Independence

How we choose what to test

We test AI tools that are available for screening job applicants. A provider is selected for testing because of how its tool can be used in recruitment, never because it is or could become a client.

How we are funded

Angoff is funded by its owner and, over time, by audits sold to organizations. We disclose external grants and financial relationships that may be relevant to assessing our independence, with the source, the type of funding and any conditions that could affect the choice of test, method, analysis or publication. No provider pays to be included in or left out of our public tests, or to change how a public result is presented. Providers can commission a confidential audit of their own tool. Such an audit never affects which tools we test publicly or how a published result reads, and it is never published without the provider's written consent. Anyone who commissions such an audit accepts that the assignment is disclosed if the provider's tool is later tested publicly. From the time a provider is selected for testing until publication, or until the test is closed without publication, we do not accept assignments from the provider and do not negotiate new ones. Previous and ongoing business relationships with a tested provider are disclosed together with the result. The same applies to assignments in the past 24 months for a client that we know offers the tested model in a service of its own. If the client has not agreed to be named, we state only that such an assignment exists. We do not select a provider for testing if we cannot disclose our business relationships with it.

The same rules for every provider

Every tested provider receives access to the results and method details it needs and at least 30 calendar days to comment on the test before first publication. Confirmed factual errors in reports already published are corrected under the correction policy without waiting for a new comment period, and the provider concerned is informed of the correction. If a withdrawn report is replaced by a substantially new analysis with new conclusions about the provider, the provider receives a new period of at least 30 calendar days before publication. If a tested provider later becomes a client, the published result stays as it is. It is changed only under our correction policy, the same way as for every other provider.

What we do not do

We do not develop or sell the AI models we review. In our internal work we use AI services, including from providers whose models we test. Which services are used and their role in the study are described in the method description. Our conclusions are not determined by the tested provider or by a client.

Clients and published tests

Results from client assignments are never published without the client's written consent. Aggregated statistics from at least three clients, where neither the client, the system nor individual responses can be identified, may be published under the agreement with the client. The price of an audit does not depend on whether the result is favourable to the client, and no client receives a discount in return for being named as a reference. If a test cannot be interpreted, the price stated in the quote for that case applies. Published tests are in a separate part of the website, without prices or order links, and we never mix a named test result with an offer. When we name a client as a reference, it means only that the client has engaged us, not that the client's system has been audited with a particular result. If we deviate from a rule in the independence policy, we document the deviation, the reason and any consequences for the independence of the audit, and we correct the published information concerned where necessary.