). Read a sample: each week, a small sample of AI answers headed to a client, a decision or a budget. Report: each person gets a private note on their own week; leaders get team numbers, never names.

Built to be trusted

Private to the person. Counts for groups of five or more, never names or rankings. Never used in hiring, pay or discipline. Designed to run inside your own cloud account. Read-only. Checked by your own reviewers first.

The pilot

One team, six weeks: agree the rules and connect one AI tool read-only; two silent weeks checked against your reviewers; private weekly notes; your first team reading in week 6.

The white paper

Approved by Nobody: why organisations can see everything about their AI except whether their people still direct it, the Direction framework, evidence from 29 sources, and how any team can estimate its own Direction Rate in a week. Contact: hello@ivanooo.com.